Aixin Sun

Xiang Li, Yiqun Yao, Xin Jiang et al. · tencent-ai, tsinghua

Large language models (LLMs) are considered important approaches towards foundational machine intelligence, achieving remarkable success in Natural Language Processing and multimodal tasks, among others. However, the carbon footprints and financial costs originating from heavy pre-training computation is a non-negligible issue. Progressive training methods, inspired by the neurogenesis process that grows neural structures, have shown potential to accelerate LLM pre-training. However, the algorithms, implementation, and practices for progressively training LLMs beyond 100B parameters remain underexplored. In this paper, we show that our model, namely FLM-101B, trained with our growth strategy under a budget of \$100K, reaches 80\% of the baselines' performances with only 10\% of their floating-point operations. We believe that further studies on progressive training will benefit the community by cutting down the costs and promoting green AI. The checkpoint of FLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.

Yubo Ma, Yuhang Zang, Liangyu Chen et al. · pku, stanford

Understanding documents with rich layouts and multi-modal components is a long-standing and practical task. Recent Large Vision-Language Models (LVLMs) have made remarkable strides in various tasks, particularly in single-page document understanding (DU). However, their abilities on long-context DU remain an open problem. This work presents MMLongBench-Doc, a long-context, multi-modal benchmark comprising 1,062 expert-annotated questions. Distinct from previous datasets, it is constructed upon 130 lengthy PDF-formatted documents with an average of 49.4 pages and 20,971 textual tokens. Towards comprehensive evaluation, answers to these questions rely on pieces of evidence from (1) different sources (text, image, chart, table, and layout structure) and (2) various locations (i.e. page number). Moreover, 33.2% of the questions are cross-page questions requiring evidence across multiple pages. 22.8% of the questions are designed to be unanswerable for detecting potential hallucinations. Experiments on 14 LVLMs demonstrate that long-context DU greatly challenges current models. Notably, the best-performing model, GPT-4o, achieves an F1 score of only 42.7%, while the second-best, GPT-4V, scores 31.4%. Furthermore, 12 LVLMs (all except GPT-4o and GPT-4V) even present worse performance than their LLM counterparts which are fed with lossy-parsed OCR documents. These results validate the necessity of future research toward more capable long-context LVLMs. Project Page: https://mayubo2333.github.io/MMLongBench-Doc

Mathieu Ravaut, Aixin Sun, Nancy F. Chen et al.

Large language models (LLMs) excel in abstractive summarization tasks, delivering fluent and pertinent summaries. Recent advancements have extended their capabilities to handle long-input contexts, exceeding 100k tokens. However, in question answering, language models exhibit uneven utilization of their input context. They tend to favor the initial and final segments, resulting in a U-shaped performance pattern concerning where the answer is located within the input. This bias raises concerns, particularly in summarization where crucial content may be dispersed throughout the source document(s). Besides, in summarization, mapping facts from the source to the summary is not trivial as salient content is usually re-phrased. In this paper, we conduct the first comprehensive study on context utilization and position bias in summarization. Our analysis encompasses 6 LLMs, 10 datasets, and 5 evaluation metrics. We introduce a new evaluation benchmark called MiddleSum on the which we benchmark two alternative inference methods to alleviate position bias: hierarchical summarization and incremental summarization. Our code and data can be found here: https://github.com/ntunlp/MiddleSum.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research.

Xin Jiang, Xiang Li, Wenjia Ma et al.

Large language models (LLMs) represented by GPT family have achieved remarkable success. The characteristics of LLMs lie in their ability to accommodate a wide range of tasks through a generative approach. However, the flexibility of their output format poses challenges in controlling and harnessing the model's outputs, thereby constraining the application of LLMs in various domains. In this work, we present Sketch, an innovative toolkit designed to streamline LLM operations across diverse fields. Sketch comprises the following components: (1) a suite of task description schemas and prompt templates encompassing various NLP tasks; (2) a user-friendly, interactive process for building structured output LLM services tailored to various NLP tasks; (3) an open-source dataset for output format control, along with tools for dataset construction; and (4) an open-source model based on LLaMA3-8B-Instruct that adeptly comprehends and adheres to output formatting instructions. We anticipate this initiative to bring considerable convenience to LLM users, achieving the goal of ''plug-and-play'' for various applications. The components of Sketch will be progressively open-sourced at https://github.com/cofe-ai/Sketch.

Yiqun Yao, Wenjia Ma, Xuezhi Fang et al.

Controlling the format of outputs generated by large language models (LLMs) is a critical functionality in various applications. Current methods typically employ constrained decoding with rule-based automata or fine-tuning with manually crafted format instructions, both of which struggle with open-domain format requirements. To address this limitation, we introduce a novel framework for controlled generation in LLMs, leveraging user-provided, one-shot QA pairs. This study investigates LLMs' capabilities to follow open-domain, one-shot constraints and replicate the format of the example answers. We observe that this is a non-trivial problem for current LLMs. We also develop a dataset collection methodology for supervised fine-tuning that enhances the open-domain format control of LLMs without degrading output quality, as well as a benchmark on which we evaluate both the helpfulness and format correctness of LLM outputs. The resulting datasets, named OIFC-SFT, along with the related code, will be made publicly available at https://github.com/cofe-ai/OIFC.

Renjie Liang, Li Li, Chongzhi Zhang et al.

In this paper, we propose the task of \textit{Ranked Video Moment Retrieval} (RVMR) to locate a ranked list of matching moments from a collection of videos, through queries in natural language. Although a few related tasks have been proposed and studied by CV, NLP, and IR communities, RVMR is the task that best reflects the practical setting of moment search. To facilitate research in RVMR, we develop the TVR-Ranking dataset, based on the raw videos and existing moment annotations provided in the TVR dataset. Our key contribution is the manual annotation of relevance levels for 94,442 query-moment pairs. We then develop the $NDCG@K, IoU\geq μ$ evaluation metric for this new task and conduct experiments to evaluate three baseline models. Our experiments show that the new RVMR task brings new challenges to existing models and we believe this new dataset contributes to the research on multi-modality search. The dataset is available at \url{https://github.com/Ranking-VMR/TVR-Ranking}

Yubo Ma, Yixin Cao, YongChing Hong et al.

Large Language Models (LLMs) have made remarkable strides in various tasks. Whether LLMs are competitive few-shot solvers for information extraction (IE) tasks, however, remains an open problem. In this work, we aim to provide a thorough answer to this question. Through extensive experiments on nine datasets across four IE tasks, we demonstrate that current advanced LLMs consistently exhibit inferior performance, higher latency, and increased budget requirements compared to fine-tuned SLMs under most settings. Therefore, we conclude that LLMs are not effective few-shot information extractors in general. Nonetheless, we illustrate that with appropriate prompting strategies, LLMs can effectively complement SLMs and tackle challenging samples that SLMs struggle with. And moreover, we propose an adaptive filter-then-rerank paradigm to combine the strengths of LLMs and SLMs. In this paradigm, SLMs serve as filters and LLMs serve as rerankers. By prompting LLMs to rerank a small portion of difficult samples identified by SLMs, our preliminary system consistently achieves promising improvements (2.4% F1-gain on average) on various IE tasks, with an acceptable time and cost investment.

Yequan Wang, Xuying Meng, Yiyi Liu et al. · tencent-ai, tsinghua

Many studies on dialog emotion analysis focus on utterance-level emotion only. These models hence are not optimized for dialog-level emotion detection, i.e. to predict the emotion category of a dialog as a whole. More importantly, these models cannot benefit from the context provided by the whole dialog. In real-world applications, annotations to dialog could fine-grained, including both utterance-level tags (e.g. speaker type, intent category, and emotion category), and dialog-level tags (e.g. user satisfaction, and emotion curve category). In this paper, we propose a Context-based Hierarchical Attention Capsule~(Chat-Capsule) model, which models both utterance-level and dialog-level emotions and their interrelations. On a dialog dataset collected from customer support of an e-commerce platform, our model is also able to predict user satisfaction and emotion curve category. Emotion curve refers to the change of emotions along the development of a conversation. Experiments show that the proposed Chat-Capsule outperform state-of-the-art baselines on both benchmark dataset and proprietary dataset. Source code will be released upon acceptance.

Chen Chen, Yufei Wang, Aixin Sun et al.

Knowledge Graph Completion (KGC) often requires both KG structural and textual information to be effective. Pre-trained Language Models (PLMs) have been used to learn the textual information, usually under the fine-tune paradigm for the KGC task. However, the fine-tuned PLMs often overwhelmingly focus on the textual information and overlook structural knowledge. To tackle this issue, this paper proposes CSProm-KG (Conditional Soft Prompts for KGC) which maintains a balance between structural information and textual knowledge. CSProm-KG only tunes the parameters of Conditional Soft Prompts that are generated by the entities and relations representations. We verify the effectiveness of CSProm-KG on three popular static KGC benchmarks WN18RR, FB15K-237 and Wikidata5M, and two temporal KGC benchmarks ICEWS14 and ICEWS05-15. CSProm-KG outperforms competitive baseline models and sets new state-of-the-art on these benchmarks. We conduct further analysis to show (i) the effectiveness of our proposed components, (ii) the efficiency of CSProm-KG, and (iii) the flexibility of CSProm-KG.

Xinyu Zou, Zhi Hu, Yiming Zhao et al.

Multi-scenario learning (MSL) enables a service provider to cater for users' fine-grained demands by separating services for different user sectors, e.g., by user's geographical region. Under each scenario there is a need to optimize multiple task-specific targets e.g., click through rate and conversion rate, known as multi-task learning (MTL). Recent solutions for MSL and MTL are mostly based on the multi-gate mixture-of-experts (MMoE) architecture. MMoE structure is typically static and its design requires domain-specific knowledge, making it less effective in handling both MSL and MTL. In this paper, we propose a novel Automatic Expert Selection framework for Multi-scenario and Multi-task search, named AESM^{2}. AESM^{2} integrates both MSL and MTL into a unified framework with an automatic structure learning. Specifically, AESM^{2} stacks multi-task layers over multi-scenario layers. This hierarchical design enables us to flexibly establish intrinsic connections between different scenarios, and at the same time also supports high-level feature extraction for different tasks. At each multi-scenario/multi-task layer, a novel expert selection algorithm is proposed to automatically identify scenario-/task-specific and shared experts for each input. Experiments over two real-world large-scale datasets demonstrate the effectiveness of AESM^{2} over a battery of strong baselines. Online A/B test also shows substantial performance gain on multiple metrics. Currently, AESM^{2} has been deployed online for serving major traffic.

Shaowen Zhou, Bowen Yu, Aixin Sun et al.

Open Information Extraction (OpenIE) facilitates domain-independent discovery of relational facts from large corpora. The technique well suits many open-world natural language understanding scenarios, such as automatic knowledge base construction, open-domain question answering, and explicit reasoning. Thanks to the rapid development in deep learning technologies, numerous neural OpenIE architectures have been proposed and achieve considerable performance improvement. In this survey, we provide an extensive overview of the-state-of-the-art neural OpenIE models, their key design decisions, strengths and weakness. Then, we discuss limitations of current solutions and the open issues in OpenIE problem itself. Finally we list recent trends that could help expand its scope and applicability, setting up promising directions for future research in OpenIE. To our best knowledge, this paper is the first review on this specific topic.

Xinze Li, Yixin Cao, Liangming Pan et al.

Although achieving great success, Large Language Models (LLMs) usually suffer from unreliable hallucinations. Although language attribution can be a potential solution, there are no suitable benchmarks and evaluation metrics to attribute LLMs to structured knowledge. In this paper, we define a new task of Knowledge-aware Language Model Attribution (KaLMA) that improves upon three core concerns with conventional attributed LMs. First, we extend attribution source from unstructured texts to Knowledge Graph (KG), whose rich structures benefit both the attribution performance and working scenarios. Second, we propose a new ``Conscious Incompetence" setting considering the incomplete knowledge repository, where the model identifies the need for supporting knowledge beyond the provided KG. Third, we propose a comprehensive automatic evaluation metric encompassing text quality, citation quality, and text citation alignment. To implement the above innovations, we build a dataset in biography domain BioKaLMA via evolutionary question generation strategy, to control the question complexity and necessary knowledge to the answer. For evaluation, we develop a baseline solution and demonstrate the room for improvement in LLMs' citation generation, emphasizing the importance of incorporating the "Conscious Incompetence" setting, and the critical role of retrieval accuracy.

Yu-chen Fan, Yitong Ji, Jie Zhang et al.

A typical benchmark dataset for recommender system (RecSys) evaluation consists of user-item interactions generated on a platform within a time period. The interaction generation mechanism partially explains why a user interacts with (e.g., like, purchase, rate) an item, and the context of when a particular interaction happened. In this study, we conduct a meticulous analysis of the MovieLens dataset and explain the potential impact of using the dataset for evaluating recommendation algorithms. We make a few main findings from our analysis. First, there are significant differences in user interactions at the different stages when a user interacts with the MovieLens platform. The early interactions largely define the user portrait which affects the subsequent interactions. Second, user interactions are highly affected by the candidate movies that are recommended by the platform's internal recommendation algorithm(s). Third, changing the order of user interactions makes it more difficult for sequential algorithms to capture the progressive interaction process. We further discuss the discrepancy between the interaction generation mechanism that is employed by the MovieLens system and that of typical real-world recommendation scenarios. In summary, the MovieLens platform demonstrates an efficient and effective way of collecting user preferences to address cold-starts. However, models that achieve excellent recommendation accuracy on the MovieLens dataset may not demonstrate superior performance in practice, for at least two kinds of differences: (i) the differences in the contexts of user-item interaction generation, and (ii) the differences in user knowledge about the item collections. While results on MovieLens can be useful as a reference, they should not be solely relied upon as the primary justification for the effectiveness of a recommendation system model.

Fuzhao Xue, Jianghai Chen, Aixin Sun et al.

Transformer-based models have delivered impressive results on many tasks, particularly vision and language tasks. In many model training situations, conventional configurations are typically adopted. For example, we often set the base model with hidden dimensions (i.e. model width) to be 768 and the number of transformer layers (i.e. model depth) to be 12. In this paper, we revisit these conventional configurations. Through theoretical analysis and experimental evaluation, we show that the masked autoencoder is effective in alleviating the over-smoothing issue in deep transformer training. Based on this finding, we propose Bamboo, an idea of using deeper and narrower transformer configurations, for masked autoencoder training. On ImageNet, with such a simple change in configuration, re-designed model achieves 87.1% top-1 accuracy and outperforms SoTA models like MAE and BEiT. On language tasks, re-designed model outperforms BERT with default setting by 1.1 points on average, on GLUE datasets.

Kuicai Dong, Aixin Sun, Jung-Jae Kim et al.

Open Information Extraction (OpenIE) aims to extract relational tuples from open-domain sentences. Traditional rule-based or statistical models have been developed based on syntactic structures of sentences, identified by syntactic parsers. However, previous neural OpenIE models under-explore the useful syntactic information. In this paper, we model both constituency and dependency trees into word-level graphs, and enable neural OpenIE to learn from the syntactic structures. To better fuse heterogeneous information from both graphs, we adopt multi-view learning to capture multiple relationships from them. Finally, the finetuned constituency and dependency representations are aggregated with sentential semantic representations for tuple generation. Experiments show that both constituency and dependency information, and the multi-view learning are effective.

Yequan Wang, Hengran Zhang, Aixin Sun et al.

Given comparative text, comparative relation extraction aims to extract two targets (\eg two cameras) in comparison and the aspect they are compared for (\eg image quality). The extracted comparative relations form the basis of further opinion analysis.Existing solutions formulate this task as a sequence labeling task, to extract targets and aspects. However, they cannot directly extract comparative relation(s) from text. In this paper, we show that comparative relations can be directly extracted with high accuracy, by generative model. Based on GPT-2, we propose a Generation-based Comparative Relation Extractor (GCRE-GPT). Experiment results show that \modelname achieves state-of-the-art accuracy on two datasets.

Yequan Wang, Xiang Li, Aixin Sun et al.

Quotation extraction aims to extract quotations from written text. There are three components in a quotation: source refers to the holder of the quotation, cue is the trigger word(s), and content is the main body. Existing solutions for quotation extraction mainly utilize rule-based approaches and sequence labeling models. While rule-based approaches often lead to low recalls, sequence labeling models cannot well handle quotations with complicated structures. In this paper, we propose the Context and Former-Label Enhanced Net (CofeNet) for quotation extraction. CofeNet is able to extract complicated quotations with components of variable lengths and complicated structures. On two public datasets (i.e., PolNeAR and Riqua) and one proprietary dataset (i.e., PoliticsZH), we show that our CofeNet achieves state-of-the-art performance on complicated quotation extraction.

Yequan Wang, Jiawen Deng, Aixin Sun et al.

Recently, amounts of works utilize perplexity~(PPL) to evaluate the quality of the generated text. They suppose that if the value of PPL is smaller, the quality(i.e. fluency) of the text to be evaluated is better. However, we find that the PPL referee is unqualified and it cannot evaluate the generated text fairly for the following reasons: (i) The PPL of short text is larger than long text, which goes against common sense, (ii) The repeated text span could damage the performance of PPL, and (iii) The punctuation marks could affect the performance of PPL heavily. Experiments show that the PPL is unreliable for evaluating the quality of given text. Last, we discuss the key problems with evaluating text quality using language models.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Mathieu Ravaut, Hao Zhang, Lu Xu et al.

Conversational recommender systems (CRS) aim to recommend relevant items to users by eliciting user preference through natural language conversation. Prior work often utilizes external knowledge graphs for items' semantic information, a language model for dialogue generation, and a recommendation module for ranking relevant items. This combination of multiple components suffers from a cumbersome training process, and leads to semantic misalignment issues between dialogue generation and item recommendation. In this paper, we represent items in natural language and formulate CRS as a natural language processing task. Accordingly, we leverage the power of pre-trained language models to encode items, understand user intent via conversation, perform item recommendation through semantic matching, and generate dialogues. As a unified model, our PECRS (Parameter-Efficient CRS), can be optimized in a single stage, without relying on non-textual metadata such as a knowledge graph. Experiments on two benchmark CRS datasets, ReDial and INSPIRED, demonstrate the effectiveness of PECRS on recommendation and conversation. Our code is available at: https://github.com/Ravoxsg/efficient_unified_crs.

Shuai Zhang, Yi Tay, Lina Yao et al.

Deep learning based recommender systems have been extensively explored in recent years. However, the large number of models proposed each year poses a big challenge for both researchers and practitioners in reproducing the results for further comparisons. Although a portion of papers provides source code, they adopted different programming languages or different deep learning packages, which also raises the bar in grasping the ideas. To alleviate this problem, we released the open source project: \textbf{DeepRec}. In this toolkit, we have implemented a number of deep learning based recommendation algorithms using Python and the widely used deep learning package - Tensorflow. Three major recommendation scenarios: rating prediction, top-N recommendation (item ranking) and sequential recommendation, were considered. Meanwhile, DeepRec maintains good modularity and extensibility to easily incorporate new models into the framework. It is distributed under the terms of the GNU General Public License. The source code is available at github: \url{https://github.com/cheungdaven/DeepRec}

Siqi Fan, Xin Jiang, Xiang Li et al.

Due to the large number of parameters, the inference phase of Large Language Models (LLMs) is resource-intensive. However, not all requests posed to LLMs are equally difficult to handle. Through analysis, we show that for some tasks, LLMs can achieve results comparable to the final output at some intermediate layers. That is, not all layers of LLMs are necessary during inference. If we can predict at which layer the inferred results match the final results (produced by evaluating all layers), we could significantly reduce the inference cost. To this end, we propose a simple yet effective algorithm named AdaInfer to adaptively terminate the inference process for an input instance. AdaInfer relies on easily obtainable statistical features and classic classifiers like SVM. Experiments on well-known LLMs like the Llama2 series and OPT, show that AdaInfer can achieve an average of 17.8% pruning ratio, and up to 43% on sentiment tasks, with nearly no performance drop (<1%). Because AdaInfer does not alter LLM parameters, the LLMs incorporated with AdaInfer maintain generalizability across tasks.

Yixin Cao, Shibo Hong, Xinze Li et al.

Large Language Models (LLMs) are advancing at an amazing speed and have become indispensable across academia, industry, and daily applications. To keep pace with the status quo, this survey probes the core challenges that the rise of LLMs poses for evaluation. We identify and analyze two pivotal transitions: (i) from task-specific to capability-based evaluation, which reorganizes benchmarks around core competencies such as knowledge, reasoning, instruction following, multi-modal understanding, and safety; and (ii) from manual to automated evaluation, encompassing dynamic dataset curation and "LLM-as-a-judge" scoring. Yet, even with these transitions, a crucial obstacle persists: the evaluation generalization issue. Bounded test sets cannot scale alongside models whose abilities grow seemingly without limit. We will dissect this issue, along with the core challenges of the above two transitions, from the perspectives of methods, datasets, evaluators, and metrics. Due to the fast evolving of this field, we will maintain a living GitHub repository (links are in each section) to crowd-source updates and corrections, and warmly invite contributors and collaborators.

Xinze Li, Yixin Cao, Yubo Ma et al.

Extending context windows (i.e., Long Context, LC) and using retrievers to selectively access relevant information (i.e., Retrieval-Augmented Generation, RAG) are the two main strategies to enable LLMs to incorporate extremely long external contexts. This paper revisits recent studies on this topic, highlighting their key insights and discrepancies. We then provide a more comprehensive evaluation by filtering out questions answerable without external context, identifying the most effective retrieval methods, and expanding the datasets. We show that LC generally outperforms RAG in question-answering benchmarks, especially for Wikipedia-based questions. Summarization-based retrieval performs comparably to LC, while chunk-based retrieval lags behind. However, RAG has advantages in dialogue-based and general question queries. These insights underscore the trade-offs between RAG and LC strategies, offering guidance for future optimization of LLMs with external knowledge sources. We also provide an in-depth discussion on this topic, highlighting the overlooked importance of context relevance in existing studies.

Jie Zou, Aixin Sun, Cheng Long et al.

In conversational recommender systems (CRSs), conversations usually involve a set of items and item-related entities or attributes, e.g., director is a related entity of a movie. These items and item-related entities are often mentioned along the development of a dialog, leading to potential sequential dependencies among them. However, most of existing CRSs neglect these potential sequential dependencies. In this article, we first propose a Transformer-based sequential conversational recommendation method, named TSCR, to model the sequential dependencies in the conversations to improve CRS. In TSCR, we represent conversations by items and the item-related entities, and construct user sequences to discover user preferences by considering both the mentioned items and item-related entities. Based on the constructed sequences, we deploy a Cloze task to predict the recommended items along a sequence. Meanwhile, in certain domains, knowledge graphs formed by the items and their related entities are readily available, which provide various different kinds of associations among them. Given that TSCR does not benefit from such knowledge graphs, we then propose a knowledge graph enhanced version of TSCR, called TSCRKG. In specific, we leverage the knowledge graph to offline initialize our model TSCRKG, and augment the user sequence of conversations (i.e., sequence of the mentioned items and item-related entities in the conversation) with multi-hop paths in the knowledge graph. Experimental results demonstrate that our TSCR model significantly outperforms state-of-the-art baselines, and the enhanced version TSCRKG further improves recommendation performance on top of TSCR.

Siqi Fan, Bowen Qin, Peng Han et al.

Recent thinking models trained with reinforcement learning and backward-checking CoT often suffer from overthinking: they produce excessively long outputs even on simple problems, wasting computation. Existing evaluations, based on token efficiency, give an incomplete view as they neglect problem difficulty and intermediate computation costs. We formalize reasoning efficiency as a relative measure between thinking and instruct models, treating instruct models as the minimal-effort baseline. A systematic study across four thinking models and multiple benchmarks reveals two consistent patterns: (i) instruct models achieve higher efficiency overall, and (ii) problem difficulty affects efficiency, with thinking models wasting computation on easy problems but providing value on harder ones. Building on this insight, we propose COTHINK, a simple two-stage pipeline: an instruct model drafts a brief outline, and a thinking model expands it. On GSM8K, MATH500, and AIME24, COTHINK cuts token usage by 21.1% while keeping accuracy on four thinking models, and remains competitive with strong efficiency baselines.

Yubo Ma, Jinsong Li, Yuhang Zang et al. · pku

Despite the strong performance of ColPali/ColQwen2 in Visualized Document Retrieval (VDR), it encodes each page into multiple patch-level embeddings and leads to excessive memory usage. This empirical study investigates methods to reduce patch embeddings per page at minimum performance degradation. We evaluate two token-reduction strategies: token pruning and token merging. Regarding token pruning, we surprisingly observe that a simple random strategy outperforms other sophisticated pruning methods, though still far from satisfactory. Further analysis reveals that pruning is inherently unsuitable for VDR as it requires removing certain page embeddings without query-specific information. Turning to token merging (more suitable for VDR), we search for the optimal combinations of merging strategy across three dimensions and develop Light-ColPali/ColQwen2. It maintains 98.2% of retrieval performance with only 11.8% of original memory usage, and preserves 94.6% effectiveness at 2.8% memory footprint. We expect our empirical findings and resulting Light-ColPali/ColQwen2 offer valuable insights and establish a competitive baseline for future research towards efficient VDR.

Jing Wang, Fengzhuo Zhang, Xiaoli Li et al.

A variety of Auto-Regressive Video Diffusion Models (ARVDM) have achieved remarkable successes in generating realistic long-form videos. However, theoretical analyses of these models remain scant. In this work, we develop theoretical underpinnings for these models and use our insights to improve the performance of existing models. We first develop Meta-ARVDM, a unified framework of ARVDMs that subsumes most existing methods. Using Meta-ARVDM, we analyze the KL-divergence between the videos generated by Meta-ARVDM and the true videos. Our analysis uncovers two important phenomena inherent to ARVDM -- error accumulation and memory bottleneck. By deriving an information-theoretic impossibility result, we show that the memory bottleneck phenomenon cannot be avoided. To mitigate the memory bottleneck, we design various network structures to explicitly use more past frames. We also achieve a significantly improved trade-off between the mitigation of the memory bottleneck and the inference efficiency by compressing the frames. Experimental results on DMLab and Minecraft validate the efficacy of our methods. Our experiments also demonstrate a Pareto-frontier between the error accumulation and memory bottleneck across different methods.

Yiqun Yao, Xiang Li, Xin Jiang et al.

Humans naturally process real-world multimodal information in a full-duplex manner. In artificial intelligence, replicating this capability is essential for advancing model development and deployment, particularly in embodied contexts. The development of multimodal models faces two primary challenges: (1) effectively handling more than three modalities-such as vision, audio, and text; and (2) delivering full-duplex responses to rapidly evolving human instructions. To facilitate research on models that support both omnimodal processing and full duplexity, we present RoboEgo (alias: FLM-Ego), a unified model system designed to address both challenges. RoboEgo incorporates a backbone architecture and algorithms that natively support full duplexity, achieving a theoretical duplex latency of 80 ms. In streaming visually grounded conversations under real-world conditions, RoboEgo exhibits superior responsiveness and speech naturalness, while maintaining comparable content qualities to state-of-the-art semi-duplex omnimodal models-a feat previously considered unattainable by native full-duplex systems.

Siqi Fan, Xiusheng Huang, Yiqun Yao et al.

Large language models (LLMs) can carry out human-like dialogue, but unlike humans, they are stateless due to the superposition property. However, during multi-turn, multi-agent interactions, LLMs begin to exhibit consistent, character-like behaviors, hinting at a form of emergent lifelong learning. Despite this, existing benchmarks often fail to capture these dynamics, primarily focusing on static, open-ended evaluations. To address this gap, we introduce LIFESTATE-BENCH, a benchmark designed to assess lifelong learning in LLMs. It features two episodic datasets: Hamlet and a synthetic script collection, rich in narrative structure and character interactions. Our fact checking evaluation probes models' self-awareness, episodic memory retrieval, and relationship tracking, across both parametric and non-parametric approaches. Experiments on models like Llama3.1-8B, GPT-4-turbo, and DeepSeek R1, we demonstrate that nonparametric methods significantly outperform parametric ones in managing stateful learning. However, all models exhibit challenges with catastrophic forgetting as interactions extend, highlighting the need for further advancements in lifelong learning.

Yiqun Yao, Naitong Yu, Xiang Li et al.

We introduce EgoMem, the first lifelong memory agent tailored for full-duplex models that process real-time omnimodal streams. EgoMem enables real-time models to recognize multiple users directly from raw audiovisual streams, to provide personalized response, and to maintain long-term knowledge of users' facts, preferences, and social relationships extracted from audiovisual history. EgoMem operates with three asynchronous processes: (i) a retrieval process that dynamically identifies user via face and voice, and gathers relevant context from a long-term memory; (ii) an omnimodal dialog process that generates personalized audio responses based on the retrieved context; and (iii) a memory management process that automatically detects dialog boundaries from omnimodal streams, and extracts necessary information to update the long-term memory. Unlike existing memory agents for LLMs, EgoMem relies entirely on raw audiovisual streams, making it especially suitable for lifelong, real-time, and embodied scenarios. Experimental results demonstrate that EgoMem's retrieval and memory management modules achieve over 95% accuracy on the test set. When integrated with a fine-tuned RoboEgo omnimodal chatbot, the system achieves fact-consistency scores above 87% in real-time personalized dialogs, establishing a strong baseline for future research.

Qishuai Zhong, Zongmin Li, Siqi Fan et al.

Effective engagement by large language models (LLMs) requires adapting responses to users' sociodemographic characteristics, such as age, occupation, and education level. While many real-world applications leverage dialogue history for contextualization, existing evaluations of LLMs' behavioral adaptation often focus on single-turn prompts. In this paper, we propose a framework to evaluate LLM adaptation when attributes are introduced either (1) explicitly via user profiles in the prompt or (2) implicitly through multi-turn dialogue history. We assess the consistency of model behavior across these modalities. Using a multi-agent pipeline, we construct a synthetic dataset pairing dialogue histories with distinct user profiles and employ questions from the Value Survey Module (VSM 2013) (Hofstede and Hofstede, 2016) to probe value expression. Our findings indicate that most models adjust their expressed values in response to demographic changes, particularly in age and education level, but consistency varies. Models with stronger reasoning capabilities demonstrate greater alignment, indicating the importance of reasoning in robust sociodemographic adaptation.

KM Pooja, Cheng Long, Aixin Sun

The task of entity linking, which involves associating mentions with their respective entities in a knowledge graph, has received significant attention due to its numerous potential applications. Recently, various multimodal entity linking (MEL) techniques have been proposed, targeted to learn comprehensive embeddings by leveraging both text and vision modalities. The selection of high-quality negative samples can potentially play a crucial role in metric/representation learning. However, to the best of our knowledge, this possibility remains unexplored in existing literature within the framework of MEL. To fill this gap, we address the multimodal entity linking problem in a generative adversarial setting where the generator is responsible for generating high-quality negative samples, and the discriminator is assigned the responsibility for the metric learning tasks. Since the generator is involved in generating samples, which is a discrete process, we optimize it using policy gradient techniques and propose a policy gradient-based generative adversarial network for multimodal entity linking (PGMEL). Experimental results based on Wiki-MEL, Richpedia-MEL and WikiDiverse datasets demonstrate that PGMEL learns meaningful representation by selecting challenging negative samples and outperforms state-of-the-art methods.

Guoxiu He, Xin Song, Futing Wang et al.

Knowledge editing aims to update the embedded knowledge within Large Language Models (LLMs). However, existing approaches, whether through parameter modification or external memory integration, often suffer from inconsistent evaluation objectives and experimental setups. To address this gap, we conduct a comprehensive benchmarking study. In addition to fact-level datasets, we introduce more complex event-based datasets and general-purpose datasets drawn from other tasks. Our evaluation covers both instruction-tuned and reasoning-oriented LLMs, under a realistic autoregressive inference setting rather than teacher-forced decoding. Beyond single-edit assessments, we also evaluate multi-edit scenarios to better reflect practical demands. We employ four evaluation dimensions, including portability, and compare all recent methods against a simple and straightforward baseline named Selective Contextual Reasoning (SCR). Empirical results reveal that parameter-based editing methods perform poorly under realistic conditions. In contrast, SCR consistently outperforms them across all settings. This study offers new insights into the limitations of current knowledge editing methods and highlights the potential of context-based reasoning as a more robust alternative.

Chongzhi Zhang, Xizhou Zhu, Aixin Sun

Video moment search, the process of finding relevant moments in a video corpus to match a user's query, is crucial for various applications. Existing solutions, however, often assume a single perfect matching moment, struggle with inefficient inference, and have limitations with hour-long videos. This paper introduces a flexible and scalable framework for retrieving a ranked list of moments from collection of videos in any length to match a text query, a task termed Ranked Video Moment Retrieval (RVMR). Our framework, called Segment-Proposal-Ranking (SPR), simplifies the search process into three independent stages: segment retrieval, proposal generation, and moment refinement with re-ranking. Specifically, videos are divided into equal-length segments with precomputed embeddings indexed offline, allowing efficient retrieval regardless of video length. For scalable online retrieval, both segments and queries are projected into a shared feature space to enable approximate nearest neighbor (ANN) search. Retrieved segments are then merged into coarse-grained moment proposals. Then a refinement and re-ranking module is designed to reorder and adjust timestamps of the coarse-grained proposals. Evaluations on the TVR-Ranking dataset demonstrate that our framework achieves state-of-the-art performance with significant reductions in computational cost and processing time. The flexible design also allows for independent improvements to each stage, making SPR highly adaptable for large-scale applications.

Chongzhi Zhang, Mingyuan Zhang, Zhiyang Teng et al.

Natural Language Video Localization (NLVL), grounding phrases from natural language descriptions to corresponding video segments, is a complex yet critical task in video understanding. Despite ongoing advancements, many existing solutions lack the capability to globally capture temporal dynamics of the video data. In this study, we present a novel approach to NLVL that aims to address this issue. Our method involves the direct generation of a global 2D temporal map via a conditional denoising diffusion process, based on the input video and language query. The main challenges are the inherent sparsity and discontinuity of a 2D temporal map in devising the diffusion decoder. To address these challenges, we introduce a multi-scale technique and develop an innovative diffusion decoder. Our approach effectively encapsulates the interaction between the query and video data across various time scales. Experiments on the Charades and DiDeMo datasets underscore the potency of our design.

Jing Wang, Aixin Sun, Hao Zhang et al.

Given a query, the task of Natural Language Video Localization (NLVL) is to localize a temporal moment in an untrimmed video that semantically matches the query. In this paper, we adopt a proposal-based solution that generates proposals (i.e., candidate moments) and then select the best matching proposal. On top of modeling the cross-modal interaction between candidate moments and the query, our proposed Moment Sampling DETR (MS-DETR) enables efficient moment-moment relation modeling. The core idea is to sample a subset of moments guided by the learnable templates with an adopted DETR (DEtection TRansformer) framework. To achieve this, we design a multi-scale visual-linguistic encoder, and an anchor-guided moment decoder paired with a set of learnable templates. Experimental results on three public datasets demonstrate the superior performance of MS-DETR.

Xinze Li, Yixin Cao, Muhao Chen et al.

Goal-oriented Script Generation is a new task of generating a list of steps that can fulfill the given goal. In this paper, we propose to extend the task from the perspective of cognitive theory. Instead of a simple flat structure, the steps are typically organized hierarchically - Human often decompose a complex task into subgoals, where each subgoal can be further decomposed into steps. To establish the benchmark, we contribute a new dataset, propose several baseline methods, and set up evaluation metrics. Both automatic and human evaluation verify the high-quality of dataset, as well as the effectiveness of incorporating subgoals into hierarchical script generation. Furthermore, We also design and evaluate the model to discover subgoal, and find that it is a bit more difficult to decompose the goals than summarizing from segmented steps.

Kuicai Dong, Aixin Sun, Jung-Jae Kim et al.

Open Information Extraction (OIE) aims to extract factual relational tuples from open-domain sentences. Downstream tasks use the extracted OIE tuples as facts, without examining the certainty of these facts. However, uncertainty/speculation is a common linguistic phenomenon. Existing studies on speculation detection are defined at sentence level, but even if a sentence is determined to be speculative, not all tuples extracted from it may be speculative. In this paper, we propose to study speculations in OIE and aim to determine whether an extracted tuple is speculative. We formally define the research problem of tuple-level speculation detection and conduct a detailed data analysis on the LSOIE dataset which contains labels for speculative tuples. Lastly, we propose a baseline model OIE-Spec for this new research task.

Kuicai Dong, Aixin Sun, Jung-Jae Kim et al.

Open Information Extraction (OIE) aims to extract relational tuples from open-domain sentences. Existing OIE systems split a sentence into tokens and recognize token spans as tuple relations and arguments. We instead propose Sentence as Chunk sequence (SaC) and recognize chunk spans as tuple relations and arguments. We argue that SaC has better quantitative and qualitative properties for OIE than sentence as token sequence, and evaluate four choices of chunks (i.e., CoNLL chunks, simple phrases, NP chunks, and spans from SpanOIE) against gold OIE tuples. Accordingly, we propose a simple BERT-based model for sentence chunking, and propose Chunk-OIE for tuple extraction on top of SaC. Chunk-OIE achieves state-of-the-art results on multiple OIE datasets, showing that SaC benefits OIE task.

Yubo Ma, Zehao Wang, Yixin Cao et al.

Few-shot event detection (ED) has been widely studied, while this brings noticeable discrepancies, e.g., various motivations, tasks, and experimental settings, that hinder the understanding of models for future progress.This paper presents a thorough empirical study, a unified view of ED models, and a better unified baseline. For fair evaluation, we compare 12 representative methods on three datasets, which are roughly grouped into prompt-based and prototype-based models for detailed analysis. Experiments consistently demonstrate that prompt-based methods, including ChatGPT, still significantly trail prototype-based methods in terms of overall performance. To investigate their superior performance, we break down their design elements along several dimensions and build a unified framework on prototype-based methods. Under such unified view, each prototype-method can be viewed a combination of different modules from these design elements. We further combine all advantageous modules and propose a simple yet effective baseline, which outperforms existing methods by a large margin (e.g., 2.7% F1 gains under low-resource setting).

Xiang Li, Xin Jiang, Xuying Meng et al.

Pre-trained language models (PLMs) have achieved remarkable success in NLP tasks. Despite the great success, mainstream solutions largely follow the pre-training then finetuning paradigm, which brings in both high deployment costs and low training efficiency. Nevertheless, fine-tuning on a specific task is essential because PLMs are only pre-trained with language signal from large raw data. In this paper, we propose a novel fine-tuning-free strategy for language models, to consider both language signal and teacher signal. Teacher signal is an abstraction of a battery of downstream tasks, provided in a unified proposition format. Trained with both language and strong task-aware teacher signals in an interactive manner, our FreeLM model demonstrates strong generalization and robustness. FreeLM outperforms large models e.g., GPT-3 and InstructGPT, on a range of language understanding tasks in experiments. FreeLM is much smaller with 0.3B parameters, compared to 175B in these models.

Ting Lin, Aixin Sun, Yequan Wang

A sentence may express sentiments on multiple aspects. When these aspects are associated with different sentiment polarities, a model's accuracy is often adversely affected. We observe that multiple aspects in such hard sentences are mostly expressed through multiple clauses, or formally known as elementary discourse units (EDUs), and one EDU tends to express a single aspect with unitary sentiment towards that aspect. In this paper, we propose to consider EDU boundaries in sentence modeling, with attentions at both word and EDU levels. Specifically, we highlight sentiment-bearing words in EDU through word-level sparse attention. Then at EDU level, we force the model to attend to the right EDU for the right aspect, by using EDU-level sparse attention and orthogonal regularization. Experiments on three benchmark datasets show that our simple EDU-Attention model outperforms state-of-the-art baselines. Because EDU can be automatically segmented with high accuracy, our model can be applied to sentences directly without the need of manual EDU boundary annotation.

Hao Zhang, Aixin Sun, Wei Jing et al.

Temporal sentence grounding in videos (TSGV), \aka natural language video localization (NLVL) or video moment retrieval (VMR), aims to retrieve a temporal moment that semantically corresponds to a language query from an untrimmed video. Connecting computer vision and natural language, TSGV has drawn significant attention from researchers in both communities. This survey attempts to provide a summary of fundamental concepts in TSGV and current research status, as well as future research directions. As the background, we present a common structure of functional components in TSGV, in a tutorial style: from feature extraction from raw video and language query, to answer prediction of the target moment. Then we review the techniques for multimodal understanding and interaction, which is the key focus of TSGV for effective alignment between the two modalities. We construct a taxonomy of TSGV techniques and elaborate the methods in different categories with their strengths and weaknesses. Lastly, we discuss issues with the current TSGV research and share our insights about promising research directions.

Grace E. Lee, Aixin Sun

Conducting a systematic review (SR) is comprised of multiple tasks: (i) collect documents (studies) that are likely to be relevant from digital libraries (eg., PubMed), (ii) manually read and label the documents as relevant or irrelevant, (iii) extract information from the relevant studies, and (iv) analyze and synthesize the information and derive a conclusion of SR. When researchers label studies, they can screen ranked documents where relevant documents are higher than irrelevant ones. This practice, known as screening prioritization (ie., document ranking approach), speeds up the process of conducting a SR as the documents labelled as relevant can move to the next tasks earlier. However, the approach is limited in reducing the manual workload because the total number of documents to screen remains the same. Towards reducing the manual workload in the screening process, we investigate the quality of document ranking of SR. This can signal researchers whereabouts in the ranking relevant studies are located and let them decide where to stop the screening. After extensive analysis on SR document rankings from different ranking models, we hypothesize 'topic broadness' as a factor that affects the ranking quality of SR. Finally, we propose a measure that estimates the topic broadness and demonstrate that the proposed measure is a simple yet effective method to predict the qualities of document rankings for SRs.

Grace E. Lee, Aixin Sun

When medical researchers conduct a systematic review (SR), screening studies is the most time-consuming process: researchers read several thousands of medical literature and manually label them relevant or irrelevant. Screening prioritization (ie., document ranking) is an approach for assisting researchers by providing document rankings where relevant documents are ranked higher than irrelevant ones. Seed-driven document ranking (SDR) uses a known relevant document (ie., seed) as a query and generates such rankings. Previous work on SDR seeks ways to identify different term weights in a query document and utilizes them in a retrieval model to compute ranking scores. Alternatively, we formulate the SDR task as finding similar documents to a query document and produce rankings based on similarity scores. We propose a document matching measure named Mirror Matching, which calculates matching scores between medical abstract texts by incorporating common writing patterns, such as background, method, result, and conclusion in order. We conduct experiments on CLEF 2019 eHealth Task 2 TAR dataset, and the empirical results show this simple approach achieves the higher performance than traditional and neural retrieval models on Average Precision and Precision-focused metrics.

Hao Zhang, Aixin Sun, Wei Jing et al.

The temporal sentence grounding in video (TSGV) task is to locate a temporal moment from an untrimmed video, to match a language query, i.e., a sentence. Without considering bias in moment annotations (e.g., start and end positions in a video), many models tend to capture statistical regularities of the moment annotations, and do not well learn cross-modal reasoning between video and language query. In this paper, we propose two debiasing strategies, data debiasing and model debiasing, to "force" a TSGV model to capture cross-modal interactions. Data debiasing performs data oversampling through video truncation to balance moment temporal distribution in train set. Model debiasing leverages video-only and query-only models to capture the distribution bias, and forces the model to learn cross-modal interactions. Using VSLNet as the base model, we evaluate impact of the two strategies on two datasets that contain out-of-distribution test instances. Results show that both strategies are effective in improving model generalization capability. Equipped with both debiasing strategies, VSLNet achieves best results on both datasets.

Yiyi Liu, Yequan Wang, Aixin Sun et al.

Sarcasm employs ambivalence, where one says something positive but actually means negative, and vice versa. The essence of sarcasm, which is also a sufficient and necessary condition, is the conflict between literal and implied sentiments expressed in one sentence. However, it is difficult to recognize such sentiment conflict because the sentiments are mixed or even implicit. As a result, the recognition of sophisticated and obscure sentiment brings in a great challenge to sarcasm detection. In this paper, we propose a Dual-Channel Framework by modeling both literal and implied sentiments separately. Based on this dual-channel framework, we design the Dual-Channel Network~(DC-Net) to recognize sentiment conflict. Experiments on political debates (i.e. IAC-V1 and IAC-V2) and Twitter datasets show that our proposed DC-Net achieves state-of-the-art performance on sarcasm recognition. Our code is released to support research.

Xin Zhou, Aixin Sun, Yong Liu et al.

Collaborative filtering (CF) is widely used to learn informative latent representations of users and items from observed interactions. Existing CF-based methods commonly adopt negative sampling to discriminate different items. Training with negative sampling on large datasets is computationally expensive. Further, negative items should be carefully sampled under the defined distribution, in order to avoid selecting an observed positive item in the training dataset. Unavoidably, some negative items sampled from the training dataset could be positive in the test set. In this paper, we propose a self-supervised collaborative filtering framework (SelfCF), that is specially designed for recommender scenario with implicit feedback. The proposed SelfCF framework simplifies the Siamese networks and can be easily applied to existing deep-learning based CF models, which we refer to as backbone networks. The main idea of SelfCF is to augment the output embeddings generated by backbone networks, because it is infeasible to augment raw input of user/item ids. We propose and study three output perturbation techniques that can be applied to different types of backbone networks including both traditional CF models and graph-based models. The framework enables learning informative representations of users and items without negative samples, and is agnostic to the encapsulated backbones. We conduct comprehensive experiments on four datasets to show that our framework may achieve even better recommendation accuracy than the encapsulated supervised counterpart with a 2$\times$--4$\times$ faster training speed. We also show that SelfCF can boost up the accuracy by up to 17.79% on average, compared with a self-supervised framework BUIR.