Yanlin Wang

Anji Li, Mingwei Liu, Zhenxi Chen et al.

Automated unit test generation using large language models (LLMs) holds great promise but often struggles with generating tests that are both correct and maintainable in real-world projects. This paper presents KTester, a novel framework that integrates project-specific knowledge and testing domain knowledge to enhance LLM-based test generation. Our approach first extracts project structure and usage knowledge through static analysis, which provides rich context for the model. It then employs a testing-domain-knowledge-guided separation of test case design and test method generation, combined with a multi-perspective prompting strategy that guides the LLM to consider diverse testing heuristics. The generated tests follow structured templates, improving clarity and maintainability. We evaluate KTester on multiple open-source projects, comparing it against state-of-the-art LLM-based baselines using automatic correctness and coverage metrics, as well as a human study assessing readability and maintainability. Results demonstrate that KTester significantly outperforms existing methods across six key metrics, improving execution pass rate by 5.69% and line coverage by 8.83% over the strongest baseline, while requiring less time and generating fewer test cases. Human evaluators also rate the tests produced by KTester significantly higher in terms of correctness, readability, and maintainability, confirming the practical advantages of our knowledge-driven framework.

Wanjun Zhong, Ruixiang Cui, Yiduo Guo et al.

Evaluating the general abilities of foundation models to tackle human-level tasks is a vital aspect of their development and application in the pursuit of Artificial General Intelligence (AGI). Traditional benchmarks, which rely on artificial datasets, may not accurately represent human-level capabilities. In this paper, we introduce AGIEval, a novel benchmark specifically designed to assess foundation model in the context of human-centric standardized exams, such as college entrance exams, law school admission tests, math competitions, and lawyer qualification tests. We evaluate several state-of-the-art foundation models, including GPT-4, ChatGPT, and Text-Davinci-003, using this benchmark. Impressively, GPT-4 surpasses average human performance on SAT, LSAT, and math competitions, attaining a 95% accuracy rate on the SAT Math test and a 92.5% accuracy on the English test of the Chinese national college entrance exam. This demonstrates the extraordinary performance of contemporary foundation models. In contrast, we also find that GPT-4 is less proficient in tasks that require complex reasoning or specific domain knowledge. Our comprehensive analyses of model capabilities (understanding, knowledge, reasoning, and calculation) reveal these models' strengths and limitations, providing valuable insights into future directions for enhancing their general capabilities. By concentrating on tasks pertinent to human cognition and decision-making, our benchmark delivers a more meaningful and robust evaluation of foundation models' performance in real-world scenarios. The data, code, and all model outputs are released in https://github.com/ruixiangcui/AGIEval.

Daya Guo, Shuai Lu, Nan Duan et al.

Pre-trained models for programming languages have recently demonstrated great success on code intelligence. To support both code-related understanding and generation tasks, recent works attempt to pre-train unified encoder-decoder models. However, such encoder-decoder framework is sub-optimal for auto-regressive tasks, especially code completion that requires a decoder-only manner for efficient inference. In this paper, we present UniXcoder, a unified cross-modal pre-trained model for programming language. The model utilizes mask attention matrices with prefix adapters to control the behavior of the model and leverages cross-modal contents like AST and code comment to enhance code representation. To encode AST that is represented as a tree in parallel, we propose a one-to-one mapping method to transform AST in a sequence structure that retains all structural information from the tree. Furthermore, we propose to utilize multi-modal contents to learn representation of code fragment with contrastive learning, and then align representations among programming languages using a cross-modal generation task. We evaluate UniXcoder on five code-related tasks over nine datasets. To further evaluate the performance of code fragment representation, we also construct a dataset for a new task, called zero-shot code-to-code search. Results show that our model achieves state-of-the-art performance on most tasks and analysis reveals that comment and AST can both enhance UniXcoder.

Ensheng Shi, Yanlin Wang, Hongyu Zhang et al.

Recently, fine-tuning pre-trained code models such as CodeBERT on downstream tasks has achieved great success in many software testing and analysis tasks. While effective and prevalent, fine-tuning the pre-trained parameters incurs a large computational cost. In this paper, we conduct an extensive experimental study to explore what happens to layer-wise pre-trained representations and their encoded code knowledge during fine-tuning. We then propose efficient alternatives to fine-tune the large pre-trained code model based on the above findings. Our experimental study shows that (1) lexical, syntactic and structural properties of source code are encoded in the lower, intermediate, and higher layers, respectively, while the semantic property spans across the entire model. (2) The process of fine-tuning preserves most of the code properties. Specifically, the basic code properties captured by lower and intermediate layers are still preserved during fine-tuning. Furthermore, we find that only the representations of the top two layers change most during fine-tuning for various downstream tasks. (3) Based on the above findings, we propose Telly to efficiently fine-tune pre-trained code models via layer freezing. The extensive experimental results on five various downstream tasks demonstrate that training parameters and the corresponding time cost are greatly reduced, while performances are similar or better. Replication package including source code, datasets, and online Appendix is available at: \url{https://github.com/DeepSoftwareAnalytics/Telly}.

Haochen Li, Chunyan Miao, Cyril Leung et al.

Code search, which aims at retrieving the most relevant code fragment for a given natural language query, is a common activity in software development practice. Recently, contrastive learning is widely used in code search research, where many data augmentation approaches for source code (e.g., semantic-preserving program transformation) are proposed to learn better representations. However, these augmentations are at the raw-data level, which requires additional code analysis in the preprocessing stage and additional training costs in the training stage. In this paper, we explore augmentation methods that augment data (both code and query) at representation level which does not require additional data processing and training, and based on this we propose a general format of representation-level augmentation that unifies existing methods. Then, we propose three new augmentation methods (linear extrapolation, binary interpolation, and Gaussian scaling) based on the general format. Furthermore, we theoretically analyze the advantages of the proposed augmentation methods over traditional contrastive learning methods on code search. We experimentally evaluate the proposed representation-level augmentation methods with state-of-the-art code search models on a large-scale public dataset consisting of six programming languages. The experimental results show that our approach can consistently boost the performance of the studied code search models. Our source code is available at https://github.com/Alex-HaochenLi/RACS.

Ziyao Zhang, Yanlin Wang, Chong Wang et al.

Code generation aims to automatically generate code from input requirements, significantly enhancing development efficiency. Recent large language models (LLMs) based approaches have shown promising results and revolutionized code generation task. Despite the promising performance, LLMs often generate contents with hallucinations, especially for the code generation scenario requiring the handling of complex contextual dependencies in practical development process. Although previous study has analyzed hallucinations in LLM-powered code generation, the study is limited to standalone function generation. In this paper, we conduct an empirical study to study the phenomena, mechanism, and mitigation of LLM hallucinations within more practical and complex development contexts in repository-level generation scenario. First, we manually examine the code generation results from six mainstream LLMs to establish a hallucination taxonomy of LLM-generated code. Next, we elaborate on the phenomenon of hallucinations, analyze their distribution across different models. We then analyze causes of hallucinations and identify four potential factors contributing to hallucinations. Finally, we propose an RAG-based mitigation method, which demonstrates consistent effectiveness in all studied LLMs. The replication package including code, data, and experimental results is available at https://github.com/DeepSoftwareAnalytics/LLMCodingHallucination

Ensheng Shi, Yanlin Wang, Wei Tao et al. · tencent-ai

Commit messages are important for software development and maintenance. Many neural network-based approaches have been proposed and shown promising results on automatic commit message generation. However, the generated commit messages could be repetitive or redundant. In this paper, we propose RACE, a new retrieval-augmented neural commit message generation method, which treats the retrieved similar commit as an exemplar and leverages it to generate an accurate commit message. As the retrieved commit message may not always accurately describe the content/intent of the current code diff, we also propose an exemplar guider, which learns the semantic similarity between the retrieved and current code diff and then guides the generation of commit message based on the similarity. We conduct extensive experiments on a large public dataset with five programming languages. Experimental results show that RACE can outperform all baselines. Furthermore, RACE can boost the performance of existing Seq2Seq models in commit message generation.

Caihua Li, Lianghong Guo, Yanlin Wang et al. · tencent-ai

Issue resolution, a complex Software Engineering (SWE) task integral to real-world development, has emerged as a compelling challenge for artificial intelligence. The establishment of benchmarks like SWE-bench revealed this task as profoundly difficult for large language models, thereby significantly accelerating the evolution of autonomous coding agents. This paper presents a systematic survey of this emerging domain. We begin by examining data construction pipelines, covering automated collection and synthesis approaches. We then provide a comprehensive analysis of methodologies, spanning training-free frameworks with their modular components to training-based techniques, including supervised fine-tuning and reinforcement learning. Subsequently, we discuss critical analyses of data quality and agent behavior, alongside practical applications. Finally, we identify key challenges and outline promising directions for future research. An open-source repository is maintained at https://github.com/DeepSoftwareAnalytics/Awesome-Issue-Resolution to serve as a dynamic resource in this field.

Yanlin Wang, Wanjun Zhong, Yanxian Huang et al.

In recent years, Large Language Models (LLMs) have achieved remarkable success and have been widely used in various downstream tasks, especially in the tasks of the software engineering (SE) field. We find that many studies combining LLMs with SE have employed the concept of agents either explicitly or implicitly. However, there is a lack of an in-depth survey to sort out the development context of existing works, analyze how existing works combine the LLM-based agent technologies to optimize various tasks, and clarify the framework of LLM-based agents in SE. In this paper, we conduct the first survey of the studies on combining LLM-based agents with SE and present a framework of LLM-based agents in SE which includes three key modules: perception, memory, and action. We also summarize the current challenges in combining the two fields and propose future opportunities in response to existing challenges. We maintain a GitHub repository of the related papers at: https://github.com/DeepSoftwareAnalytics/Awesome-Agent4SE.

Ensheng Shi, Fengji Zhang, Yanlin Wang et al.

Software development plays a crucial role in driving innovation and efficiency across modern societies. To meet the demands of this dynamic field, there is a growing need for an effective software development assistant. However, existing large language models represented by ChatGPT suffer from limited accessibility, including training data and model weights. Although other large open-source models like LLaMA have shown promise, they still struggle with understanding human intent. In this paper, we present SoTaNa, an open-source software development assistant. SoTaNa utilizes ChatGPT to generate high-quality instruction-based data for the domain of software engineering and employs a parameter-efficient fine-tuning approach to enhance the open-source foundation model, LLaMA. We evaluate the effectiveness of \our{} in answering Stack Overflow questions and demonstrate its capabilities. Additionally, we discuss its capabilities in code summarization and generation, as well as the impact of varying the volume of generated data on model performance. Notably, SoTaNa can run on a single GPU, making it accessible to a broader range of researchers. Our code, model weights, and data are public at \url{https://github.com/DeepSoftwareAnalytics/SoTaNa}.

Jianpeng Zhou, Wanjun Zhong, Yanlin Wang et al.

Large Language Models (LLMs) demonstrate impressive ability in handling reasoning tasks. However, unlike humans who can instinctively adapt their problem-solving strategies to the complexity of task, most LLM-based methods adopt a one-size-fits-all approach. These methods employ consistent models, sample sizes, prompting methods and levels of problem decomposition, regardless of the problem complexity. The inflexibility of these methods can bring unnecessary computational overhead or sub-optimal performance. To address this limitation, we introduce an Adaptive-Solver (AS) framework tha dynamically adapts solving strategies to suit various problems, enabling the flexible allocation of test-time computational resources. The framework functions with two primary modules. The initial evaluation module assesses the reliability of the current solution using answer consistency. If the solution is deemed unreliable, the subsequent adaptation module comes into play. Within this module, various types of adaptation strategies are employed collaboratively. Through such dynamic and multi-faceted adaptations, our framework can help reduce computational consumption and improve performance. Experimental results from complex reasoning benchmarks reveal that our method can significantly reduce API costs (up to 85%) while maintaining original performance. Alternatively, it achieves up to 4.5% higher accuracy compared to the baselines at the same cost. The code and dataset are available at https://github.com/john1226966735/Adaptive-Solver.

Wenchao Gu, Yanlin Wang, Lun Du et al.

Code search is to search reusable code snippets from source code corpus based on natural languages queries. Deep learning-based methods of code search have shown promising results. However, previous methods focus on retrieval accuracy but lacked attention to the efficiency of the retrieval process. We propose a novel method CoSHC to accelerate code search with deep hashing and code classification, aiming to perform an efficient code search without sacrificing too much accuracy. To evaluate the effectiveness of CoSHC, we apply our method to five code search models. Extensive experimental results indicate that compared with previous code search baselines, CoSHC can save more than 90% of retrieval time meanwhile preserving at least 99% of retrieval accuracy.

Ensheng Shi, Yanlin Wang, Wenchao Gu et al.

Code search aims to retrieve semantically relevant code snippets for a given natural language query. Recently, many approaches employing contrastive learning have shown promising results on code representation learning and greatly improved the performance of code search. However, there is still a lot of room for improvement in using contrastive learning for code search. In this paper, we propose CoCoSoDa to effectively utilize contrastive learning for code search via two key factors in contrastive learning: data augmentation and negative samples. Specifically, soft data augmentation is to dynamically masking or replacing some tokens with their types for input sequences to generate positive samples. Momentum mechanism is used to generate large and consistent representations of negative samples in a mini-batch through maintaining a queue and a momentum encoder. In addition, multimodal contrastive learning is used to pull together representations of code-query pairs and push apart the unpaired code snippets and queries. We conduct extensive experiments to evaluate the effectiveness of our approach on a large-scale dataset with six programming languages. Experimental results show that: (1) CoCoSoDa outperforms 14 baselines and especially exceeds CodeBERT, GraphCodeBERT, and UniXcoder by 13.3%, 10.5%, and 5.9% on average MRR scores, respectively. (2) The ablation studies show the effectiveness of each component of our approach. (3) We adapt our techniques to several different pre-trained models such as RoBERTa, CodeBERT, and GraphCodeBERT and observe a significant boost in their performance in code search. (4) Our model performs robustly under different hyper-parameters. Furthermore, we perform qualitative and quantitative analyses to explore reasons behind the good performance of our model.

Kaifeng He, Xiaojun Zhang, Peiliang Cai et al.

Large language models (LLMs) frequently generate defective outputs in code generation tasks, ranging from logical bugs to security vulnerabilities. While these generation failures are often treated as model-level limitations, empirical evidence increasingly traces their root causes to imperfections within the training corpora. Yet, the specific mechanisms linking training data quality issues to generated code quality issues remain largely unmapped. This paper presents a systematic literature review of 114 primary studies to investigate how training data quality issues propagate into code generation. We establish a unified taxonomy that categorizes generated code quality issues across nine dimensions and training data quality issues into code and non-code attributes. Based on this taxonomy, we formalize a causal framework detailing 18 typical propagation mapping mechanisms. Furthermore, we synthesize state-of-the-art detection and mitigation techniques across the data, model, and generation lifecycles. The reviewed literature reveals a clear methodological shift: quality assurance is transitioning from reactive, heuristic-based post-generation filtering toward proactive, data-centric governance and closed-loop repair. Finally, we identify open challenges and outline research directions for developing reliable LLMs for code through integrated data curation and continuous evaluation. Our repository is available at https://github.com/SYSUSELab/From-Data-to-Code.

Fangyu Zhang, Yanjie Wei, Jin Liu et al.

The development of noninvasive brain imaging such as resting-state functional magnetic resonance imaging (rs-fMRI) and its combination with AI algorithm provides a promising solution for the early diagnosis of Autism spectrum disorder (ASD). However, the performance of the current ASD classification based on rs-fMRI still needs to be improved. This paper introduces a classification framework to aid ASD diagnosis based on rs-fMRI. In the framework, we proposed a novel filter feature selection method based on the difference between step distribution curves (DSDC) to select remarkable functional connectivities (FCs) and utilized a multilayer perceptron (MLP) which was pretrained by a simplified Variational Autoencoder (VAE) for classification. We also designed a pipeline consisting of a normalization procedure and a modified hyperbolic tangent (tanh) activation function to replace the original tanh function, further improving the model accuracy. Our model was evaluated by 10 times 10-fold cross-validation and achieved an average accuracy of 78.12%, outperforming the state-of-the-art methods reported on the same dataset. Given the importance of sensitivity and specificity in disease diagnosis, two constraints were designed in our model which can improve the model's sensitivity and specificity by up to 9.32% and 10.21%, respectively. The added constraints allow our model to handle different application scenarios and can be used broadly.

Wendong Bi, Lun Du, Qiang Fu et al.

Graph Neural Networks (GNNs) are popular machine learning methods for modeling graph data. A lot of GNNs perform well on homophily graphs while having unsatisfactory performance on heterophily graphs. Recently, some researchers turn their attention to designing GNNs for heterophily graphs by adjusting the message passing mechanism or enlarging the receptive field of the message passing. Different from existing works that mitigate the issues of heterophily from model design perspective, we propose to study heterophily graphs from an orthogonal perspective by rewiring the graph structure to reduce heterophily and making the traditional GNNs perform better. Through comprehensive empirical studies and analysis, we verify the potential of the rewiring methods. To fully exploit its potential, we propose a method named Deep Heterophily Graph Rewiring (DHGR) to rewire graphs by adding homophilic edges and pruning heterophilic edges. The detailed way of rewiring is determined by comparing the similarity of label/feature-distribution of node neighbors. Besides, we design a scalable implementation for DHGR to guarantee high efficiency. DHRG can be easily used as a plug-in module, i.e., a graph pre-processing step, for any GNNs, including both GNN for homophily and heterophily, to boost their performance on the node classification task. To the best of our knowledge, it is the first work studying graph rewiring for heterophily graphs. Extensive experiments on 11 public graph datasets demonstrate the superiority of our proposed methods.

Wendong Bi, Lun Du, Qiang Fu et al.

Graph Neural Networks (GNNs) have shown expressive performance on graph representation learning by aggregating information from neighbors. Recently, some studies have discussed the importance of modeling neighborhood distribution on the graph. However, most existing GNNs aggregate neighbors' features through single statistic (e.g., mean, max, sum), which loses the information related to neighbor's feature distribution and therefore degrades the model performance. In this paper, inspired by the method of moment in statistical theory, we propose to model neighbor's feature distribution with multi-order moments. We design a novel GNN model, namely Mix-Moment Graph Neural Network (MM-GNN), which includes a Multi-order Moment Embedding (MME) module and an Element-wise Attention-based Moment Adaptor module. MM-GNN first calculates the multi-order moments of the neighbors for each node as signatures, and then use an Element-wise Attention-based Moment Adaptor to assign larger weights to important moments for each node and update node representations. We conduct extensive experiments on 15 real-world graphs (including social networks, citation networks and web-page networks etc.) to evaluate our model, and the results demonstrate the superiority of MM-GNN over existing state-of-the-art models.

Jiachi Chen, Qingyuan Zhong, Yanlin Wang et al.

The emergence of Large Language Models (LLMs) has significantly influenced various aspects of software development activities. Despite their benefits, LLMs also pose notable risks, including the potential to generate harmful content and being abused by malicious developers to create malicious code. Several previous studies have focused on the ability of LLMs to resist the generation of harmful content that violates human ethical standards, such as biased or offensive content. However, there is no research evaluating the ability of LLMs to resist malicious code generation. To fill this gap, we propose RMCBench, the first benchmark comprising 473 prompts designed to assess the ability of LLMs to resist malicious code generation. This benchmark employs two scenarios: a text-to-code scenario, where LLMs are prompted with descriptions to generate code, and a code-to-code scenario, where LLMs translate or complete existing malicious code. Based on RMCBench, we conduct an empirical study on 11 representative LLMs to assess their ability to resist malicious code generation. Our findings indicate that current LLMs have a limited ability to resist malicious code generation with an average refusal rate of 40.36% in text-to-code scenario and 11.52% in code-to-code scenario. The average refusal rate of all LLMs in RMCBench is only 28.71%; ChatGPT-4 has a refusal rate of only 35.73%. We also analyze the factors that affect LLMs' ability to resist malicious code generation and provide implications for developers to enhance model robustness.

Lunyiu Nie, Jiuding Sun, Yanlin Wang et al.

The recent prevalence of pretrained language models (PLMs) has dramatically shifted the paradigm of semantic parsing, where the mapping from natural language utterances to structured logical forms is now formulated as a Seq2Seq task. Despite the promising performance, previous PLM-based approaches often suffer from hallucination problems due to their negligence of the structural information contained in the sentence, which essentially constitutes the key semantics of the logical forms. Furthermore, most works treat PLM as a black box in which the generation process of the target logical form is hidden beneath the decoder modules, which greatly hinders the model's intrinsic interpretability. To address these two issues, we propose to incorporate the current PLMs with a hierarchical decoder network. By taking the first-principle structures as the semantic anchors, we propose two novel intermediate supervision tasks, namely Semantic Anchor Extraction and Semantic Anchor Alignment, for training the hierarchical decoders and probing the model intermediate representations in a self-adaptive manner alongside the fine-tuning process. We conduct intensive experiments on several semantic parsing benchmarks and demonstrate that our approach can consistently outperform the baselines. More importantly, by analyzing the intermediate representations of the hierarchical decoders, our approach also makes a huge step toward the intrinsic interpretability of PLMs in the domain of semantic parsing.

Yuxuan Chen, Mingwei Liu, Guangsheng Ou et al.

Code search is essential for code reuse, allowing developers to efficiently locate relevant code snippets. The advent of powerful decoder-only Large Language Models (LLMs) has revolutionized many code intelligence tasks. However, their effectiveness for the retrieval-based task of code search, particularly compared to established encoder-based models, remains underexplored. This paper addresses this gap by presenting a large-scale systematic evaluation of eleven decoder-only LLMs, analyzing their performance across zero-shot and fine-tuned settings. Our results show that fine-tuned decoder-only models, particularly CodeGemma, significantly outperform encoder-only models like UniXcoder, achieving a 40.4% higher Mean Average Precision (MAP) on the CoSQA$^+$ benchmark. Our analysis further reveals two crucial nuances for practitioners: first, the relationship between model size and performance is non-monotonic, with mid-sized models often outperforming larger variants; second, the composition of the training data is critical, as a multilingual dataset enhances generalization while a small amount of data from a specific language can act as noise and interfere with model effectiveness. These findings offer a comprehensive guide to selecting and optimizing modern LLMs for code search.

Kaifeng He, Mingwei Liu, Chong Wang et al.

Code generation with large language models (LLMs) is highly sensitive to token selection during decoding, particularly at uncertain decision points that influence program logic. While standard strategies such as greedy decoding treat all tokens uniformly, they overlook code-specific uncertainty patterns, leading to suboptimal performance. This paper presents an empirical study revealing that many generation errors stem from token ranking mistakes at high-uncertainty steps, where the correct token is present but not top-ranked. Motivated by these findings, we propose AdaDec, a lookahead-based uncertainty-guided adaptive decoding framework that integrates a token-level pause-then-rerank mechanism driven by token uncertainty. AdaDec learns model-specific uncertainty thresholds and applies a lookahead-based reranking strategy when uncertainty is high. Experiments on HumanEval+, MBPP+, and DevEval benchmarks show that AdaDec improves Pass@1 accuracy by up to 20.9% in absolute terms over greedy decoding. More importantly, it consistently outperforms both competitive baselines like Beam Search and state-of-the-art adaptive decoding methods such as AdapT, while maintaining high efficiency through selective, uncertainty-triggered pausing. Our results highlight the promise of uncertainty-aware adaptive decoding for improving both the reliability and efficiency of LLM-based code generation.

Xinyu Li, Yintong Huo, Chenxi Mao et al.

Log-based anomaly detection is fundamentally constrained by training data sparsity. Our empirical study reveals that public benchmark datasets cover less than 10% of source code log templates. Consequently, models frequently misclassify unseen but valid execution paths as anomalies, leading to false alarms. To address this, we propose AnomalyGen, a novel framework that augments training data by synthesizing labeled log sequences from source code. AnomalyGen combines log-oriented static analysis with Large Language Model (LLM) reasoning in three stages: (1) building Log-Oriented Control Flow Graphs (LCFGs) to enumerate structurally valid execution paths; (2) applying LLM Chain-of-Thought (CoT) reasoning to verify logical consistency and generate realistic runtime parameters (e.g., block IDs, IP addresses); and (3) labeling generated sequences with domain heuristics. Evaluations on HDFS and Zookeeper across 12 diverse anomaly detection models show AnomalyGen consistently improves performance. Deep learning models achieved average F1-score gains of 2.18% (HDFS) and 1.69% (Zookeeper), with an unsupervised Transformer on HDFS jumping from 0.818 to 0.970. Ablation results show that both static analysis and LLM-based verification are necessary: removing them reduces F1 by up to 8.7 and 10.7 percentage points, respectively. Our framework and datasets are publicly available to facilitate future research.

Juntao Chen, Tingting Bi, Yanlin Wang et al.

Modern software systems rely on dependency networks of reusable libraries, where breaking changes propagate and cause downstream consumers to fail. Despite growing research across ecosystems, no comprehensive synthesis exists. We conduct a systematic literature review of 97 primary studies, answering four research questions across five ecosystems: Maven/Java, npm/JavaScript, Python, Web APIs, and Linux distributions. The synthesis yields four results. First, a four-dimensional taxonomy along Nature, Detectability, Scope, and Visibility. Second, five reason categories and five impact dimensions, where maintenance and design improvements account for a larger share of breaking changes than new feature work. Third, 43 detection approaches that reach high accuracy on syntactic breaks but limited coverage on behavioral ones. Fourth, 66 strategies for communicating, preventing, and recovering from breaking changes, organized by the actor's role. Based on these findings, we identify three open challenges and three research opportunities. The challenges are behavioral break detection at scale, the failure of semantic versioning as a trust mechanism, and transitive dependency propagation under information asymmetry. The opportunities are LLM-augmented behavioral contract inference, ecosystem-level dependency graph intelligence, and domain-specific tooling for ML and data science.

Ruimin Ma, Yanlin Wang, Yanjie Wei et al.

Accurate diagnosis of autism spectrum disorder (ASD) based on neuroimaging data has significant implications, as extracting useful information from neuroimaging data for ASD detection is challenging. Even though machine learning techniques have been leveraged to improve the information extraction from neuroimaging data, the varying data quality caused by different meta-data conditions (i.e., data collection strategies) limits the effective information that can be extracted, thus leading to data-dependent predictive accuracies in ASD detection, which can be worse than random guess in some cases. In this work, we systematically investigate the impact of three kinds of meta-data on the predictive accuracy of classifying ASD based on structural MRI collected from 20 different sites, where meta-data conditions vary.

Yanlin Wang, Xinyi Xu, Jiachi Chen et al.

The rise of large language models (LLMs) has sparked a surge of interest in agents, leading to the rapid growth of agent frameworks. Agent frameworks are software toolkits and libraries that provide standardized components, abstractions, and orchestration mechanisms to simplify agent development. Despite widespread use of agent frameworks, their practical applications and how they influence the agent development process remain underexplored. Different agent frameworks encounter similar problems during use, indicating that these recurring issues deserve greater attention and call for further improvements in agent framework design. Meanwhile, as the number of agent frameworks continues to grow and evolve, more than 80% of developers report difficulties in identifying the frameworks that best meet their specific development requirements. In this paper, we conduct the first empirical study of LLM-based agent frameworks, exploring real-world experiences of developers in building AI agents. To compare how well the agent frameworks meet developer needs, we further collect developer discussions for the ten previously identified agent frameworks, resulting in a total of 11,910 discussions. Finally, by analyzing these discussions, we compare the frameworks across five dimensions: development efficiency, functional abstraction, learning cost, performance optimization, and maintainability, which refers to how easily developers can update and extend both the framework itself and the agents built upon it over time. Our comparative analysis reveals significant differences among frameworks in how they meet the needs of agent developers. Overall, we provide a set of findings and implications for the LLM-driven AI agent framework ecosystem and offer insights for the design of future LLM-based agent frameworks and agent developers.

Sicong Liu, Yanxian Huang, Mingwei Liu et al.

Code generation tasks aim to automate the conversion of user requirements into executable code, significantly reducing manual development efforts and enhancing software productivity. The emergence of large language models (LLMs) has significantly advanced code generation, though their efficiency is still impacted by certain inherent architectural constraints. Each token generation necessitates a complete inference pass, requiring persistent retention of contextual information in memory and escalating resource consumption. While existing research prioritizes inference-phase optimizations such as prompt compression and model quantization, the generation phase remains underexplored. To tackle these challenges, we propose a knowledge-infused framework named ShortCoder, which optimizes code generation efficiency while preserving semantic equivalence and readability. In particular, we introduce: (1) ten syntax-level simplification rules for Python, derived from AST-preserving transformations, achieving 18.1% token reduction without functional compromise; (2) a hybrid data synthesis pipeline integrating rule-based rewriting with LLM-guided refinement, producing ShorterCodeBench, a corpus of validated tuples of original code and simplified code with semantic consistency; (3) a fine-tuning strategy that injects conciseness awareness into the base LLMs. Extensive experimental results demonstrate that ShortCoder consistently outperforms state-of-the-art methods on HumanEval, achieving an improvement of 18.1%-37.8% in generation efficiency over previous methods while ensuring the performance of code generation.

Tianyue Jiang, Yanli Wang, Yanlin Wang et al.

Repository-level code completion remains a challenging task for existing code large language models (code LLMs) due to their limited understanding of repository-specific context and domain knowledge. While retrieval-augmented generation (RAG) approaches have shown promise by retrieving relevant code snippets as cross-file context, they suffer from two fundamental problems: misalignment between the query and the target code in the retrieval process, and the inability of existing retrieval methods to effectively utilize the inference information. To address these challenges, we propose AlignCoder, a repository-level code completion framework that introduces a query enhancement mechanism and a reinforcement learning based retriever training method. Our approach generates multiple candidate completions to construct an enhanced query that bridges the semantic gap between the initial query and the target code. Additionally, we employ reinforcement learning to train an AlignRetriever that learns to leverage inference information in the enhanced query for more accurate retrieval. We evaluate AlignCoder on two widely-used benchmarks (CrossCodeEval and RepoEval) across five backbone code LLMs, demonstrating an 18.1% improvement in EM score compared to baselines on the CrossCodeEval benchmark. The results show that our framework achieves superior performance and exhibits high generalizability across various code LLMs and programming languages.

Mingwei Liu, Zihao Wang, Zhenxi Chen et al.

Translating natural language descriptions into viable code fixes remains a fundamental challenge in software engineering. While the proliferation of agentic large language models (LLMs) has vastly improved automated repository-level debugging, current frameworks hit a ceiling when dealing with sophisticated bugs like implicit type degradations and complex polymorphic control flows. Because these methods rely heavily on static analysis and superficial execution feedback, they lack visibility into intermediate runtime states. Consequently, agents are forced into costly, speculative trial-and-error loops, wasting computational tokens without successfully isolating the root cause. To bridge this gap, we propose DAIRA (Dynamic Analysis-enhanced Issue Resolution Agent), a pioneering automated repair framework that natively embeds dynamic analysis into the agent's reasoning cycle. Driven by a Test Tracing-Driven methodology, DAIRA utilizes lightweight monitors to extract critical runtime data -- such as variable mutations and call stacks -- and synthesizes them into structured semantic reports. This mechanism fundamentally shifts the agent's behavior from blind guesswork to evidence-based, deterministic deduction. When powered by Gemini 3 Flash Preview, DAIRA establishes a new state-of-the-art (SOTA) performance, achieving a 79.4% resolution rate on the SWE-bench Verified dataset. Compared to existing baselines, our framework not only conquers highly complex defects but also cuts overall inference expenses by roughly 10% and decreases input token consumption by approximately 25%.

Ruimin Ma, Ruitao Xie, Yanlin Wang et al.

Autism spectrum disorder (ASD) is a highly disabling mental disease that brings significant impairments of social interaction ability to the patients, making early screening and intervention of ASD critical. With the development of the machine learning and neuroimaging technology, extensive research has been conducted on machine classification of ASD based on structural Magnetic Resonance Imaging (s-MRI). However, most studies involve with datasets where participants' age are above 5 and lack interpretability. In this paper, we propose a machine learning method for ASD classification in children with age range from 0.92 to 4.83 years, based on s-MRI features extracted using contrastive variational autoencoder (CVAE). 78 s-MRIs, collected from Shenzhen Children's Hospital, are used for training CVAE, which consists of both ASD-specific feature channel and common shared feature channel. The ASD participants represented by ASD-specific features can be easily discriminated from TC participants represented by the common shared features. In case of degraded predictive accuracy when data size is extremely small, a transfer learning strategy is proposed here as a potential solution. Finally, we conduct neuroanatomical interpretation based on the correlation between s-MRI features extracted from CVAE and surface area of different cortical regions, which discloses potential biomarkers that could help target treatments of ASD in the future.

Yanlin Wang, Di Yuan, Shani Dettman et al.

Cochlear implants (CI) significantly improve spoken language in children with severe-to-profound sensorineural hearing loss (SNHL), yet outcomes remain more variable than in children with normal hearing. This variability cannot be reliably predicted for individual children using age at implantation or residual hearing. This study aims to compare the accuracy of traditional machine learning (ML) to deep transfer learning (DTL) algorithms to predict post-CI spoken language development of children with bilateral SNHL using a binary classification model of high versus low language improvers. A total of 278 implanted children enrolled from three centers. The accuracy, sensitivity and specificity of prediction models based upon brain neuroanatomic features using traditional ML and DTL learning. DTL prediction models using bilinear attention-based fusion strategy achieved: accuracy of 92.39% (95% CI, 90.70%-94.07%), sensitivity of 91.22% (95% CI, 89.98%-92.47%), specificity of 93.56% (95% CI, 90.91%-96.21%), and area under the curve (AUC) of 0.977 (95% CI, 0.969-0.986). DTL outperformed traditional ML models in all outcome measures. DTL was significantly improved by direct capture of discriminative and task-specific information that are advantages of representation learning enabled by this approach over ML. The results support the feasibility of a single DTL prediction model for language prediction of children served by CI programs worldwide.

Shengbo Wang, Mingwei Liu, Guangsheng Ou et al.

Automated C-to-Rust migration encounters systemic obstacles when scaling from code snippets to industrial projects, mainly because build context is often unavailable ("dependency hell") and domain-specific evolutionary knowledge is missing. As a result, current LLM-based methods frequently cannot reconstruct precise type definitions under complex build systems or infer idiomatic API correspondences, which in turn leads to hallucinated dependencies and unproductive repair loops.To tackle these issues, we introduce His2Trans, a framework that combines a deterministic, build-aware skeleton with self-evolving knowledge extraction to support stable, incremental migration. On the structural side, His2Trans performs build tracing to create a compilable Project-Level Skeleton Graph, providing a strictly typed environment that separates global verification from local logic generation. On the cognitive side, it derives fine-grained API and code-fragment rules from historical migration traces and uses a Retrieval-Augmented Generation (RAG) system to steer the LLM toward idiomatic interface reuse.Experiments on industrial OpenHarmony modules show that His2Trans reaches a 97.51% incremental compilation pass rate, effectively fixing build failures where baselines struggle. On general-purpose benchmarks, it reduces the unsafe code ratio by 25.23 percentage points compared with C2Rust while also lowering warning counts, although cross-domain functional correctness remains challenging. Finally, knowledge accumulation studies demonstrate the framework's evolutionary behavior: by continuously integrating verified patterns, His2Trans cuts repair overhead on unseen tasks by about 60%.

Mingyu Zhao, Xingyu Huang, Ziyu Lyu et al.

Graph learning methods have been extensively applied in diverse application areas. However, what kind of inherent graph properties e.g. graph proximity, graph structural information has been encoded into graph representation learning for downstream tasks is still under-explored. In this paper, we propose a novel graph probing framework (GraphProbe) to investigate and interpret whether the family of graph learning methods has encoded different levels of knowledge in graph representation learning. Based on the intrinsic properties of graphs, we design three probes to systematically investigate the graph representation learning process from different perspectives, respectively the node-wise level, the path-wise level, and the structural level. We construct a thorough evaluation benchmark with nine representative graph learning methods from random walk based approaches, basic graph neural networks and self-supervised graph methods, and probe them on six benchmark datasets for node classification, link prediction and graph classification. The experimental evaluation verify that GraphProbe can estimate the capability of graph representation learning. Remaking results have been concluded: GCN and WeightedGCN methods are relatively versatile methods achieving better results with respect to different tasks.

Sheng Zhang, Hui Li, Yanlin Wang et al.

Code search engine is an essential tool in software development. Many code search methods have sprung up, focusing on the overall ranking performance of code search. In this paper, we study code search from another perspective by analyzing the bias of code search models. Biased code search engines provide poor user experience, even though they show promising overall performance. Due to different development conventions (e.g., prefer long queries or abbreviations), some programmers will find the engine useful, while others may find it hard to get desirable search results. To mitigate biases, we develop a general debiasing framework that employs reranking to calibrate search results. It can be easily plugged into existing engines and handle new code search biases discovered in the future. Experiments show that our framework can effectively reduce biases. Meanwhile, the overall ranking performance of code search gets improved after debiasing.

Lianghong Guo, Yanlin Wang, Caihua Li et al. · tencent-ai

Constructing large-scale datasets for the GitHub issue resolution task is crucial for both training and evaluating the software engineering capabilities of Large Language Models (LLMs). However, the traditional process for creating such benchmarks is notoriously challenging and labor-intensive, particularly in the stages of setting up evaluation environments, grading test outcomes, and validating task instances. In this paper, we propose SWE-Factory, an automated pipeline designed to address these challenges. To tackle these issues, our pipeline integrates three core automated components. First, we introduce SWE-Builder, a multi-agent system that automates evaluation environment construction, which employs four specialized agents that work in a collaborative, iterative loop and leverages an environment memory pool to enhance efficiency. Second, we introduce a standardized, exit-code-based grading method that eliminates the need for manually writing custom parsers. Finally, we automate the fail2pass validation process using these reliable exit code signals. Experiments on 671 issues across four programming languages show that our pipeline can effectively construct valid task instances; for example, with GPT-4.1-mini, our SWE-Builder constructs 269 valid instances at $0.045 per instance, while with Gemini-2.5-flash, it achieves comparable performance at the lowest cost of $0.024 per instance. We also demonstrate that our exit-code-based grading achieves 100% accuracy compared to manual inspection, and our automated fail2pass validation reaches a precision of 0.92 and a recall of 1.00. We hope our automated pipeline will accelerate the collection of large-scale, high-quality GitHub issue resolution datasets for both training and evaluation. Our code and datasets are released at https://github.com/DeepSoftwareAnalytics/swe-factory.

Linxi Liang, Jing Gong, Mingwei Liu et al.

Large Language Models (LLMs) have become pivotal tools for automating code generation in software development. However, these models face significant challenges in producing version-aware code for rapidly evolving languages like Rust, where frequent Application Programming Interfaces (API) changes across versions lead to compatibility issues and correctness errors. Existing benchmarks lack systematic evaluation of how models navigate API transitions, relying on labor-intensive manual curation and offering limited version-specific insights. To address this gap, we present RustEvo, a novel framework for constructing dynamic benchmarks that evaluate the ability of LLMs to adapt to evolving Rust APIs. RustEvo automates dataset creation by synthesizing 588 API changes (380 from Rust standard libraries, 208 from 15 third-party crates) into programming tasks mirroring real-world challenges. These tasks cover four API evolution categories: Stabilizations, Signature Changes, Behavioral Changes, and Deprecations, reflecting their actual distribution in the Rust ecosystem. Experiments on state-of-the-art (SOTA) LLMs reveal significant performance variations: models achieve a 65.8% average success rate on stabilized APIs but only 38.0% on behavioral changes, highlighting difficulties in detecting semantic shifts without signature alterations. Knowledge cutoff dates strongly influence performance, with models scoring 56.1% on before-cutoff APIs versus 32.5% on after-cutoff tasks. Retrieval-Augmented Generation (RAG) mitigates this gap, improving success rates by 13.5% on average for APIs released after model training. Our findings underscore the necessity of our evolution-aware benchmarks to advance the adaptability of LLMs in fast-paced software ecosystems. The framework and the benchmarks are publicly released at https://github.com/SYSUSELab/RustEvo.

Dewu Zheng, Yanlin Wang, Ensheng Shi et al.

With the rapid advancement of large language models (LLMs), extensive research has been conducted to investigate the code generation capabilities of LLMs. However, existing efforts primarily focus on general-domain tasks, leaving LLMs' code generation performance in real-world application domains underexplored. This raises a critical question: can a model's general-domain coding ability reliably represent its ability in specialized domains? In this paper, we introduce DomainCodeBench, a multi-domain code generation benchmark designed to systematically evaluate LLMs across 12 software application domains and 15 programming languages. DomainCodeBench contains 2,400 manually verified tasks with ground truth, human-annotated docstrings, and fine-grained dependency information to ensure more coverage of domain-specific challenges. Specifically, we first identify the most popular application domains by topic mining. Then, we curate coding tasks based on commonly used frameworks and platforms in each domain. We obtain several findings through extensive experiments on DomainCodeBench with ten mainstream LLMs. (1) Performance decoupling: experiments reveal that top general-domain models do not consistently excel in specific application domains; (2) Domain-specific weaknesses: LLMs often fail due to domain knowledge gaps and third-party library misusage; (3) Contextual enhancement: we show that augmenting prompts with domain-specific knowledge improves performance by around 38.17%, providing actionable insights for performance optimization. Our replication package, including the benchmark, source code, and experimental results, is available at https://github.com/DeepSoftwareAnalytics/DomainCodeBench.

Zekai Zhang, Mingwei Liu, Zhenxi Chen et al.

Code editing plays a vital role in software engineering, requiring developers to adjust existing code according to natural language instructions while keeping functionality intact and avoiding unnecessary modifications. However, commit-based datasets commonly used for this task are often noisy, lack diversity, and fail to reflect the style of real-world edit instructions. To address this, we introduce OpenCodeEdit, an open-source pipeline that leverages multiple LLMs to synthesize realistic code-edit triplets. The pipeline produces both concise "lazy" instructions and more detailed "descriptive" ones, and applies filtering based on diffs and topics to guarantee data quality and variety. Using this process, we construct OCEDataFT, a curated dataset of 20K samples. Fine-tuning three advanced base models on OCEDataFT leads to significant performance boosts on the CanItEdit benchmark, with relative pass@1 improvements ranging from 4.50% to 20.79%. Notably, the resulting models achieve performance close to closed-source systems, narrowing the gap to GPT-4 to just 3.54%, without relying on proprietary resources or manual annotation.

Jing Gong, Yanghui Wu, Linxi Liang et al.

Semantic code search, retrieving code that matches a given natural language query, is an important task to improve productivity in software engineering. Existing code search datasets face limitations: they rely on human annotators who assess code primarily through semantic understanding rather than functional verification, leading to potential inaccuracies and scalability issues. Additionally, current evaluation metrics often overlook the multi-choice nature of code search. This paper introduces CoSQA+, pairing high-quality queries from CoSQA with multiple suitable codes. We develop an automated pipeline featuring multiple model-based candidate selections and the novel test-driven agent annotation system. Among a single Large Language Model (LLM) annotator and Python expert annotators (without test-based verification), agents leverage test-based verification and achieve the highest accuracy of 93.9%. Through extensive experiments, CoSQA+ has demonstrated superior quality over CoSQA. Models trained on CoSQA+ exhibit improved performance. We publicly release both CoSQA+_all, which contains 412,080 agent-annotated pairs, and CoSQA+_verified, which contains 1,000 human-verified pairs, at https://github.com/DeepSoftwareAnalytics/CoSQA_Plus.

Wei Tao, Yucheng Zhou, Yanlin Wang et al.

Commit messages are natural language descriptions of code changes, which are important for software evolution such as code understanding and maintenance. However, previous methods are trained on the entire dataset without considering the fact that a portion of commit messages adhere to good practice (i.e., good-practice commits), while the rest do not. On the basis of our empirical study, we discover that training on good-practice commits significantly contributes to the commit message generation. Motivated by this finding, we propose a novel knowledge-aware denoising learning method called KADEL. Considering that good-practice commits constitute only a small proportion of the dataset, we align the remaining training samples with these good-practice commits. To achieve this, we propose a model that learns the commit knowledge by training on good-practice commits. This knowledge model enables supplementing more information for training samples that do not conform to good practice. However, since the supplementary information may contain noise or prediction errors, we propose a dynamic denoising training method. This method composes a distribution-aware confidence function and a dynamic distribution list, which enhances the effectiveness of the training process. Experimental results on the whole MCMD dataset demonstrate that our method overall achieves state-of-the-art performance compared with previous methods. Our source code and data are available at https://github.com/DeepSoftwareAnalytics/KADEL

Wanjun Zhong, Lianghong Guo, Qiqi Gao et al.

Revolutionary advancements in Large Language Models have drastically reshaped our interactions with artificial intelligence systems. Despite this, a notable hindrance remains-the deficiency of a long-term memory mechanism within these models. This shortfall becomes increasingly evident in situations demanding sustained interaction, such as personal companion systems and psychological counseling. Therefore, we propose MemoryBank, a novel memory mechanism tailored for LLMs. MemoryBank enables the models to summon relevant memories, continually evolve through continuous memory updates, comprehend, and adapt to a user personality by synthesizing information from past interactions. To mimic anthropomorphic behaviors and selectively preserve memory, MemoryBank incorporates a memory updating mechanism, inspired by the Ebbinghaus Forgetting Curve theory, which permits the AI to forget and reinforce memory based on time elapsed and the relative significance of the memory, thereby offering a human-like memory mechanism. MemoryBank is versatile in accommodating both closed-source models like ChatGPT and open-source models like ChatGLM. We exemplify application of MemoryBank through the creation of an LLM-based chatbot named SiliconFriend in a long-term AI Companion scenario. Further tuned with psychological dialogs, SiliconFriend displays heightened empathy in its interactions. Experiment involves both qualitative analysis with real-world user dialogs and quantitative analysis with simulated dialogs. In the latter, ChatGPT acts as users with diverse characteristics and generates long-term dialog contexts covering a wide array of topics. The results of our analysis reveal that SiliconFriend, equipped with MemoryBank, exhibits a strong capability for long-term companionship as it can provide emphatic response, recall relevant memories and understand user personality.

Wei Tao, Yucheng Zhou, Yanlin Wang et al. · tencent-ai

In software development, resolving the emergent issues within GitHub repositories is a complex challenge that involves not only the incorporation of new code but also the maintenance of existing code. Large Language Models (LLMs) have shown promise in code generation but face difficulties in resolving Github issues, particularly at the repository level. To overcome this challenge, we empirically study the reason why LLMs fail to resolve GitHub issues and analyze the major factors. Motivated by the empirical findings, we propose a novel LLM-based Multi-Agent framework for GitHub Issue reSolution, MAGIS, consisting of four agents customized for software evolution: Manager, Repository Custodian, Developer, and Quality Assurance Engineer agents. This framework leverages the collaboration of various agents in the planning and coding process to unlock the potential of LLMs to resolve GitHub issues. In experiments, we employ the SWE-bench benchmark to compare MAGIS with popular LLMs, including GPT-3.5, GPT-4, and Claude-2. MAGIS can resolve 13.94% GitHub issues, significantly outperforming the baselines. Specifically, MAGIS achieves an eight-fold increase in resolved ratio over the direct application of GPT-4, the advanced LLM.

Yanli Wang, Yanlin Wang, Suiquan Wang et al.

Repository-level code translation refers to translating an entire code repository from one programming language to another while preserving the functionality of the source repository. Many benchmarks have been proposed to evaluate the performance of such code translators. However, previous benchmarks mostly provide fine-grained samples, focusing at either code snippet, function, or file-level code translation. Such benchmarks do not accurately reflect real-world demands, where entire repositories often need to be translated, involving longer code length and more complex functionalities. To address this gap, we propose a new benchmark, named RepoTransBench, which is a real-world repository-level code translation benchmark with an automatically executable test suite. We conduct experiments on RepoTransBench to evaluate the translation performance of 11 advanced LLMs. We find that the Success@1 score (test success in one attempt) of the best-performing LLM is only 7.33%. To further explore the potential of LLMs for repository-level code translation, we provide LLMs with error-related feedback to perform iterative debugging and observe an average 7.09% improvement on Success@1. However, even with this improvement, the Success@1 score of the best-performing LLM is only 21%, which may not meet the need for reliable automatic repository-level code translation. Finally, we conduct a detailed error analysis and highlight current LLMs' deficiencies in repository-level code translation, which could provide a reference for further improvements.

Yanlin Wang, Nancy M Young, Patrick C M Wong

Neural prediction offers a promising approach to forecasting the individual variability of neurocognitive functions and disorders and providing prognostic indicators for personalized invention. However, it is challenging to translate neural predictive models into medical artificial intelligent applications due to the limitations of domain shift and label scarcity. Here, we propose the directed evolution model (DEM), a novel computational model that mimics the trial-and-error processes of biological directed evolution to approximate optimal solutions for predictive modeling tasks. We demonstrated that the directed evolution algorithm is an effective strategy for uncertainty exploration, enhancing generalization in reinforcement learning. Furthermore, by incorporating replay buffer and continual backpropagate methods into DEM, we provide evidence of achieving better trade-off between exploitation and exploration in continuous learning settings. We conducted experiments on four different datasets for children with cochlear implants whose spoken language developmental outcomes vary considerably on the individual-child level. Preoperative neural MRI data has shown to accurately predict the post-operative outcome of these children within but not across datasets. Our results show that DEM can efficiently improve the performance of cross-domain pre-implantation neural predictions while addressing the challenge of label scarcity in target domain.

Mingwei Liu, Zheng Pei, Yanlin Wang et al.

In the context of software frameworks with limited resources (such as HarmonyOS), large language models (LLMs) often exhibit poor code generation performance because they lack sufficient exposure to such environments during pre-training. Although LLMs can usually maintain correct logical structures across programming languages, they frequently struggle when dealing with framework-specific APIs or syntax, resulting in errors. This indicates that while pre-training equips LLMs with general algorithmic capabilities, they remain unfamiliar with the distinctive syntax and API usage of underrepresented frameworks. As a result, even advanced commercial models like GPT-4o cannot reliably generate correct code without prior adaptation. To address this issue, we propose APIKG4SYN, a framework designed to exploit API knowledge graphs for the construction of API-oriented question-code pairs, specifically tailored for low-resource frameworks without requiring executable code. APIKG4SYN integrates both single-API and multi-API knowledge, where the latter is derived through uncertainty estimation (UE)-driven Monte Carlo Tree Search (MCTS), enabling the creation of a diverse and informative dataset for fine-tuning LLMs. Using HarmonyOS as a case study, we build the first benchmark for HarmonyOS code generation. Experimental results show that fine-tuning Qwen with APIKG4SYN raises pass@1 accuracy to 25.00%, compared with 17.59% for the baseline GPT model. These results confirm that API-oriented data significantly enhance LLM performance in low-resource software development scenarios.

Zengbin Wang, Feng Xiong, Liang Lin et al.

Reinforcement learning with verifiable rewards (RLVR) has significantly advanced the reasoning ability of vision-language models (VLMs). However, the inherent text-dominated nature of VLMs often leads to insufficient visual faithfulness, characterized by sparse attention activation to visual tokens. More importantly, our empirical analysis reveals that temporal visual forgetting along reasoning steps exacerbates this deficiency. To bridge this gap, we propose Visually-Guided Policy Optimization (VGPO), a novel framework to reinforce visual focus during policy optimization. Specifically, VGPO initially introduces a Visual Attention Compensation mechanism that leverages visual similarity to localize and amplify visual cues, while progressively elevating visual expectations in later steps to counteract visual forgetting. Building on this mechanism, we implement a dual-grained advantage re-weighting strategy: the intra-trajectory level highlights tokens exhibiting relatively high visual activation, while the inter-trajectory level prioritizes trajectories demonstrating superior visual accumulation. Extensive experiments demonstrate that VGPO achieves better visual activation and superior performance in mathematical multimodal reasoning and visual-dependent tasks.

Jianqiao Lu, Wanjun Zhong, Yufei Wang et al.

Although large language models (LLMs) have demonstrated adeptness in a range of tasks, they still lag behind human learning efficiency. This disparity is often linked to the inherent human capacity to learn from basic examples, gradually generalize and handle more complex problems, and refine their skills with continuous feedback. Inspired by this, this paper introduces YODA, a novel teacher-student progressive learning framework that emulates the teacher-student education process to improve the efficacy of model fine-tuning. The framework operates on an interactive \textit{basic-generalized-harder} loop. The teacher agent provides tailored feedback on the student's answers, and systematically organizes the education process. This process unfolds by teaching the student basic examples, reinforcing understanding through generalized questions, and then enhancing learning by posing questions with progressively enhanced complexity. With the teacher's guidance, the student learns to iteratively refine its answer with feedback, and forms a robust and comprehensive understanding of the posed questions. The systematic procedural data, which reflects the progressive learning process of humans, is then utilized for model training. Taking math reasoning as a testbed, experiments show that training LLaMA2 with data from YODA improves SFT with significant performance gain (+17.01\% on GSM8K and +9.98\% on MATH). In addition, we find that training with curriculum learning further improves learning robustness.

Yanlin Wang, Kefeng Duan, Dewu Zheng et al.

Code intelligence is an emerging domain in software engineering, aiming to improve the effectiveness and efficiency of various code-related tasks. Recent research suggests that incorporating contextual information beyond the basic original task inputs (i.e., source code) can substantially enhance model performance. Such contextual signals may be obtained directly or indirectly from sources such as API documentation or intermediate representations like abstract syntax trees can significantly improve the effectiveness of code intelligence. Despite growing academic interest, there is a lack of systematic analysis of context in code intelligence. To address this gap, we conduct an extensive literature review of 146 relevant studies published between September 2007 and August 2024. Our investigation yields four main contributions. (1) A quantitative analysis of the research landscape, including publication trends, venues, and the explored domains; (2) A novel taxonomy of context types used in code intelligence; (3) A task-oriented analysis investigating context integration strategies across diverse code intelligence tasks; (4) A critical evaluation of evaluation methodologies for context-aware methods. Based on these findings, we identify fundamental challenges in context utilization in current code intelligence systems and propose a research roadmap that outlines key opportunities for future research.

Shengbo Wang, Mingwei Liu, Zike Li et al.

The rapid advancement of Large Language Models (LLMs) poses a significant challenge to existing mathematical reasoning benchmarks. However, these benchmarks tend to become easier over time as LLMs can learn from the published benchmarks. This limitation hinder the precise evaluation of the true capabilities of SOTA models. To address this challenge, this paper introduces EvolMathEval, an automated mathematical benchmark generation and evolution framework based on evolutionary testing. Experimental results demonstrate that EvolMathEval can not only generate a large volume of high-difficulty problems through continuous self-iteration, but it can also significantly enhance the complexity of public datasets like GSM8K through evolution, reducing model accuracy by an average of 48\%. Deeper investigation reveals that when solving these evolved problems, LLMs tend to bypass complex multi-step logical reasoning by relying on simplistic and fuzzy conditions, consequently leading to incorrect solutions. We define this phenomenon as the ``Pseudo Aha Moment", which we find accounts for 77\% to 100\% of errors on targeted problems. Code and resources are available at: https://anonymous.4open.science/r/EvolMathEval

Wenchao Gu, Zongyi Lyu, Yanlin Wang et al.

Code retrieval aims to provide users with desired code snippets based on users' natural language queries. With the development of deep learning technologies, adopting pre-trained models for this task has become mainstream. Considering the retrieval efficiency, most of the previous approaches adopt a dual-encoder for this task, which encodes the description and code snippet into representation vectors, respectively. However, the model structure of the dual-encoder tends to limit the model's performance, since it lacks the interaction between the code snippet and description at the bottom layer of the model during training. To improve the model's effectiveness while preserving its efficiency, we propose a framework, which adopts Self-AdaPtive Model Distillation for Efficient CodE Retrieval, named SPENCER. SPENCER first adopts the dual-encoder to narrow the search space and then adopts the cross-encoder to improve accuracy. To improve the efficiency of SPENCER, we propose a novel model distillation technique, which can greatly reduce the inference time of the dual-encoder while maintaining the overall performance. We also propose a teaching assistant selection strategy for our model distillation, which can adaptively select the suitable teaching assistant models for different pre-trained models during the model distillation to ensure the model performance. Extensive experiments demonstrate that the combination of dual-encoder and cross-encoder improves overall performance compared to solely dual-encoder-based models for code retrieval. Besides, our model distillation technique retains over 98% of the overall performance while reducing the inference time of the dual-encoder by 70%.

Yupeng Qi, Ziyu Lyu, Min Yang et al.

As large language models (LLMs) are increasingly applied across various domains, enhancing safety while maintaining the helpfulness of LLMs has become a critical challenge. Recent studies solve this problem through safety-constrained online preference optimization or safety-constrained offline preference optimization. However, the safety-constrained online methods often suffer from excessive safety, which might reduce helpfulness, while the safety-constrained offline methods perform poorly in adaptively balancing safety and helpfulness. To address these limitations, we propose MidPO, a \textbf{\underline{Mi}}xture of Experts (MoE) framework for safety-helpfulness \textbf{\underline{d}}ual \textbf{\underline{P}}reference \textbf{\underline{O}}ptimization. Firstly, MidPO devises single-preference enhanced direct preference optimization approach to transform the base model into two independent experts, termed safety and helpfulness experts, and fine-tunes the two independent experts for optimal safety or helpfulness performance. Secondly, to achieve an effective balance between safety and helpfulness, MidPO incorporates the two experts into the MoE framework and designs a dynamic routing mechanism to allocate contributions from each expert adaptively. We conduct quantitative and qualitative experiments on three popular datasets to demonstrate the proposed MidPO significantly outperforms state-of-the-art approaches in both safety and helpfulness. The code and models will be released.