Bang Liu

Yang Ding, Jing Yu, Bang Liu et al.

Knowledge-based visual question answering requires the ability of associating external knowledge for open-ended cross-modal scene understanding. One limitation of existing solutions is that they capture relevant knowledge from text-only knowledge bases, which merely contain facts expressed by first-order predicates or language descriptions while lacking complex but indispensable multimodal knowledge for visual understanding. How to construct vision-relevant and explainable multimodal knowledge for the VQA scenario has been less studied. In this paper, we propose MuKEA to represent multimodal knowledge by an explicit triplet to correlate visual objects and fact answers with implicit relations. To bridge the heterogeneous gap, we propose three objective losses to learn the triplet representations from complementary views: embedding structure, topological relation and semantic space. By adopting a pre-training and fine-tuning learning strategy, both basic and domain-specific multimodal knowledge are progressively accumulated for answer prediction. We outperform the state-of-the-art by 3.35% and 6.08% respectively on two challenging knowledge-required datasets: OK-VQA and KRVQA. Experimental results prove the complementary benefits of the multimodal knowledge with existing knowledge bases and the advantages of our end-to-end framework over the existing pipeline methods. The code is available at https://github.com/AndersonStra/MuKEA.

Yi Cheng, Wenge Liu, Wenjie Li et al.

Providing Emotional Support (ES) to soothe people in emotional distress is an essential capability in social interactions. Most existing researches on building ES conversation systems only considered single-turn interactions with users, which was over-simplified. In comparison, multi-turn ES conversation systems can provide ES more effectively, but face several new technical challenges, including: (1) how to adopt appropriate support strategies to achieve the long-term dialogue goal of comforting the user's emotion; (2) how to dynamically model the user's state. In this paper, we propose a novel system MultiESC to address these issues. For strategy planning, drawing inspiration from the A* search algorithm, we propose lookahead heuristics to estimate the future user feedback after using particular strategies, which helps to select strategies that can lead to the best long-term effects. For user state modeling, MultiESC focuses on capturing users' subtle emotional expressions and understanding their emotion causes. Extensive experiments show that MultiESC significantly outperforms competitive baselines in both dialogue generation and strategy planning. Our codes are available at https://github.com/lwgkzl/MultiESC.

Sijie Cheng, Zhouhong Gu, Bang Liu et al. · tsinghua

Taxonomies have been widely used in various domains to underpin numerous applications. Specially, product taxonomies serve an essential role in the e-commerce domain for the recommendation, browsing, and query understanding. However, taxonomies need to constantly capture the newly emerged terms or concepts in e-commerce platforms to keep up-to-date, which is expensive and labor-intensive if it relies on manual maintenance and updates. Therefore, we target the taxonomy expansion task to attach new concepts to existing taxonomies automatically. In this paper, we present a self-supervised and user behavior-oriented product taxonomy expansion framework to append new concepts into existing taxonomies. Our framework extracts hyponymy relations that conform to users' intentions and cognition. Specifically, i) to fully exploit user behavioral information, we extract candidate hyponymy relations that match user interests from query-click concepts; ii) to enhance the semantic information of new concepts and better detect hyponymy relations, we model concepts and relations through both user-generated content and structural information in existing taxonomies and user click logs, by leveraging Pre-trained Language Models and Graph Neural Network combined with Contrastive Learning; iii) to reduce the cost of dataset construction and overcome data skews, we construct a high-quality and balanced training dataset from existing taxonomy with no supervision. Extensive experiments on real-world product taxonomies in Meituan Platform, a leading Chinese vertical e-commerce platform to order take-out with more than 70 million daily active users, demonstrate the superiority of our proposed framework over state-of-the-art methods. Notably, our method enlarges the size of real-world product taxonomies from 39,263 to 94,698 relations with 88% precision.

Xiaoqiang Wang, Bang Liu, Siliang Tang et al. · mila

Existing metrics for assessing question generation not only require costly human reference but also fail to take into account the input context of generation, rendering the lack of deep understanding of the relevance between the generated questions and input contexts. As a result, they may wrongly penalize a legitimate and reasonable candidate question when it (i) involves complicated reasoning with the context or (ii) can be grounded by multiple evidences in the context. In this paper, we propose $\textbf{QRelScore}$, a context-aware $\underline{\textbf{Rel}}$evance evaluation metric for $\underline{\textbf{Q}}$uestion Generation. Based on off-the-shelf language models such as BERT and GPT2, QRelScore employs both word-level hierarchical matching and sentence-level prompt-based generation to cope with the complicated reasoning and diverse generation from multiple evidences, respectively. Compared with existing metrics, our experiments demonstrate that QRelScore is able to achieve a higher correlation with human judgments while being much more robust to adversarial samples.

Xiaoqiang Wang, Bang Liu, Fangli Xu et al. · mila

Machine Reading Comprehension (MRC) reveals the ability to understand a given text passage and answer questions based on it. Existing research works in MRC rely heavily on large-size models and corpus to improve the performance evaluated by metrics such as Exact Match ($EM$) and $F_1$. However, such a paradigm lacks sufficient interpretation to model capability and can not efficiently train a model with a large corpus. In this paper, we argue that a deep understanding of model capabilities and data properties can help us feed a model with appropriate training data based on its learning status. Specifically, we design an MRC capability assessment framework that assesses model capabilities in an explainable and multi-dimensional manner. Based on it, we further uncover and disentangle the connections between various data properties and model performance. Finally, to verify the effectiveness of the proposed MRC capability assessment framework, we incorporate it into a curriculum learning pipeline and devise a Capability Boundary Breakthrough Curriculum (CBBC) strategy, which performs a model capability-based training to maximize the data value and improve training efficiency. Extensive experiments demonstrate that our approach significantly improves performance, achieving up to an 11.22% / 8.71% improvement of $EM$ / $F_1$ on MRC tasks.

Yu Song, Santiago Miret, Huan Zhang et al.

We propose an instruction-based process for trustworthy data curation in materials science (MatSci-Instruct), which we then apply to finetune a LLaMa-based language model targeted for materials science (HoneyBee). MatSci-Instruct helps alleviate the scarcity of relevant, high-quality materials science textual data available in the open literature, and HoneyBee is the first billion-parameter language model specialized to materials science. In MatSci-Instruct we improve the trustworthiness of generated data by prompting multiple commercially available large language models for generation with an Instructor module (e.g. Chat-GPT) and verification from an independent Verifier module (e.g. Claude). Using MatSci-Instruct, we construct a dataset of multiple tasks and measure the quality of our dataset along multiple dimensions, including accuracy against known facts, relevance to materials science, as well as completeness and reasonableness of the data. Moreover, we iteratively generate more targeted instructions and instruction-data in a finetuning-evaluation-feedback loop leading to progressively better performance for our finetuned HoneyBee models. Our evaluation on the MatSci-NLP benchmark shows HoneyBee's outperformance of existing language models on materials science tasks and iterative improvement in successive stages of instruction-data refinement. We study the quality of HoneyBee's language modeling through automatic evaluation and analyze case studies to further understand the model's capabilities and limitations. Our code and relevant datasets are publicly available at \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-HoneyBee}.

Wangjie Jiang, Zhihao Ye, Zijing Ou et al.

Chinese Spelling Correction (CSC) is gaining increasing attention due to its promise of automatically detecting and correcting spelling errors in Chinese texts. Despite its extensive use in many applications, like search engines and optical character recognition systems, little has been explored in medical scenarios in which complex and uncommon medical entities are easily misspelled. Correcting the misspellings of medical entities is arguably more difficult than those in the open domain due to its requirements of specificdomain knowledge. In this work, we define the task of Medical-domain Chinese Spelling Correction and propose MCSCSet, a large scale specialist-annotated dataset that contains about 200k samples. In contrast to the existing open-domain CSC datasets, MCSCSet involves: i) extensive real-world medical queries collected from Tencent Yidian, ii) corresponding misspelled sentences manually annotated by medical specialists. To ensure automated dataset curation, MCSCSet further offers a medical confusion set consisting of the commonly misspelled characters of given Chinese medical terms. This enables one to create the medical misspelling dataset automatically. Extensive empirical studies have shown significant performance gaps between the open-domain and medical-domain spelling correction, highlighting the need to develop high-quality datasets that allow for Chinese spelling correction in specific domains. Moreover, our work benchmarks several representative Chinese spelling correction models, establishing baselines for future work.

Shengyao Lu, Bang Liu, Keith G. Mills et al.

Systematicity, i.e., the ability to recombine known parts and rules to form new sequences while reasoning over relational data, is critical to machine intelligence. A model with strong systematicity is able to train on small-scale tasks and generalize to large-scale tasks. In this paper, we propose R5, a relational reasoning framework based on reinforcement learning that reasons over relational graph data and explicitly mines underlying compositional logical rules from observations. R5 has strong systematicity and being robust to noisy data. It consists of a policy value network equipped with Monte Carlo Tree Search to perform recurrent relational prediction and a backtrack rewriting mechanism for rule mining. By alternately applying the two components, R5 progressively learns a set of explicit rules from data and performs explainable and generalizable relation prediction. We conduct extensive evaluations on multiple datasets. Experimental results show that R5 outperforms various embedding-based and rule induction baselines on relation prediction tasks while achieving a high recall rate in discovering ground truth rules.

Ailisi Li, Xueyao Jiang, Bang Liu et al.

Math Word Problems (MWP) is an important task that requires the ability of understanding and reasoning over mathematical text. Existing approaches mostly formalize it as a generation task by adopting Seq2Seq or Seq2Tree models to encode an input math problem in natural language as a global representation and generate the output mathematical expression. Such approaches only learn shallow heuristics and fail to capture fine-grained variations in inputs. In this paper, we propose to model a math word problem in a fine-to-coarse manner to capture both the local fine-grained information and the global logical structure of it. Instead of generating a complete equation sequence or expression tree from the global features, we iteratively combine low-level operands to predict a higher-level operator, abstracting the problem and reasoning about the solving operators from bottom to up. Our model is naturally more sensitive to local variations and can better generalize to unseen problem types. Extensive evaluations on Math23k and SVAMP datasets demonstrate the accuracy and robustness of our method.

Xinlin Li, Bang Liu, Rui Heng Yang et al.

Efficiently deploying deep neural networks on low-resource edge devices is challenging due to their ever-increasing resource requirements. To address this issue, researchers have proposed multiplication-free neural networks, such as Power-of-Two quantization, or also known as Shift networks, which aim to reduce memory usage and simplify computation. However, existing low-bit Shift networks are not as accurate as their full-precision counterparts, typically suffering from limited weight range encoding schemes and quantization loss. In this paper, we propose the DenseShift network, which significantly improves the accuracy of Shift networks, achieving competitive performance to full-precision networks for vision and speech applications. In addition, we introduce a method to deploy an efficient DenseShift network using non-quantized floating-point activations, while obtaining 1.6X speed-up over existing methods. To achieve this, we demonstrate that zero-weight values in low-bit Shift networks do not contribute to model capacity and negatively impact inference computation. To address this issue, we propose a zero-free shifting mechanism that simplifies inference and increases model capacity. We further propose a sign-scale decomposition design to enhance training efficiency and a low-variance random initialization strategy to improve the model's transfer learning performance. Our extensive experiments on various computer vision and speech tasks demonstrate that DenseShift outperforms existing low-bit multiplication-free networks and achieves competitive performance compared to full-precision networks. Furthermore, our proposed approach exhibits strong transfer learning performance without a drop in accuracy. Our code was released on GitHub.

Jeremy Qin, Bang Liu, Quoc Dinh Nguyen

Black-box large language models (LLMs) are increasingly deployed in various environments, making it essential for these models to effectively convey their confidence and uncertainty, especially in high-stakes settings. However, these models often exhibit overconfidence, leading to potential risks and misjudgments. Existing techniques for eliciting and calibrating LLM confidence have primarily focused on general reasoning datasets, yielding only modest improvements. Accurate calibration is crucial for informed decision-making and preventing adverse outcomes but remains challenging due to the complexity and variability of tasks these models perform. In this work, we investigate the miscalibration behavior of black-box LLMs within the healthcare setting. We propose a novel method, \textit{Atypical Presentations Recalibration}, which leverages atypical presentations to adjust the model's confidence estimates. Our approach significantly improves calibration, reducing calibration errors by approximately 60\% on three medical question answering datasets and outperforming existing methods such as vanilla verbalized confidence, CoT verbalized confidence and others. Additionally, we provide an in-depth analysis of the role of atypicality within the recalibration framework.

Huan Zhang, Yu Song, Ziyu Hou et al.

The emergence of specialized large language models (LLMs) has shown promise in addressing complex tasks for materials science. Many LLMs, however, often struggle with distinct complexities of material science tasks, such as materials science computational tasks, and often rely heavily on outdated implicit knowledge, leading to inaccuracies and hallucinations. To address these challenges, we introduce HoneyComb, the first LLM-based agent system specifically designed for materials science. HoneyComb leverages a novel, high-quality materials science knowledge base (MatSciKB) and a sophisticated tool hub (ToolHub) to enhance its reasoning and computational capabilities tailored to materials science. MatSciKB is a curated, structured knowledge collection based on reliable literature, while ToolHub employs an Inductive Tool Construction method to generate, decompose, and refine API tools for materials science. Additionally, HoneyComb leverages a retriever module that adaptively selects the appropriate knowledge source or tools for specific tasks, thereby ensuring accuracy and relevance. Our results demonstrate that HoneyComb significantly outperforms baseline models across various tasks in materials science, effectively bridging the gap between current LLM capabilities and the specialized needs of this domain. Furthermore, our adaptable framework can be easily extended to other scientific domains, highlighting its potential for broad applicability in advancing scientific research and applications.

Hossein Rajabzadeh, Suyuchen Wang, Hyock Ju Kwon et al.

We employ a tool-interacting divide-and-conquer strategy enabling large language models (LLMs) to answer complex multimodal multi-hop questions. In particular, we harness the power of large language models to divide a given multimodal multi-hop question into unimodal single-hop sub-questions to be answered by the appropriate tool from a predefined set of tools. After all corresponding tools provide the LLM with their answers, the LLM generates the next relevant unimodal single-hop question. To increase the reasoning ability of LLMs, we prompt chatGPT to generate a tool-interacting divide-and-conquer dataset. This dataset is then used to efficiently finetune the corresponding LLM. To assess the effectiveness of this approach, we conduct an evaluation on two recently introduced complex question-answering datasets. The experimental analysis demonstrate substantial improvements over existing state-of-the-art solutions, indicating the efficacy and generality of our strategy

Yu Song, Santiago Miret, Bang Liu

We present MatSci-NLP, a natural language benchmark for evaluating the performance of natural language processing (NLP) models on materials science text. We construct the benchmark from publicly available materials science text data to encompass seven different NLP tasks, including conventional NLP tasks like named entity recognition and relation classification, as well as NLP tasks specific to materials science, such as synthesis action retrieval which relates to creating synthesis procedures for materials. We study various BERT-based models pretrained on different scientific text corpora on MatSci-NLP to understand the impact of pretraining strategies on understanding materials science text. Given the scarcity of high-quality annotated data in the materials science domain, we perform our fine-tuning experiments with limited training data to encourage the generalize across MatSci-NLP tasks. Our experiments in this low-resource training setting show that language models pretrained on scientific text outperform BERT trained on general text. MatBERT, a model pretrained specifically on materials science journals, generally performs best for most tasks. Moreover, we propose a unified text-to-schema for multitask learning on \benchmark and compare its performance with traditional fine-tuning methods. In our analysis of different training methods, we find that our proposed text-to-schema methods inspired by question-answering consistently outperform single and multitask NLP fine-tuning methods. The code and datasets are publicly available at \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-ACL23}.

Jiuding Yang, Weidong Guo, Bang Liu et al.

Although conceptualization has been widely studied in semantics and knowledge representation, it is still challenging to find the most accurate concept phrases to characterize the main idea of a text snippet on the fast-growing social media. This is partly attributed to the fact that most knowledge bases contain general terms of the world, such as trees and cars, which do not have the defining power or are not interesting enough to social media app users. Another reason is that the intricacy of natural language allows the use of tense, negation and grammar to change the logic or emphasis of language, thus conveying completely different meanings. In this paper, we present TAG, a high-quality concept matching dataset consisting of 10,000 labeled pairs of fine-grained concepts and web-styled natural language sentences, mined from the open-domain social media. The concepts we consider represent the trending interests of online users. Associated with TAG is a concept graph of these fine-grained concepts and entities to provide the structural context information. We evaluate a wide range of popular neural text matching models as well as pre-trained language models on TAG, and point out their insufficiency to tag social media content with the most appropriate concept. We further propose a novel graph-graph matching method that demonstrates superior abstraction and generalization performance by better utilizing both the structural context in the concept graph and logic interactions between semantic units in the sentence via syntactic dependency parsing. We open-source both the TAG dataset and the proposed methods to facilitate further research.

Qianren Mao, Xi Li, Bang Liu et al.

Hashtag generation aims to generate short and informal topical tags from a microblog post, in which tokens or phrases form the hashtags. These tokens or phrases may originate from primary fragmental textual pieces (e.g., segments) in the original text and are separated into different segments. However, conventional sequence-to-sequence generation methods are hard to filter out secondary information from different textual granularity and are not good at selecting crucial tokens. Thus, they are suboptimal in generating more condensed hashtags. In this work, we propose a modified Transformer-based generation model with adding a segments-selection procedure for the original encoding and decoding phases. The segments-selection phase is based on a novel Segments Selection Mechanism (SSM) to model different textual granularity on global text, local segments, and tokens, contributing to generating condensed hashtags. Specifically, it first attends to primary semantic segments and then transforms discontinuous segments from the source text into a sequence of hashtags by selecting crucial tokens. Extensive evaluations on the two datasets reveal our approach's superiority with significant improvements to the extraction and generation baselines. The code and datasets are available at https://github.com/OpenSUM/HashtagGen.

Bang Liu, Xinfeng Li, Jiayi Zhang et al. · microsoft-research

The advent of large language models (LLMs) has catalyzed a transformative shift in artificial intelligence, paving the way for advanced intelligent agents capable of sophisticated reasoning, robust perception, and versatile action across diverse domains. As these agents increasingly drive AI research and practical applications, their design, evaluation, and continuous improvement present intricate, multifaceted challenges. This book provides a comprehensive overview, framing intelligent agents within modular, brain-inspired architectures that integrate principles from cognitive science, neuroscience, and computational research. We structure our exploration into four interconnected parts. First, we systematically investigate the modular foundation of intelligent agents, systematically mapping their cognitive, perceptual, and operational modules onto analogous human brain functionalities and elucidating core components such as memory, world modeling, reward processing, goal, and emotion. Second, we discuss self-enhancement and adaptive evolution mechanisms, exploring how agents autonomously refine their capabilities, adapt to dynamic environments, and achieve continual learning through automated optimization paradigms. Third, we examine multi-agent systems, investigating the collective intelligence emerging from agent interactions, cooperation, and societal structures. Finally, we address the critical imperative of building safe and beneficial AI systems, emphasizing intrinsic and extrinsic security threats, ethical alignment, robustness, and practical mitigation strategies necessary for trustworthy real-world deployment. By synthesizing modular AI architectures with insights from different disciplines, this survey identifies key research challenges and opportunities, encouraging innovations that harmonize technological advancement with meaningful societal benefit.

Suyuchen Wang, Ivan Kobyzev, Peng Lu et al.

This paper addresses the challenge of train-short-test-long (TSTL) scenarios in Large Language Models (LLMs) equipped with Rotary Position Embedding (RoPE), where models pre-trained on shorter sequences face difficulty with out-of-distribution (OOD) token positions in longer sequences. We introduce Resonance RoPE, a novel approach designed to narrow the generalization gap in TSTL scenarios by refining the interpolation of RoPE features for OOD positions, significantly improving the model performance without additional online computational costs. Furthermore, we present PosGen, a new synthetic benchmark specifically designed for fine-grained behavior analysis in TSTL scenarios, aiming to isolate the constantly increasing difficulty of token generation on long contexts from the challenges of recognizing new token positions. Our experiments on synthetic tasks show that after applying Resonance RoPE, Transformers recognize OOD position better and more robustly. Our extensive LLM experiments also show superior performance after applying Resonance RoPE to the current state-of-the-art RoPE scaling method, YaRN, on both upstream language modeling tasks and a variety of downstream long-text applications.

Haochen Shi, Zhiyuan Sun, Xingdi Yuan et al. · microsoft-research

Embodied Instruction Following (EIF) is a crucial task in embodied learning, requiring agents to interact with their environment through egocentric observations to fulfill natural language instructions. Recent advancements have seen a surge in employing large language models (LLMs) within a framework-centric approach to enhance performance in embodied learning tasks, including EIF. Despite these efforts, there exists a lack of a unified understanding regarding the impact of various components-ranging from visual perception to action execution-on task performance. To address this gap, we introduce OPEx, a comprehensive framework that delineates the core components essential for solving embodied learning tasks: Observer, Planner, and Executor. Through extensive evaluations, we provide a deep analysis of how each component influences EIF task performance. Furthermore, we innovate within this space by deploying a multi-agent dialogue strategy on a TextWorld counterpart, further enhancing task performance. Our findings reveal that LLM-centric design markedly improves EIF outcomes, identify visual perception and low-level action execution as critical bottlenecks, and demonstrate that augmenting LLMs with a multi-agent framework further elevates performance.

K Roth, Rushil Gupta, Simon Halle et al.

Large language models struggle to synthesize disparate pieces of information into a coherent plan when approaching a complex procedural task. In this work, we introduce a novel formalism and structure for such procedural knowledge. Based on this formalism, we present a novel procedural knowledge dataset called LCStep, which we created from LangChain tutorials. To leverage this procedural knowledge to solve new tasks, we propose analogy-augmented generation (AAG), which draws inspiration from the human ability to assimilate past experiences to solve unfamiliar problems. AAG uses a custom procedure memory store to retrieve and adapt specialized domain knowledge to answer new procedural tasks. We demonstrate that AAG outperforms few-shot and RAG baselines on LCStep, RecipeNLG, and CHAMP datasets under a pairwise LLM-based evaluation, corroborated by human evaluation in the case of RecipeNLG.

Shengyao Lu, Bang Liu, Keith G. Mills et al.

Understanding and explaining the predictions of Graph Neural Networks (GNNs), is crucial for enhancing their safety and trustworthiness. Subgraph-level explanations are gaining attention for their intuitive appeal. However, most existing subgraph-level explainers face efficiency challenges in explaining GNNs due to complex search processes. The key challenge is to find a balance between intuitiveness and efficiency while ensuring transparency. Additionally, these explainers usually induce subgraphs by nodes, which may introduce less-intuitive disconnected nodes in the subgraph-level explanations or omit many important subgraph structures. In this paper, we reveal that inducing subgraph explanations by edges is more comprehensive than other subgraph inducing techniques. We also emphasize the need of determining the subgraph explanation size for each data instance, as different data instances may involve different important substructures. Building upon these considerations, we introduce a training-free approach, named EiG-Search. We employ an efficient linear-time search algorithm over the edge-induced subgraphs, where the edges are ranked by an enhanced gradient-based importance. We conduct extensive experiments on a total of seven datasets, demonstrating its superior performance and efficiency both quantitatively and qualitatively over the leading baselines.

Jinlin Wang, Suyuchen Wang, Ziwen Xia et al.

Large Language Models (LLMs) are proficient at retrieving single facts from extended contexts, yet they struggle with tasks requiring the simultaneous retrieval of multiple facts, especially during generation. This paper identifies a novel "lost-in-the-middle" phenomenon, where LLMs progressively lose track of critical information throughout the generation process, resulting in incomplete or inaccurate retrieval. To address this challenge, we introduce Find All Crucial Texts (FACT), an iterative retrieval method that refines context through successive rounds of rewriting. This approach enables models to capture essential facts incrementally, which are often overlooked in single-pass retrieval. Experiments demonstrate that FACT substantially enhances multi-fact retrieval performance across various tasks, though improvements are less notable in general-purpose QA scenarios. Our findings shed light on the limitations of LLMs in multi-fact retrieval and underscore the need for more resilient long-context retrieval strategies.

Dekun Wu, Frederik Brudy, Bang Liu et al.

Virtual environments are essential to AI agent research. Existing environments for LLM agent research typically focus on either physical task solving or social simulation, with the former oversimplifying agent individuality and social dynamics, and the latter lacking physical grounding of social behaviors. We introduce IndoorWorld, a heterogeneous multi-agent environment that tightly integrates physical and social dynamics. By introducing novel challenges for LLM-driven agents in orchestrating social dynamics to influence physical environments and anchoring social interactions within world states, IndoorWorld opens up possibilities of LLM-based building occupant simulation for architectural design. We demonstrate the potential with a series of experiments within an office setting to examine the impact of multi-agent collaboration, resource competition, and spatial layout on agent behavior.

Xiaoqiang Wang, Lingfei Wu, Tengfei Ma et al. · mila

Large language models (LLMs) are primarily evaluated by overall performance on various text understanding and generation tasks. However, such a paradigm fails to comprehensively differentiate the fine-grained language and cognitive skills, rendering the lack of sufficient interpretation to LLMs' capabilities. In this paper, we present FAC$^2$E, a framework for Fine-grAined and Cognition-grounded LLMs' Capability Evaluation. Specifically, we formulate LLMs' evaluation in a multi-dimensional and explainable manner by dissociating the language-related capabilities and the cognition-related ones. Besides, through extracting the intermediate reasoning from LLMs, we further break down the process of applying a specific capability into three sub-steps: recalling relevant knowledge, utilizing knowledge, and solving problems. Finally, FAC$^2$E evaluates each sub-step of each fine-grained capability, providing a two-faceted diagnosis for LLMs. Utilizing FAC$^2$E, we identify a common shortfall in knowledge utilization among models and propose a straightforward, knowledge-enhanced method to mitigate this issue. Our results not only showcase promising performance enhancements but also highlight a direction for future LLM advancements.

Shengyao Lu, Keith G. Mills, Jiao He et al.

Understanding the decision-making process of Graph Neural Networks (GNNs) is crucial to their interpretability. Most existing methods for explaining GNNs typically rely on training auxiliary models, resulting in the explanations remain black-boxed. This paper introduces Graph Output Attribution (GOAt), a novel method to attribute graph outputs to input graph features, creating GNN explanations that are faithful, discriminative, as well as stable across similar samples. By expanding the GNN as a sum of scalar products involving node features, edge features and activation patterns, we propose an efficient analytical method to compute contribution of each node or edge feature to each scalar product and aggregate the contributions from all scalar products in the expansion form to derive the importance of each node and edge. Through extensive experiments on synthetic and real-world data, we show that our method not only outperforms various state-ofthe-art GNN explainers in terms of the commonly used fidelity metric, but also exhibits stronger discriminability, and stability by a remarkable margin.

Zhong Zhang, Bang Liu, Junming Shao

Pre-trained language models (PLMs) are known to be overly parameterized and have significant redundancy, indicating a small degree of freedom of the PLMs. Motivated by the observation, in this paper, we study the problem of re-parameterizing and fine-tuning PLMs from a new perspective: Discovery of intrinsic task-specific subspace. Specifically, by exploiting the dynamics of the fine-tuning process for a given task, the parameter optimization trajectory is learned to uncover its intrinsic task-specific subspace. A key finding is that PLMs can be effectively fine-tuned in the subspace with a small number of free parameters. Beyond, we observe some outlier dimensions emerging during fine-tuning in the subspace. Disabling these dimensions degrades the model performance significantly. This suggests that these dimensions are crucial to induce task-specific knowledge to downstream tasks.

Xiaoqiang Wang, Bang Liu, Siliang Tang et al.

We present $\textbf{$\texttt{SkillQG}$}$: a question generation framework with controllable comprehension types for assessing and improving machine reading comprehension models. Existing question generation systems widely differentiate questions by $\textit{literal}$ information such as question words and answer types to generate semantically relevant questions for a given context. However, they rarely consider the $\textit{comprehension}$ nature of questions, i.e. the different comprehension capabilities embodied by different questions. In comparison, our $\texttt{SkillQG}$ is able to tailor a fine-grained assessment and improvement to the capabilities of question answering models built on it. Specifically, we first frame the comprehension type of questions based on a hierarchical skill-based schema, then formulate $\texttt{SkillQG}$ as a skill-conditioned question generator. Furthermore, to improve the controllability of generation, we augment the input text with question focus and skill-specific knowledge, which are constructed by iteratively prompting the pre-trained language models. Empirical results demonstrate that $\texttt{SkillQG}$ outperforms baselines in terms of quality, relevance, and skill-controllability while showing a promising performance boost in downstream question answering task.

Yuyan Chen, Yanghua Xiao, Bang Liu

Interpreting the predictions of existing Question Answering (QA) models is critical to many real-world intelligent applications, such as QA systems for healthcare, education, and finance. However, existing QA models lack interpretability and provide no feedback or explanation for end-users to help them understand why a specific prediction is the answer to a question. In this research, we argue that the evidences of an answer is critical to enhancing the interpretability of QA models. Unlike previous research that simply extracts several sentence(s) in the context as evidence, we are the first to explicitly define the concept of evidence as the supporting facts in a context which are informative, concise, and readable. Besides, we provide effective strategies to quantitatively measure the informativeness, conciseness and readability of evidence. Furthermore, we propose Grow-and-Clip Evidence Distillation (GCED) algorithm to extract evidences from the contexts by trade-off informativeness, conciseness, and readability. We conduct extensive experiments on the SQuAD and TriviaQA datasets with several baseline models to evaluate the effect of GCED on interpreting answers to questions. Human evaluation are also carried out to check the quality of distilled evidences. Experimental results show that automatic distilled evidences have human-like informativeness, conciseness and readability, which can enhance the interpretability of the answers to questions.

Martin Weyssow, Houari Sahraoui, Bang Liu

The progress made in code modeling has been tremendous in recent years thanks to the design of natural language processing learning approaches based on state-of-the-art model architectures. Nevertheless, we believe that the current state-of-the-art does not focus enough on the full potential that data may bring to a learning process in software engineering. Our vision articulates on the idea of leveraging multi-modal learning approaches to modeling the programming world. In this paper, we investigate one of the underlying idea of our vision whose objective based on concept graphs of identifiers aims at leveraging high-level relationships between domain concepts manipulated through particular language constructs. In particular, we propose to enhance an existing pretrained language model of code by joint-learning it with a graph neural network based on our concept graphs. We conducted a preliminary evaluation that shows gain of effectiveness of the models for code search using a simple joint-learning method and prompts us to further investigate our research vision.

Zijing Ou, Qinliang Su, Jianxing Yu et al.

Existing unsupervised document hashing methods are mostly established on generative models. Due to the difficulties of capturing long dependency structures, these methods rarely model the raw documents directly, but instead to model the features extracted from them (e.g. bag-of-words (BOW), TFIDF). In this paper, we propose to learn hash codes from BERT embeddings after observing their tremendous successes on downstream tasks. As a first try, we modify existing generative hashing models to accommodate the BERT embeddings. However, little improvement is observed over the codes learned from the old BOW or TFIDF features. We attribute this to the reconstruction requirement in the generative hashing, which will enforce irrelevant information that is abundant in the BERT embeddings also compressed into the codes. To remedy this issue, a new unsupervised hashing paradigm is further proposed based on the mutual information (MI) maximization principle. Specifically, the method first constructs appropriate global and local codes from the documents and then seeks to maximize their mutual information. Experimental results on three benchmark datasets demonstrate that the proposed method is able to generate hash codes that outperform existing ones learned from BOW features by a substantial margin.

Zhexue Chen, Hong Huang, Bang Liu et al.

Aspect Sentiment Triplet Extraction (ASTE) aims to extract triplets from sentences, where each triplet includes an entity, its associated sentiment, and the opinion span explaining the reason for the sentiment. Most existing research addresses this problem in a multi-stage pipeline manner, which neglects the mutual information between such three elements and has the problem of error propagation. In this paper, we propose a Semantic and Syntactic Enhanced aspect Sentiment triplet Extraction model (S3E2) to fully exploit the syntactic and semantic relationships between the triplet elements and jointly extract them. Specifically, we design a Graph-Sequence duel representation and modeling paradigm for the task of ASTE: we represent the semantic and syntactic relationships between word pairs in a sentence by graph and encode it by Graph Neural Networks (GNNs), as well as modeling the original sentence by LSTM to preserve the sequential information. Under this setting, we further apply a more efficient inference strategy for the extraction of triplets. Extensive evaluations on four benchmark datasets show that S3E2 significantly outperforms existing approaches, which proves our S3E2's superiority and flexibility in an end-to-end fashion.

Tong Mo, Bang Liu

Keyword spotting aims to identify specific keyword audio utterances. In recent years, deep convolutional neural networks have been widely utilized in keyword spotting systems. However, their model architectures are mainly based on off-the shelfbackbones such as VGG-Net or ResNet, instead of specially designed for the task. In this paper, we utilize neural architecture search to design convolutional neural network models that can boost the performance of keyword spotting while maintaining an acceptable memory footprint. Specifically, we search the model operators and their connections in a specific search space with Encoder-Decoder neural architecture optimization. Extensive evaluations on Google's Speech Commands Dataset show that the model architecture searched by our approach achieves a state-of-the-art accuracy of over 97%.

Junnan Liu, Qianren Mao, Bang Liu et al.

Neural abstractive summarization methods often require large quantities of labeled training data. However, labeling large amounts of summarization data is often prohibitive due to time, financial, and expertise constraints, which has limited the usefulness of summarization systems to practical applications. In this paper, we argue that this limitation can be overcome by a semi-supervised approach: consistency training which is to leverage large amounts of unlabeled data to improve the performance of supervised learning over a small corpus. The consistency regularization semi-supervised learning can regularize model predictions to be invariant to small noise applied to input articles. By adding noised unlabeled corpus to help regularize consistency training, this framework obtains comparative performance without using the full dataset. In particular, we have verified that leveraging large amounts of unlabeled data decently improves the performance of supervised learning over an insufficient labeled dataset.

Zijing Ou, Qinliang Su, Jianxing Yu et al.

With the need of fast retrieval speed and small memory footprint, document hashing has been playing a crucial role in large-scale information retrieval. To generate high-quality hashing code, both semantics and neighborhood information are crucial. However, most existing methods leverage only one of them or simply combine them via some intuitive criteria, lacking a theoretical principle to guide the integration process. In this paper, we encode the neighborhood information with a graph-induced Gaussian distribution, and propose to integrate the two types of information with a graph-driven generative model. To deal with the complicated correlations among documents, we further propose a tree-structured approximation method for learning. Under the approximation, we prove that the training objective can be decomposed into terms involving only singleton or pairwise documents, enabling the model to be trained as efficiently as uncorrelated ones. Extensive experimental results on three benchmark datasets show that our method achieves superior performance over state-of-the-art methods, demonstrating the effectiveness of the proposed model for simultaneously preserving semantic and neighborhood information.\

Yi Cheng, Siyao Li, Bang Liu et al.

This paper explores the task of Difficulty-Controllable Question Generation (DCQG), which aims at generating questions with required difficulty levels. Previous research on this task mainly defines the difficulty of a question as whether it can be correctly answered by a Question Answering (QA) system, lacking interpretability and controllability. In our work, we redefine question difficulty as the number of inference steps required to answer it and argue that Question Generation (QG) systems should have stronger control over the logic of generated questions. To this end, we propose a novel framework that progressively increases question difficulty through step-by-step rewriting under the guidance of an extracted reasoning chain. A dataset is automatically constructed to facilitate the research, on which extensive experiments are conducted to test the performance of our method.

Zhengxu Hou, Bang Liu, Ruihui Zhao et al.

For task-oriented dialog systems, training a Reinforcement Learning (RL) based Dialog Management module suffers from low sample efficiency and slow convergence speed due to the sparse rewards in RL.To solve this problem, many strategies have been proposed to give proper rewards when training RL, but their rewards lack interpretability and cannot accurately estimate the distribution of state-action pairs in real dialogs. In this paper, we propose a multi-level reward modeling approach that factorizes a reward into a three-level hierarchy: domain, act, and slot. Based on inverse adversarial reinforcement learning, our designed reward model can provide more accurate and explainable reward signals for state-action pairs.Extensive evaluations show that our approach can be applied to a wide range of reinforcement learning-based dialog systems and significantly improves both the performance and the speed of convergence.

Suyuchen Wang, Ruihui Zhao, Xi Chen et al.

Taxonomy is a hierarchically structured knowledge graph that plays a crucial role in machine intelligence. The taxonomy expansion task aims to find a position for a new term in an existing taxonomy to capture the emerging knowledge in the world and keep the taxonomy dynamically updated. Previous taxonomy expansion solutions neglect valuable information brought by the hierarchical structure and evaluate the correctness of merely an added edge, which downgrade the problem to node-pair scoring or mini-path classification. In this paper, we propose the Hierarchy Expansion Framework (HEF), which fully exploits the hierarchical structure's properties to maximize the coherence of expanded taxonomy. HEF makes use of taxonomy's hierarchical structure in multiple aspects: i) HEF utilizes subtrees containing most relevant nodes as self-supervision data for a complete comparison of parental and sibling relations; ii) HEF adopts a coherence modeling module to evaluate the coherence of a taxonomy's subtree by integrating hypernymy relation detection and several tree-exclusive features; iii) HEF introduces the Fitting Score for position selection, which explicitly evaluates both path and level selections and takes full advantage of parental relations to interchange information for disambiguation and self-correction. Extensive experiments show that by better exploiting the hierarchical structure and optimizing taxonomy's coherence, HEF vastly surpasses the prior state-of-the-art on three benchmark datasets by an average improvement of 46.7% in accuracy and 32.3% in mean reciprocal rank.

Mingjun Zhao, Shengli Yan, Bang Liu et al.

Query-based document summarization aims to extract or generate a summary of a document which directly answers or is relevant to the search query. It is an important technique that can be beneficial to a variety of applications such as search engines, document-level machine reading comprehension, and chatbots. Currently, datasets designed for query-based summarization are short in numbers and existing datasets are also limited in both scale and quality. Moreover, to the best of our knowledge, there is no publicly available dataset for Chinese query-based document summarization. In this paper, we present QBSUM, a high-quality large-scale dataset consisting of 49,000+ data samples for the task of Chinese query-based document summarization. We also propose multiple unsupervised and supervised solutions to the task and demonstrate their high-speed inference and superior performance via both offline experiments and online A/B tests. The QBSUM dataset is released in order to facilitate future advancement of this research field.

Ting Zhang, Bang Liu, Di Niu et al.

A large number of deep learning models have been proposed for the text matching problem, which is at the core of various typical natural language processing (NLP) tasks. However, existing deep models are mainly designed for the semantic matching between a pair of short texts, such as paraphrase identification and question answering, and do not perform well on the task of relevance matching between short-long text pairs. This is partially due to the fact that the essential characteristics of short-long text matching have not been well considered in these deep models. More specifically, these methods fail to handle extreme length discrepancy between text pieces and neither can they fully characterize the underlying structural information in long text documents. In this paper, we are especially interested in relevance matching between a piece of short text and a long document, which is critical to problems like query-document matching in information retrieval and web searching. To extract the structural information of documents, an undirected graph is constructed, with each vertex representing a keyword and the weight of an edge indicating the degree of interaction between keywords. Based on the keyword graph, we further propose a Multiresolution Graph Attention Network to learn multi-layered representations of vertices through a Graph Convolutional Network (GCN), and then match the short text snippet with the graphical representation of the document with the attention mechanisms applied over each layer of the GCN. Experimental results on two datasets demonstrate that our graph approach outperforms other state-of-the-art deep matching models.

Bang Liu, Mingjun Zhao, Di Niu et al.

Automatic question generation is an important technique that can improve the training of question answering, help chatbots to start or continue a conversation with humans, and provide assessment materials for educational purposes. Existing neural question generation models are not sufficient mainly due to their inability to properly model the process of how each word in the question is selected, i.e., whether repeating the given passage or being generated from a vocabulary. In this paper, we propose our Clue Guided Copy Network for Question Generation (CGC-QG), which is a sequence-to-sequence generative model with copying mechanism, yet employing a variety of novel components and techniques to boost the performance of question generation. In CGC-QG, we design a multi-task labeling strategy to identify whether a question word should be copied from the input passage or be generated instead, guiding the model to learn the accurate boundaries between copying and generation. Furthermore, our input passage encoder takes as input, among a diverse range of other features, the prediction made by a clue word predictor, which helps identify whether each word in the input passage is a potential clue to be copied into the target question. The clue word predictor is designed based on a novel application of Graph Convolutional Networks onto a syntactic dependency tree representation of each passage, thus being able to predict clue words only based on their context in the passage and their relative positions to the answer in the tree. We jointly train the clue prediction as well as question generation with multi-task learning and a number of practical strategies to reduce the complexity. Extensive evaluations show that our model significantly improves the performance of question generation and out-performs all previous state-of-the-art neural question generation models by a substantial margin.

Bang Liu, Di Niu, Haojie Wei et al.

Identifying the relationship between two articles, e.g., whether two articles published from different sources describe the same breaking news, is critical to many document understanding tasks. Existing approaches for modeling and matching sentence pairs do not perform well in matching longer documents, which embody more complex interactions between the enclosed entities than a sentence does. To model article pairs, we propose the Concept Interaction Graph to represent an article as a graph of concepts. We then match a pair of articles by comparing the sentences that enclose the same concept vertex through a series of encoding techniques, and aggregate the matching signals through a graph convolutional network. To facilitate the evaluation of long article matching, we have created two datasets, each consisting of about 30K pairs of breaking news articles covering diverse topics in the open domain. Extensive evaluations of the proposed methods on the two datasets demonstrate significant improvements over a wide range of state-of-the-art methods for natural language matching.