Pengjie Ren

Weiwei Sun, Lingyong Yan, Xinyu Ma et al. · baidu

Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy between the pre-training objectives of LLMs and the ranking objective poses another challenge. In this paper, we first investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR. Surprisingly, our experiments reveal that properly instructed LLMs can deliver competitive, even superior results to state-of-the-art supervised methods on popular IR benchmarks. Furthermore, to address concerns about data contamination of LLMs, we collect a new test set called NovelEval, based on the latest knowledge and aiming to verify the model's ability to rank unknown knowledge. Finally, to improve efficiency in real-world applications, we delve into the potential for distilling the ranking capabilities of ChatGPT into small specialized models using a permutation distillation scheme. Our evaluation results turn out that a distilled 440M model outperforms a 3B supervised model on the BEIR benchmark. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.

Shen Gao, Zhengliang Shi, Minghang Zhu et al. · pku

Augmenting large language models (LLMs) with external tools has emerged as a promising approach to extending the capability of LLMs. Although some works employ open-source LLMs for the tool learning task, most of them are trained in a controlled environment in which LLMs only learn to execute the human-provided tools. However, selecting proper tools from the large toolset is also a crucial ability for the tool learning model to be applied in real-world applications. Existing methods usually directly employ self-instruction methods to train the model, which ignores differences in tool complexity. In this paper, we propose the Confucius, a novel tool learning framework to train LLM to use complicated tools in real-world scenarios, which contains two main phases: (1) We first propose a multi-stage learning method to teach the LLM to use various tools from an easy-to-difficult curriculum; (2) thenceforth, we propose the Iterative Self-instruct from Introspective Feedback (ISIF) to dynamically construct the dataset to improve the ability to use the complicated tool. Extensive experiments conducted on both controlled and real-world settings demonstrate the superiority of our tool learning framework in the real-world application scenarios compared to both tuning-free (e.g. ChatGPT, Claude) and tuning-based baselines (e.g. GPT4Tools).

Weiwei Sun, Zheng Chen, Xinyu Ma et al. · baidu

Recent studies have demonstrated the great potential of Large Language Models (LLMs) serving as zero-shot relevance rankers. The typical approach involves making comparisons between pairs or lists of documents. Although effective, these listwise and pairwise methods are not efficient and also heavily rely on intricate prompt engineering. To tackle this problem, we introduce a novel instruction distillation method. The key idea is to distill the pairwise ranking ability of open-sourced LLMs to a simpler but more efficient pointwise ranking. Specifically, given the same LLM, we first rank documents using the effective pairwise approach with complex instructions, and then distill the teacher predictions to the pointwise approach with simpler instructions. Evaluation results on the BEIR, TREC, and ReDial datasets demonstrate that instruction distillation can improve efficiency by 10 to 100x and also enhance the ranking performance of LLMs. Furthermore, our approach surpasses the performance of existing supervised methods like monoT5 and is on par with the state-of-the-art zero-shot methods. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.

Qingyao Ai, Ting Bai, Zhao Cao et al. · pku, tsinghua

The research field of Information Retrieval (IR) has evolved significantly, expanding beyond traditional search to meet diverse user information needs. Recently, Large Language Models (LLMs) have demonstrated exceptional capabilities in text understanding, generation, and knowledge inference, opening up exciting avenues for IR research. LLMs not only facilitate generative retrieval but also offer improved solutions for user understanding, model evaluation, and user-system interactions. More importantly, the synergistic relationship among IR models, LLMs, and humans forms a new technical paradigm that is more powerful for information seeking. IR models provide real-time and relevant information, LLMs contribute internal knowledge, and humans play a central role of demanders and evaluators to the reliability of information services. Nevertheless, significant challenges exist, including computational costs, credibility concerns, domain-specific limitations, and ethical considerations. To thoroughly discuss the transformative impact of LLMs on IR research, the Chinese IR community conducted a strategic workshop in April 2023, yielding valuable insights. This paper provides a summary of the workshop's outcomes, including the rethinking of IR's core values, the mutual enhancement of LLMs and IR, the proposal of a novel IR technical paradigm, and open challenges.

Weiwei Sun, Zhengliang Shi, Shen Gao et al. · pku

Pre-trained language models (LMs) store knowledge in their parameters and can generate informative responses when used in conversational systems. However, LMs suffer from the problem of "hallucination:" they may generate plausible-looking statements that are irrelevant or factually incorrect. To address this problem, we propose a contrastive learning scheme, named MixCL. A novel mixed contrastive objective is proposed to explicitly optimize the implicit knowledge elicitation process of LMs, and thus reduce their hallucination in conversations. We also examine negative sampling strategies of retrieved hard negatives and model-generated negatives. We conduct experiments on Wizard-of-Wikipedia, a public, open-domain knowledge-grounded dialogue benchmark, and assess the effectiveness of MixCL. MixCL effectively reduces the hallucination of LMs in conversations and achieves the highest performance among LM-based dialogue agents in terms of relevancy and factuality. We show that MixCL achieves comparable performance to state-of-the-art KB-based approaches while enjoying notable advantages in terms of efficiency and scalability.

Weiwei Sun, Hengyi Cai, Hongshen Chen et al.

In open-domain question answering, due to the ambiguity of questions, multiple plausible answers may exist. To provide feasible answers to an ambiguous question, one approach is to directly predict all valid answers, but this can struggle with balancing relevance and diversity. An alternative is to gather candidate answers and aggregate them, but this method can be computationally costly and may neglect dependencies among answers. In this paper, we present AmbigPrompt to address the imperfections of existing approaches to answering ambiguous questions. Specifically, we integrate an answering model with a prompting model in an iterative manner. The prompting model adaptively tracks the reading process and progressively triggers the answering model to compose distinct and relevant answers. Additionally, we develop a task-specific post-pretraining approach for both the answering model and the prompting model, which greatly improves the performance of our framework. Empirical studies on two commonly-used open benchmarks show that AmbigPrompt achieves state-of-the-art or competitive results while using less memory and having a lower inference latency than competing approaches. Additionally, AmbigPrompt also performs well in low-resource settings. The code are available at: https://github.com/sunnweiwei/AmbigPrompt.

Zihan Wang, Na Huang, Fei Sun et al.

Learned recommender systems may inadvertently leak information about their training data, leading to privacy violations. We investigate privacy threats faced by recommender systems through the lens of membership inference. In such attacks, an adversary aims to infer whether a user's data is used to train the target recommender. To achieve this, previous work has used a shadow recommender to derive training data for the attack model, and then predicts the membership by calculating difference vectors between users' historical interactions and recommended items. State-of-the-art methods face two challenging problems: (1) training data for the attack model is biased due to the gap between shadow and target recommenders, and (2) hidden states in recommenders are not observational, resulting in inaccurate estimations of difference vectors. To address the above limitations, we propose a Debiasing Learning for Membership Inference Attacks against recommender systems (DL-MIA) framework that has four main components: (1) a difference vector generator, (2) a disentangled encoder, (3) a weight estimator, and (4) an attack model. To mitigate the gap between recommenders, a variational auto-encoder (VAE) based disentangled encoder is devised to identify recommender invariant and specific features. To reduce the estimation bias, we design a weight estimator, assigning a truth-level score for each difference vector to indicate estimation accuracy. We evaluate DL-MIA against both general recommenders and sequential recommenders on three real-world datasets. Experimental results show that DL-MIA effectively alleviates training and estimation biases simultaneously, and achieves state-of-the-art attack performance.

Xiaoyu Zhang, Xin Xin, Dongdong Li et al.

Conversational recommender systems (CRSs) often utilize external knowledge graphs (KGs) to introduce rich semantic information and recommend relevant items through natural language dialogues. However, original KGs employed in existing CRSs are often incomplete and sparse, which limits the reasoning capability in recommendation. Moreover, only few of existing studies exploit the dialogue context to dynamically refine knowledge from KGs for better recommendation. To address the above issues, we propose the Variational Reasoning over Incomplete KGs Conversational Recommender (VRICR). Our key idea is to incorporate the large dialogue corpus naturally accompanied with CRSs to enhance the incomplete KGs; and perform dynamic knowledge reasoning conditioned on the dialogue context. Specifically, we denote the dialogue-specific subgraphs of KGs as latent variables with categorical priors for adaptive knowledge graphs refactor. We propose a variational Bayesian method to approximate posterior distributions over dialogue-specific subgraphs, which not only leverages the dialogue corpus for restructuring missing entity relations but also dynamically selects knowledge based on the dialogue context. Finally, we infuse the dialogue-specific subgraphs to decode the recommendation and responses. We conduct experiments on two benchmark CRSs datasets. Experimental results confirm the effectiveness of our proposed method.

Yougang Lyu, Piji Li, Yechang Yang et al.

Natural language understanding (NLU) models often rely on dataset biases rather than intended task-relevant features to achieve high performance on specific datasets. As a result, these models perform poorly on datasets outside the training distribution. Some recent studies address this issue by reducing the weights of biased samples during the training process. However, these methods still encode biased latent features in representations and neglect the dynamic nature of bias, which hinders model prediction. We propose an NLU debiasing method, named debiasing contrastive learning (DCT), to simultaneously alleviate the above problems based on contrastive learning. We devise a debiasing, positive sampling strategy to mitigate biased latent features by selecting the least similar biased positive samples. We also propose a dynamic negative sampling strategy to capture the dynamic influence of biases by employing a bias-only model to dynamically select the most similar biased negative samples. We conduct experiments on three NLU benchmark datasets. Experimental results show that DCT outperforms state-of-the-art baselines on out-of-distribution datasets while maintaining in-distribution performance. We also verify that DCT can reduce biased latent features from the model's representation.

Weiwei Sun, Shuyu Guo, Shuo Zhang et al.

Task-oriented dialogue systems (TDSs) are assessed mainly in an offline setting or through human evaluation. The evaluation is often limited to single-turn or is very time-intensive. As an alternative, user simulators that mimic user behavior allow us to consider a broad set of user goals to generate human-like conversations for simulated evaluation. Employing existing user simulators to evaluate TDSs is challenging as user simulators are primarily designed to optimize dialogue policies for TDSs and have limited evaluation capabilities. Moreover, the evaluation of user simulators is an open challenge. In this work, we propose a metaphorical user simulator for end-to-end TDS evaluation, where we define a simulator to be metaphorical if it simulates user's analogical thinking in interactions with systems. We also propose a tester-based evaluation framework to generate variants, i.e., dialogue systems with different capabilities. Our user simulator constructs a metaphorical user model that assists the simulator in reasoning by referring to prior knowledge when encountering new items. We estimate the quality of simulators by checking the simulated interactions between simulators and variants. Our experiments are conducted using three TDS datasets. The proposed user simulator demonstrates better consistency with manual evaluation than an agenda-based simulator and a seq2seq model on three datasets; our tester framework demonstrates efficiency and has been tested on multiple tasks, such as conversational recommendation and e-commerce dialogues.

Weiwei Sun, Pengjie Ren, Zhaochun Ren

Knowledge selection is the key in knowledge-grounded dialogues (KGD), which aims to select an appropriate knowledge snippet to be used in the utterance based on dialogue history. Previous studies mainly employ the classification approach to classify each candidate snippet as "relevant" or "irrelevant" independently. However, such approaches neglect the interactions between snippets, leading to difficulties in inferring the meaning of snippets. Moreover, they lack modeling of the discourse structure of dialogue-knowledge interactions. We propose a simple yet effective generative approach for knowledge selection, called GenKS. GenKS learns to select snippets by generating their identifiers with a sequence-to-sequence model. GenKS therefore captures intra-knowledge interaction inherently through attention mechanisms. Meanwhile, we devise a hyperlink mechanism to model the dialogue-knowledge interactions explicitly. We conduct experiments on three benchmark datasets, and verify GenKS achieves the best results on both knowledge selection and response generation.

Xiaoyu Zhang, Ruobing Xie, Yougang Lyu et al.

Conversational recommender systems (CRSs) are able to elicit user preferences through multi-turn dialogues. They typically incorporate external knowledge and pre-trained language models to capture the dialogue context. Most CRS approaches, trained on benchmark datasets, assume that the standard items and responses in these benchmarks are optimal. However, they overlook that users may express negative emotions with the standard items and may not feel emotionally engaged by the standard responses. This issue leads to a tendency to replicate the logic of recommenders in the dataset instead of aligning with user needs. To remedy this misalignment, we introduce empathy within a CRS. With empathy we refer to a system's ability to capture and express emotions. We propose an empathetic conversational recommender (ECR) framework. ECR contains two main modules: emotion-aware item recommendation and emotion-aligned response generation. Specifically, we employ user emotions to refine user preference modeling for accurate recommendations. To generate human-like emotional responses, ECR applies retrieval-augmented prompts to fine-tune a pre-trained language model aligning with emotions and mitigating hallucination. To address the challenge of insufficient supervision labels, we enlarge our empathetic data using emotion labels annotated by large language models and emotional reviews collected from external resources. We propose novel evaluation metrics to capture user satisfaction in real-world CRS scenarios. Our experiments on the ReDial dataset validate the efficacy of our framework in enhancing recommendation accuracy and improving user satisfaction.

Zhengliang Shi, Weiwei Sun, Shuo Zhang et al.

Evaluating open-domain dialogue systems is challenging for reasons such as the one-to-many problem, i.e., many appropriate responses other than just the golden response. As of now, automatic evaluation methods need better consistency with humans, while reliable human evaluation can be time- and cost-intensive. To this end, we propose the Reference-Assisted Dialogue Evaluation (RADE) approach under the multi-task learning framework, which leverages the pre-created utterance as reference other than the gold response to relief the one-to-many problem. Specifically, RADE explicitly compares reference and the candidate response to predict their overall scores. Moreover, an auxiliary response generation task enhances prediction via a shared encoder. To support RADE, we extend three datasets with additional rated responses other than just a golden response by human annotation. Experiments on our three datasets and two existing benchmarks demonstrate the effectiveness of our method, where Pearson, Spearman, and Kendall correlations with human evaluation outperform state-of-the-art baselines.

Shanshan Wang, Zhumin Chen, Zhaochun Ren et al.

Pre-trained language models (PLM) have demonstrated their effectiveness for a broad range of information retrieval and natural language processing tasks. As the core part of PLM, multi-head self-attention is appealing for its ability to jointly attend to information from different positions. However, researchers have found that PLM always exhibits fixed attention patterns regardless of the input (e.g., excessively paying attention to [CLS] or [SEP]), which we argue might neglect important information in the other positions. In this work, we propose a simple yet effective attention guiding mechanism to improve the performance of PLM by encouraging attention towards the established goals. Specifically, we propose two kinds of attention guiding methods, i.e., map discrimination guiding (MDG) and attention pattern decorrelation guiding (PDG). The former definitely encourages the diversity among multiple self-attention heads to jointly attend to information from different representation subspaces, while the latter encourages self-attention to attend to as many different positions of the input as possible. We conduct experiments with multiple general pre-trained models (i.e., BERT, ALBERT, and Roberta) and domain-specific pre-trained models (i.e., BioBERT, ClinicalBERT, BlueBert, and SciBERT) on three benchmark datasets (i.e., MultiNLI, MedNLI, and Cross-genre-IR). Extensive experimental results demonstrate that our proposed MDG and PDG bring stable performance improvements on all datasets with high efficiency and low cost.

Wentao Deng, Jiahuan Pei, Zhaochun Ren et al.

Answer selection in open-domain dialogues aims to select an accurate answer from candidates. Recent success of answer selection models hinges on training with large amounts of labeled data. However, collecting large-scale labeled data is labor-intensive and time-consuming. In this paper, we introduce the predicted intent labels to calibrate answer labels in a self-training paradigm. Specifically, we propose the intent-calibrated self-training (ICAST) to improve the quality of pseudo answer labels through the intent-calibrated answer selection paradigm, in which we employ pseudo intent labels to help improve pseudo answer labels. We carry out extensive experiments on two benchmark datasets with open-domain dialogues. The experimental results show that ICAST outperforms baselines consistently with 1%, 5% and 10% labeled data. Specifically, it improves 2.06% and 1.00% of F1 score on the two datasets, compared with the strongest baseline with only 5% labeled data.

Mengqi Zhang, Bowen Fang, Qiang Liu et al.

Large language models (LLMs) face challenges with internal knowledge inaccuracies and outdated information. Knowledge editing has emerged as a pivotal approach to mitigate these issues. Although current knowledge editing techniques exhibit promising performance in single-hop reasoning tasks, they show limitations when applied to multi-hop reasoning. Drawing on cognitive neuroscience and the operational mechanisms of LLMs, we hypothesize that the residual single-hop knowledge after editing causes edited models to revert to their original answers when processing multi-hop questions, thereby undermining their performance in multihop reasoning tasks. To validate this hypothesis, we conduct a series of experiments that empirically confirm our assumptions. Building on the validated hypothesis, we propose a novel knowledge editing method that incorporates a Knowledge Erasure mechanism for Large language model Editing (KELE). Specifically, we design an erasure function for residual knowledge and an injection function for new knowledge. Through joint optimization, we derive the optimal recall vector, which is subsequently utilized within a rank-one editing framework to update the parameters of targeted model layers. Extensive experiments on GPT-J and GPT-2 XL demonstrate that KELE substantially enhances the multi-hop reasoning capability of edited LLMs.

Zhiwei Xu, Hangyu Mao, Nianmin Zhang et al.

In partially observable multi-agent systems, agents typically only have access to local observations. This severely hinders their ability to make precise decisions, particularly during decentralized execution. To alleviate this problem and inspired by image outpainting, we propose State Inference with Diffusion Models (SIDIFF), which uses diffusion models to reconstruct the original global state based solely on local observations. SIDIFF consists of a state generator and a state extractor, which allow agents to choose suitable actions by considering both the reconstructed global state and local observations. In addition, SIDIFF can be effortlessly incorporated into current multi-agent reinforcement learning algorithms to improve their performance. Finally, we evaluated SIDIFF on different experimental platforms, including Multi-Agent Battle City (MABC), a novel and flexible multi-agent reinforcement learning environment we developed. SIDIFF achieved desirable results and outperformed other popular algorithms.

Zhengliang Shi, Shen Gao, Xiuyi Chen et al. · baidu

Tool learning empowers large language models (LLMs) as agents to use external tools and extend their utility. Existing methods employ one single LLM-based agent to iteratively select and execute tools, thereafter incorporating execution results into the next action prediction. Despite their progress, these methods suffer from performance degradation when addressing practical tasks due to: (1) the pre-defined pipeline with restricted flexibility to calibrate incorrect actions, and (2) the struggle to adapt a general LLM-based agent to perform a variety of specialized actions. To mitigate these problems, we propose ConAgents, a Cooperative and interactive Agents framework, which coordinates three specialized agents for tool selection, tool execution, and action calibration separately. ConAgents introduces two communication protocols to enable the flexible cooperation of agents. To effectively generalize the ConAgents into open-source models, we also propose specialized action distillation, enhancing their ability to perform specialized actions in our framework. Our extensive experiments on three datasets show that the LLMs, when equipped with the ConAgents, outperform baselines with substantial improvement (i.e., up to 14% higher success rate).

Zhongkun Liu, Zheng Chen, Mengqi Zhang et al.

Fine-tuning has been demonstrated to be an effective method to improve the domain performance of large language models (LLMs). However, LLMs might fit the dataset bias and shortcuts for prediction, leading to poor generation performance. Previous works have proven that LLMs are prone to exhibit position bias, i.e., leveraging information positioned at the beginning or end, or specific positional cues within the input. Existing debiasing methods for LLMs require external bias knowledge or annotated non-biased samples, which is lacking for position debiasing and impractical in reality. In this work, we propose a self-supervised position debiasing (SOD) framework to mitigate position bias for LLMs. SOD leverages unsupervised responses from pre-trained LLMs for debiasing without relying on any external knowledge. To improve the quality of unsupervised responses, we propose an objective alignment (OAM) module to prune these responses. Experiments on eight datasets and five tasks show that SOD consistently outperforms existing methods in mitigating three types of position biases. Besides, SOD achieves this by sacrificing only a small performance on biased samples, which is general and effective. To facilitate the reproducibility of the results, we share the code of all methods and datasets on https://github.com/LZKSKY/SOD.

Ziqi Zhao, Zhaochun Ren, Liu Yang et al.

Offline reinforcement learning (RL) aims to learn policies without online explorations. To enlarge the training data, model-based offline RL learns a dynamics model which is utilized as a virtual environment to generate simulation data and enhance policy learning. However, existing data augmentation methods for offline RL suffer from (i) trivial improvement from short-horizon simulation; and (ii) the lack of evaluation and correction for generated data, leading to low-qualified augmentation. In this paper, we propose offline trajectory optimization for offline reinforcement learning (OTTO). The key motivation is to conduct long-horizon simulation and then utilize model uncertainty to evaluate and correct the augmented data. Specifically, we propose an ensemble of Transformers, a.k.a. World Transformers, to predict environment state dynamics and the reward function. Three strategies are proposed to use World Transformers to generate long-horizon trajectory simulation by perturbing the actions in the offline data. Then, an uncertainty-based World Evaluator is introduced to firstly evaluate the confidence of the generated trajectories and then perform the correction for low-confidence data. Finally, we jointly use the original data with the corrected augmentation data to train an offline RL algorithm. OTTO serves as a plug-in module and can be integrated with existing model-free offline RL methods. Experiments on various benchmarks show that OTTO can effectively improve the performance of representative offline RL algorithms, including in complex environments with sparse rewards like AntMaze. Codes are available at https://github.com/ZiqiZhao1/OTTO.

Wentao Deng, Jiahuan Pei, Zhiwei Xu et al.

A multi-agent system (MAS) enhances its capacity to solve complex natural language processing (NLP) tasks through collaboration among multiple agents, where consensus-seeking serves as a fundamental mechanism. However, existing consensus-seeking approaches typically rely on voting mechanisms to judge consensus, overlooking contradictions in system-internal beliefs that destabilize the consensus. Moreover, these methods often involve agents updating their results through indiscriminate collaboration with every other agent. Such uniform interaction fails to identify the optimal collaborators for each agent, hindering the emergence of a stable consensus. To address these challenges, we provide a theoretical framework for selecting optimal collaborators that maximize consensus stability. Based on the theorems, we propose the Belief-Calibrated Consensus Seeking (BCCS) framework to facilitate stable consensus via selecting optimal collaborators and calibrating the consensus judgment by system-internal beliefs. Experimental results on the MATH and MMLU benchmark datasets demonstrate that the proposed BCCS framework outperforms the best existing results by 2.23% and 3.95% of accuracy on challenging tasks, respectively. Our code and data are available at https://github.com/dengwentao99/BCCS.

Shuyu Guo, Shuo Zhang, Weiwei Sun et al.

Explanations in conventional recommender systems have demonstrated benefits in helping the user understand the rationality of the recommendations and improving the system's efficiency, transparency, and trustworthiness. In the conversational environment, multiple contextualized explanations need to be generated, which poses further challenges for explanations. To better measure explainability in conversational recommender systems (CRS), we propose ten evaluation perspectives based on concepts from conventional recommender systems together with the characteristics of CRS. We assess five existing CRS benchmark datasets using these metrics and observe the necessity of improving the explanation quality of CRS. To achieve this, we conduct manual and automatic approaches to extend these dialogues and construct a new CRS dataset, namely Explainable Recommendation Dialogues (E-ReDial). It includes 756 dialogues with over 2,000 high-quality rewritten explanations. We compare two baseline approaches to perform explanation generation based on E-ReDial. Experimental results suggest that models trained on E-ReDial can significantly improve explainability while introducing knowledge into the models can further improve the performance. GPT-3 in the in-context learning setting can generate more realistic and diverse movie descriptions. In contrast, T5 training on E-ReDial can better generate clear reasons for recommendations based on user preferences. E-ReDial is available at https://github.com/Superbooming/E-ReDial.

Jiabao Fang, Shen Gao, Pengjie Ren et al.

Due to strong capabilities in conducting fluent, multi-turn conversations with users, Large Language Models (LLMs) have the potential to further improve the performance of Conversational Recommender System (CRS). Unlike the aimless chit-chat that LLM excels at, CRS has a clear target. So it is imperative to control the dialogue flow in the LLM to successfully recommend appropriate items to the users. Furthermore, user feedback in CRS can assist the system in better modeling user preferences, which has been ignored by existing studies. However, simply prompting LLM to conduct conversational recommendation cannot address the above two key challenges. In this paper, we propose Multi-Agent Conversational Recommender System (MACRS) which contains two essential modules. First, we design a multi-agent act planning framework, which can control the dialogue flow based on four LLM-based agents. This cooperative multi-agent framework will generate various candidate responses based on different dialogue acts and then choose the most appropriate response as the system response, which can help MACRS plan suitable dialogue acts. Second, we propose a user feedback-aware reflection mechanism which leverages user feedback to reason errors made in previous turns to adjust the dialogue act planning, and higher-level user information from implicit semantics. We conduct extensive experiments based on user simulator to demonstrate the effectiveness of MACRS in recommendation and user preferences collection. Experimental results illustrate that MACRS demonstrates an improvement in user interaction experience compared to directly using LLMs.

Mengqi Zhang, Xiaotian Ye, Qiang Liu et al.

Large language models (LLMs) are pivotal in advancing natural language processing (NLP) tasks, yet their efficacy is hampered by inaccuracies and outdated knowledge. Model editing emerges as a promising solution to address these challenges. However, existing editing methods struggle to track and incorporate changes in knowledge associated with edits, which limits the generalization ability of postedit LLMs in processing edited knowledge. To tackle these problems, we propose a novel model editing method that leverages knowledge graphs for enhancing LLM editing, namely GLAME. Specifically, we first utilize a knowledge graph augmentation module to uncover associated knowledge that has changed due to editing, obtaining its internal representations within LLMs. This approach allows knowledge alterations within LLMs to be reflected through an external graph structure. Subsequently, we design a graph-based knowledge edit module to integrate structured knowledge into the model editing. This ensures that the updated parameters reflect not only the modifications of the edited knowledge but also the changes in other associated knowledge resulting from the editing process. Comprehensive experiments conducted on GPT-J and GPT-2 XL demonstrate that GLAME significantly improves the generalization capabilities of post-edit LLMs in employing edited knowledge.

Shiguang Wu, Wenda Wei, Mengqi Zhang et al.

Generative retrieval generates identifiers of relevant documents in an end-to-end manner using a sequence-to-sequence architecture for a given query. The relation between generative retrieval and other retrieval methods, especially those based on matching within dense retrieval models, is not yet fully comprehended. Prior work has demonstrated that generative retrieval with atomic identifiers is equivalent to single-vector dense retrieval. Accordingly, generative retrieval exhibits behavior analogous to hierarchical search within a tree index in dense retrieval when using hierarchical semantic identifiers. However, prior work focuses solely on the retrieval stage without considering the deep interactions within the decoder of generative retrieval. In this paper, we fill this gap by demonstrating that generative retrieval and multi-vector dense retrieval share the same framework for measuring the relevance to a query of a document. Specifically, we examine the attention layer and prediction head of generative retrieval, revealing that generative retrieval can be understood as a special case of multi-vector dense retrieval. Both methods compute relevance as a sum of products of query and document vectors and an alignment matrix. We then explore how generative retrieval applies this framework, employing distinct strategies for computing document token vectors and the alignment matrix. We have conducted experiments to verify our conclusions and show that both paradigms exhibit commonalities of term matching in their alignment matrix.

Jiahuan Pei, Irene Viola, Haochen Huang et al.

Autonomous artificial intelligence (AI) agents have emerged as promising protocols for automatically understanding the language-based environment, particularly with the exponential development of large language models (LLMs). However, a fine-grained, comprehensive understanding of multimodal environments remains under-explored. This work designs an autonomous workflow tailored for integrating AI agents seamlessly into extended reality (XR) applications for fine-grained training. We present a demonstration of a multimodal fine-grained training assistant for LEGO brick assembly in a pilot XR environment. Specifically, we design a cerebral language agent that integrates LLM with memory, planning, and interaction with XR tools and a vision-language agent, enabling agents to decide their actions based on past experiences. Furthermore, we introduce LEGO-MRTA, a multimodal fine-grained assembly dialogue dataset synthesized automatically in the workflow served by a commercial LLM. This dataset comprises multimodal instruction manuals, conversations, XR responses, and vision question answering. Last, we present several prevailing open-resource LLMs as benchmarks, assessing their performance with and without fine-tuning on the proposed dataset. We anticipate that the broader impact of this workflow will advance the development of smarter assistants for seamless user interaction in XR environments, fostering research in both AI and HCI communities.

Pengjie Ren, Chengshun Shi, Shiguang Wu et al.

Parameter-efficient fine-tuning (PEFT) is a popular method for tailoring pre-trained large language models (LLMs), especially as the models' scale and the diversity of tasks increase. Low-rank adaptation (LoRA) is based on the idea that the adaptation process is intrinsically low-dimensional, i.e., significant model changes can be represented with relatively few parameters. However, decreasing the rank encounters challenges with generalization errors for specific tasks when compared to full-parameter fine-tuning. We present MELoRA, a mini-ensemble low-rank adapters that uses fewer trainable parameters while maintaining a higher rank, thereby offering improved performance potential. The core idea is to freeze original pretrained weights and train a group of mini LoRAs with only a small number of parameters. This can capture a significant degree of diversity among mini LoRAs, thus promoting better generalization ability. We conduct a theoretical analysis and empirical studies on various NLP tasks. Our experimental results show that, compared to LoRA, MELoRA achieves better performance with 8 times fewer trainable parameters on natural language understanding tasks and 36 times fewer trainable parameters on instruction following tasks, which demonstrates the effectiveness of MELoRA.

Yougang Lyu, Lingyong Yan, Shuaiqiang Wang et al. · baidu

Despite their success at many natural language processing (NLP) tasks, large language models still struggle to effectively leverage knowledge for knowledge-intensive tasks, manifesting limitations such as generating incomplete, non-factual, or illogical answers. These limitations stem from inadequate knowledge awareness of LLMs during vanilla fine-tuning. To address these problems, we propose a knowledge-aware fine-tuning (KnowTuning) method to improve fine-grained and coarse-grained knowledge awareness of LLMs. We devise a fine-grained knowledge augmentation stage to train LLMs to identify difficult fine-grained knowledge in answers. We also propose a coarse-grained knowledge comparison stage to train LLMs to distinguish between reliable and unreliable knowledge, in three aspects: completeness, factuality, and logicality. Extensive experiments on both generic and medical question answering (QA) datasets confirm the effectiveness of KnowTuning, through automatic and human evaluations, across various sizes of LLMs. We further verify that KnowTuning generates more facts with less factual error rate under fine-grained facts evaluation.

Xin Xin, Liu Yang, Ziqi Zhao et al.

Session-based recommendation predicts users' future interests from previous interactions in a session. Despite the memorizing of historical samples, the request of unlearning, i.e., to remove the effect of certain training samples, also occurs for reasons such as user privacy or model fidelity. However, existing studies on unlearning are not tailored for the session-based recommendation. On the one hand, these approaches cannot achieve satisfying unlearning effects due to the collaborative correlations and sequential connections between the unlearning item and the remaining items in the session. On the other hand, seldom work has conducted the research to verify the unlearning effectiveness in the session-based recommendation scenario. In this paper, we propose SRU, a session-based recommendation unlearning framework, which enables high unlearning efficiency, accurate recommendation performance, and improved unlearning effectiveness in session-based recommendation. Specifically, we first partition the training sessions into separate sub-models according to the similarity across the sessions, then we utilize an attention-based aggregation layer to fuse the hidden states according to the correlations between the session and the centroid of the data in the sub-model. To improve the unlearning effectiveness, we further propose three extra data deletion strategies, including collaborative extra deletion (CED), neighbor extra deletion (NED), and random extra deletion (RED). Besides, we propose an evaluation metric that measures whether the unlearning sample can be inferred after the data deletion to verify the unlearning effectiveness. We implement SRU with three representative session-based recommendation models and conduct experiments on three benchmark datasets. Experimental results demonstrate the effectiveness of our methods.

Zhengliang Shi, Yuhan Wang, Lingyong Yan et al. · baidu

Tool learning aims to augment large language models (LLMs) with diverse tools, enabling them to act as agents for solving practical tasks. Due to the limited context length of tool-using LLMs, adopting information retrieval (IR) models to select useful tools from large toolsets is a critical initial step. However, the performance of IR models in tool retrieval tasks remains underexplored and unclear. Most tool-use benchmarks simplify this step by manually pre-annotating a small set of relevant tools for each task, which is far from the real-world scenarios. In this paper, we propose ToolRet, a heterogeneous tool retrieval benchmark comprising 7.6k diverse retrieval tasks, and a corpus of 43k tools, collected from existing datasets. We benchmark six types of models on ToolRet. Surprisingly, even the models with strong performance in conventional IR benchmarks, exhibit poor performance on ToolRet. This low retrieval quality degrades the task pass rate of tool-use LLMs. As a further step, we contribute a large-scale training dataset with over 200k instances, which substantially optimizes the tool retrieval ability of IR models.

Wenyu Wang, Mengqi Zhang, Xiaotian Ye et al.

Large Language Models (LLMs) inevitably acquire harmful information during training on massive datasets. LLM unlearning aims to eliminate the influence of such harmful information while maintaining the model's overall performance. Existing unlearning methods, represented by gradient ascent-based approaches, primarily focus on forgetting target data while overlooking the crucial impact of logically related knowledge on the effectiveness of unlearning. In this paper, through both theoretical and experimental analyses, we first demonstrate that a key reason for the suboptimal unlearning performance is that models can reconstruct the target content through reasoning with logically related knowledge. To address this issue, we propose Unlearning Improvement via Parameter Extrapolation (UIPE), a method that removes knowledge highly correlated with the forgetting targets. Experimental results show that UIPE significantly enhances the performance of various mainstream LLM unlearning methods on the TOFU benchmark.

Mengqi Zhang, Zisheng Zhou, Xiaotian Ye et al.

Knowledge Editing has emerged as a promising solution for efficiently updating embedded knowledge in large language models (LLMs). While existing approaches demonstrate effectiveness in integrating new knowledge and preserving the original capabilities of LLMs, they fail to maintain fine-grained irrelevant knowledge facts that share the same subject as edited knowledge but differ in relation and object. This challenge arises because subject representations inherently encode multiple attributes, causing the target and fine-grained irrelevant knowledge to become entangled in the representation space, and thus vulnerable to unintended alterations during editing. To address this, we propose DiKE, a novel approach that Disentangles Knowledge representations for LLM Editing (DiKE). DiKE consists of two key components: a Knowledge Representation Disentanglement (KRD) module that decomposes the subject representation into target-knowledgerelated and -unrelated components, and a Disentanglement-based Knowledge Edit (DKE) module that updates only the target-related component while explicitly preserving the unrelated one. We further derive a closed-form, rank-one parameter update based on matrix theory to enable efficient and minimally invasive edits. To rigorously evaluate fine-grained irrelevant knowledge preservation, we construct FINE-KED, a new benchmark comprising fine-grained irrelevant knowledge at different levels of relational similarity to the edited knowledge. Extensive experiments across multiple LLMs demonstrate that DiKE substantially improves fine-grained irrelevant knowledge preservation while maintaining competitive general editing performance.

Haochen Wang, Zhiwei Shi, Chengxi Zhu et al.

Learning-based methods, such as imitation learning (IL) and reinforcement learning (RL), can produce excel control policies over challenging agile robot tasks, such as sports robot. However, no existing work has harmonized learning-based policy with model-based methods to reduce training complexity and ensure the safety and stability for agile badminton robot control. In this paper, we introduce Hamlet, a novel hybrid control system for agile badminton robots. Specifically, we propose a model-based strategy for chassis locomotion which provides a base for arm policy. We introduce a physics-informed "IL+RL" training framework for learning-based arm policy. In this train framework, a model-based strategy with privileged information is used to guide arm policy training during both IL and RL phases. In addition, we train the critic model during IL phase to alleviate the performance drop issue when transitioning from IL to RL. We present results on our self-engineered badminton robot, achieving 94.5% success rate against the serving machine and 90.7% success rate against human players. Our system can be easily generalized to other agile mobile manipulation tasks such as agile catching and table tennis. Our project website: https://dreamstarring.github.io/HAMLET/.

Junyan Hu, Xue Xiao, Mengqi Zhang et al.

As large language models (LLMs) grow in size, traditional full fine-tuning becomes increasingly impractical due to its high computational and storage costs. Although popular parameter-efficient fine-tuning methods, such as LoRA, have significantly reduced the number of tunable parameters, there is still room for further optimization. In this work, we propose ASLoRA, a cross-layer parameter-sharing strategy combining global sharing with partial adaptive sharing. Specifically, we share the low-rank matrix A across all layers and adaptively merge matrix B during training. This sharing mechanism not only mitigates overfitting effectively but also captures inter-layer dependencies, significantly enhancing the model's representational capability. We conduct extensive experiments on various NLP tasks, showing that ASLoRA outperforms LoRA while using less than 25% of the parameters, highlighting its flexibility and superior parameter efficiency. Furthermore, in-depth analyses of the adaptive sharing strategy confirm its significant advantages in enhancing both model flexibility and task adaptability.

Yinuo Wang, Baiyang Wang, Robert E. Mercer et al.

Trustworthiness in healthcare question-answering (QA) systems is important for ensuring patient safety, clinical effectiveness, and user confidence. As large language models (LLMs) become increasingly integrated into medical settings, the reliability of their responses directly influences clinical decision-making and patient outcomes. However, achieving comprehensive trustworthiness in medical QA poses significant challenges due to the inherent complexity of healthcare data, the critical nature of clinical scenarios, and the multifaceted dimensions of trustworthy AI. In this survey, we systematically examine six key dimensions of trustworthiness in medical QA, i.e., Factuality, Robustness, Fairness, Safety, Explainability, and Calibration. We review how each dimension is evaluated in existing LLM-based medical QA systems. We compile and compare major benchmarks designed to assess these dimensions and analyze evaluation-guided techniques that drive model improvements, such as retrieval-augmented grounding, adversarial fine-tuning, and safety alignment. Finally, we identify open challenges-such as scalable expert evaluation, integrated multi-dimensional metrics, and real-world deployment studies-and propose future research directions to advance the safe, reliable, and transparent deployment of LLM-powered medical QA.

Minghang Zhu, Shen Gao, Zhengliang Shi et al.

A common training approach for language models involves using a large-scale language model to expand a human-provided dataset, which is subsequently used for model training.This method significantly reduces training costs by eliminating the need for extensive human data annotation. However, it still faces challenges such as high carbon emissions during data augmentation and the risk of data leakage when we use closed-source LLMs. To address these issues, we propose a self-evolution method for language models. First, we introduce the Multi-level Principle Generation, which enables a large-scale model to summarize task-completion principles based on a small amount of task data. Then, we propose the Principle-based Instance Generation, in which a smaller-scale language model uses these task principles to generate a large amount of data. This data is then used for model training. Experimental results show that our proposed method significantly improves model performance compared to directly using a smaller-scale language model to generate data. Additionally, since we only use the large-scale language model to generate the task-completion principles, the carbon emissions associated with training the model are greatly reduced.

Zhengliang Shi, Weiwei Sun, Shen Gao et al.

Multi-Hop Question Answering (MHQA) tasks present a significant challenge for large language models (LLMs) due to the intensive knowledge required. Current solutions, like Retrieval-Augmented Generation, typically retrieve potential documents from an external corpus to read an answer. However, the performance of this retrieve-then-read paradigm is constrained by the retriever and the inevitable noise in the retrieved documents. To mitigate these challenges, we introduce a novel generate-then-ground (GenGround) framework, synergizing the parametric knowledge of LLMs and external documents to solve a multi-hop question. GenGround empowers LLMs to alternate two phases until the final answer is derived: (1) formulate a simpler, single-hop question and directly generate the answer; (2) ground the question-answer pair in retrieved documents, amending any wrong predictions in the answer. We also propose an instructional grounding distillation method to generalize our method into smaller models. Extensive experiments conducted on four datasets illustrate the superiority of our method.

Yougang Lyu, Jitai Hao, Zihan Wang et al.

Multiple defendants in a criminal fact description generally exhibit complex interactions, and cannot be well handled by existing Legal Judgment Prediction (LJP) methods which focus on predicting judgment results (e.g., law articles, charges, and terms of penalty) for single-defendant cases. To address this problem, we propose the task of multi-defendant LJP, which aims to automatically predict the judgment results for each defendant of multi-defendant cases. Two challenges arise with the task of multi-defendant LJP: (1) indistinguishable judgment results among various defendants; and (2) the lack of a real-world dataset for training and evaluation. To tackle the first challenge, we formalize the multi-defendant judgment process as hierarchical reasoning chains and introduce a multi-defendant LJP method, named Hierarchical Reasoning Network (HRN), which follows the hierarchical reasoning chains to determine criminal relationships, sentencing circumstances, law articles, charges, and terms of penalty for each defendant. To tackle the second challenge, we collect a real-world multi-defendant LJP dataset, namely MultiLJP, to accelerate the relevant research in the future. Extensive experiments on MultiLJP verify the effectiveness of our proposed HRN.

Shen Gao, Zhitao Yao, Chongyang Tao et al.

Summarization quality evaluation is a non-trivial task in text summarization. Contemporary methods can be mainly categorized into two scenarios: (1) reference-based: evaluating with human-labeled reference summary; (2) reference-free: evaluating the summary consistency of the document. Recent studies mainly focus on one of these scenarios and explore training neural models built on PLMs to align with human criteria. However, the models from different scenarios are optimized individually, which may result in sub-optimal performance since they neglect the shared knowledge across different scenarios. Besides, designing individual models for each scenario caused inconvenience to the user. Inspired by this, we propose Unified Multi-scenario Summarization Evaluation Model (UMSE). More specifically, we propose a perturbed prefix tuning method to share cross-scenario knowledge between scenarios and use a self-supervised training paradigm to optimize the model without extra human labeling. Our UMSE is the first unified summarization evaluation framework engaged with the ability to be used in three evaluation scenarios. Experimental results across three typical scenarios on the benchmark dataset SummEval indicate that our UMSE can achieve comparable performance with several existing strong methods which are specifically designed for each scenario.

Minxing Zhang, Zhaochun Ren, Zihan Wang et al.

Recently, recommender systems have achieved promising performances and become one of the most widely used web applications. However, recommender systems are often trained on highly sensitive user data, thus potential data leakage from recommender systems may lead to severe privacy problems. In this paper, we make the first attempt on quantifying the privacy leakage of recommender systems through the lens of membership inference. In contrast with traditional membership inference against machine learning classifiers, our attack faces two main differences. First, our attack is on the user-level but not on the data sample-level. Second, the adversary can only observe the ordered recommended items from a recommender system instead of prediction results in the form of posterior probabilities. To address the above challenges, we propose a novel method by representing users from relevant items. Moreover, a shadow recommender is established to derive the labeled training data for training the attack model. Extensive experimental results show that our attack framework achieves a strong performance. In addition, we design a defense mechanism to effectively mitigate the membership inference threat of recommender systems.

Guojun Yan, Jiahuan Pei, Pengjie Ren et al.

Medical dialogue systems (MDSs) aim to assist doctors and patients with a range of professional medical services, i.e., diagnosis, treatment and consultation. The development of MDSs is hindered because of a lack of resources. In particular. (1) there is no dataset with large-scale medical dialogues that covers multiple medical services and contains fine-grained medical labels (i.e., intents, actions, slots, values), and (2) there is no set of established benchmarks for MDSs for multi-domain, multi-service medical dialogues. In this paper, we present ReMeDi, a set of resource for medical dialogues. ReMeDi consists of two parts, the ReMeDi dataset and the ReMeDi benchmarks. The ReMeDi dataset contains 96,965 conversations between doctors and patients, including 1,557 conversations with fine-gained labels. It covers 843 types of diseases, 5,228 medical entities, and 3 specialties of medical services across 40 domains. To the best of our knowledge, the ReMeDi dataset is the only medical dialogue dataset that covers multiple domains and services, and has fine-grained medical labels. The second part of the ReMeDi resources consists of a set of state-of-the-art models for (medical) dialogue generation. The ReMeDi benchmark has the following methods: (1) pretrained models (i.e., BERT-WWM, BERT-MED, GPT2, and MT5) trained, validated, and tested on the ReMeDi dataset, and (2) a self-supervised contrastive learning(SCL) method to expand the ReMeDi dataset and enhance the training of the state-of-the-art pretrained models. We describe the creation of the ReMeDi dataset, the ReMeDi benchmarking methods, and establish experimental results using the ReMeDi benchmarking methods on the ReMeDi dataset for future research to compare against. With this paper, we share the dataset, implementations of the benchmarks, and evaluation scripts.

Zhongkun Liu, Pengjie Ren, Zhumin Chen et al.

Conversational Question Simplification (CQS) aims to simplify self-contained questions into conversational ones by incorporating some conversational characteristics, e.g., anaphora and ellipsis. Existing maximum likelihood estimation (MLE) based methods often get trapped in easily learned tokens as all tokens are treated equally during training. In this work, we introduce a Reinforcement Iterative Sequence Editing (RISE) framework that optimizes the minimum Levenshtein distance (MLD) through explicit editing actions. RISE is able to pay attention to tokens that are related to conversational characteristics. To train RISE, we devise an Iterative Reinforce Training (IRT) algorithm with a Dynamic Programming based Sampling (DPS) process to improve exploration. Experimental results on two benchmark datasets show that RISE significantly outperforms state-of-the-art methods and generalizes well on unseen data.

Shanshan Wang, Pengjie Ren, Zhumin Chen et al.

Electronic health record (EHR) coding is the task of assigning ICD codes to each EHR. Most previous studies either only focus on the frequent ICD codes or treat rare and frequent ICD codes in the same way. These methods perform well on frequent ICD codes but due to the extremely unbalanced distribution of ICD codes, the performance on rare ones is far from satisfactory. We seek to improve the performance for both frequent and rare ICD codes by using a contrastive graph-based EHR coding framework, CoGraph, which re-casts EHR coding as a few-shot learning task. First, we construct a heterogeneous EHR word-entity (HEWE) graph for each EHR, where the words and entities extracted from an EHR serve as nodes and the relations between them serve as edges. Then, CoGraph learns similarities and dissimilarities between HEWE graphs from different ICD codes so that information can be transferred among them. In a few-shot learning scenario, the model only has access to frequent ICD codes during training, which might force it to encode features that are useful for frequent ICD codes only. To mitigate this risk, CoGraph devises two graph contrastive learning schemes, GSCL and GECL, that exploit the HEWE graph structures so as to encode transferable features. GSCL utilizes the intra-correlation of different sub-graphs sampled from HEWE graphs while GECL exploits the inter-correlation among HEWE graphs at different clinical stages. Experiments on the MIMIC-III benchmark dataset show that CoGraph significantly outperforms state-of-the-art methods on EHR coding, not only on frequent ICD codes, but also on rare codes, in terms of several evaluation indicators. On frequent ICD codes, GSCL and GECL improve the classification accuracy and F1 by 1.31% and 0.61%, respectively, and on rare ICD codes CoGraph has more obvious improvements by 2.12% and 2.95%.

Muyang Ma, Pengjie Ren, Zhumin Chen et al.

One of the key challenges in Sequential Recommendation (SR) is how to extract and represent user preferences. Traditional SR methods rely on the next item as the supervision signal to guide preference extraction and representation. We propose a novel learning strategy, named preference editing. The idea is to force the SR model to discriminate the common and unique preferences in different sequences of interactions between users and the recommender system. By doing so, the SR model is able to learn how to identify common and unique user preferences, and thereby do better user preference extraction and representation. We propose a transformer based SR model, named MrTransformer (Multi-preference Transformer), that concatenates some special tokens in front of the sequence to represent multiple user preferences and makes sure they capture different aspects through a preference coverage mechanism. Then, we devise a preference editing-based self-supervised learning mechanism for training MrTransformer which contains two main operations: preference separation and preference recombination. The former separates the common and unique user preferences for a given pair of sequences. The latter swaps the common preferences to obtain recombined user preferences for each sequence. Based on the preference separation and preference recombination operations, we define two types of SSL loss that require that the recombined preferences are similar to the original ones, and the common preferences are close to each other. We carry out extensive experiments on two benchmark datasets. MrTransformer with preference editing significantly outperforms state-of-the-art SR methods in terms of Recall, MRR and NDCG. We find that long sequences whose user preferences are harder to extract and represent benefit most from preference editing.

Pengjie Ren, Zhongkun Liu, Xiaomeng Song et al.

Conversational information seeking (CIS) is playing an increasingly important role in connecting people to information. Due to the lack of suitable resource, previous studies on CIS are limited to the study of theoretical/conceptual frameworks, laboratory-based user studies, or a particular aspect of CIS (e.g., asking clarifying questions). In this work, we make efforts to facilitate research on CIS from three aspects. (1) We formulate a pipeline for CIS with six sub-tasks: intent detection (ID), keyphrase extraction (KE), action prediction (AP), query selection (QS), passage selection (PS), and response generation (RG). (2) We release a benchmark dataset, called wizard of search engine (WISE), which allows for comprehensive and in-depth research on all aspects of CIS. (3) We design a neural architecture capable of training and evaluating both jointly and separately on the six sub-tasks, and devise a pre-train/fine-tune learning scheme, that can reduce the requirements of WISE in scale by making full use of available data. We report some useful characteristics of CIS based on statistics of WISE. We also show that our best performing model variant isable to achieve effective CIS as indicated by several metrics. We release the dataset, the code, as well as the evaluation scripts to facilitate future research by measuring further improvements in this important research direction.

Zihan Wang, Hongye Song, Zhaochun Ren et al.

Contract element extraction (CEE) is the novel task of automatically identifying and extracting legally relevant elements such as contract dates, payments, and legislation references from contracts. Automatic methods for this task view it as a sequence labeling problem and dramatically reduce human labor. However, as contract genres and element types may vary widely, a significant challenge for this sequence labeling task is how to transfer knowledge from one domain to another, i.e., cross-domain CEE. Cross-domain CEE differs from cross-domain named entity recognition (NER) in two important ways. First, contract elements are far more fine-grained than named entities, which hinders the transfer of extractors. Second, the extraction zones for cross-domain CEE are much larger than for cross-domain NER. As a result, the contexts of elements from different domains can be more diverse. We propose a framework, the Bi-directional Feedback cLause-Element relaTion network (Bi-FLEET), for the cross-domain CEE task that addresses the above challenges. Bi-FLEET has three main components: (1) a context encoder, (2) a clause-element relation encoder, and (3) an inference layer. To incorporate invariant knowledge about element and clause types, a clause-element graph is constructed across domains and a hierarchical graph neural network is adopted in the clause-element relation encoder. To reduce the influence of context variations, a multi-task framework with a bi-directional feedback scheme is designed in the inference layer, conducting both clause classification and element extraction. The experimental results over both cross-domain NER and CEE tasks show that Bi-FLEET significantly outperforms state-of-the-art baselines.

Dongdong Li, Zhaochun Ren, Pengjie Ren et al.

Medical dialogue generation aims to provide automatic and accurate responses to assist physicians to obtain diagnosis and treatment suggestions in an efficient manner. In medical dialogues two key characteristics are relevant for response generation: patient states (such as symptoms, medication) and physician actions (such as diagnosis, treatments). In medical scenarios large-scale human annotations are usually not available, due to the high costs and privacy requirements. Hence, current approaches to medical dialogue generation typically do not explicitly account for patient states and physician actions, and focus on implicit representation instead. We propose an end-to-end variational reasoning approach to medical dialogue generation. To be able to deal with a limited amount of labeled data, we introduce both patient state and physician action as latent variables with categorical priors for explicit patient state tracking and physician policy learning, respectively. We propose a variational Bayesian generative approach to approximate posterior distributions over patient states and physician actions. We use an efficient stochastic gradient variational Bayes estimator to optimize the derived evidence lower bound, where a 2-stage collapsed inference method is proposed to reduce the bias during model training. A physician policy network composed of an action-classifier and two reasoning detectors is proposed for augmented reasoning ability. We conduct experiments on three datasets collected from medical platforms. Our experimental results show that the proposed method outperforms state-of-the-art baselines in terms of objective and subjective evaluation metrics. Our experiments also indicate that our proposed semi-supervised reasoning method achieves a comparable performance as state-of-the-art fully supervised learning baselines for physician policy learning.

Muyang Ma, Pengjie Ren, Zhumin Chen et al.

Cross-domain sequential recommendation is the task of predict the next item that the user is most likely to interact with based on past sequential behavior from multiple domains. One of the key challenges in cross-domain sequential recommendation is to grasp and transfer the flow of information from multiple domains so as to promote recommendations in all domains. Previous studies have investigated the flow of behavioral information by exploring the connection between items from different domains. The flow of knowledge (i.e., the connection between knowledge from different domains) has so far been neglected. In this paper, we propose a mixed information flow network for cross-domain sequential recommendation to consider both the flow of behavioral information and the flow of knowledge by incorporating a behavior transfer unit and a knowledge transfer unit. The proposed mixed information flow network is able to decide when cross-domain information should be used and, if so, which cross-domain information should be used to enrich the sequence representation according to users' current preferences. Extensive experiments conducted on four e-commerce datasets demonstrate that mixed information flow network is able to further improve recommendation performance in different domains by modeling mixed information flow.

Qintong Li, Piji Li, Zhaochun Ren et al.

Lack of external knowledge makes empathetic dialogue systems difficult to perceive implicit emotions and learn emotional interactions from limited dialogue history. To address the above problems, we propose to leverage external knowledge, including commonsense knowledge and emotional lexical knowledge, to explicitly understand and express emotions in empathetic dialogue generation. We first enrich the dialogue history by jointly interacting with external knowledge and construct an emotional context graph. Then we learn emotional context representations from the knowledge-enriched emotional context graph and distill emotional signals, which are the prerequisites to predicate emotions expressed in responses. Finally, to generate the empathetic response, we propose an emotional cross-attention mechanism to learn the emotional dependencies from the emotional context graph. Extensive experiments conducted on a benchmark dataset verify the effectiveness of the proposed method. In addition, we find the performance of our method can be further improved by integrating with a pre-trained model that works orthogonally.

Yangjun Zhang, Pengjie Ren, Maarten de Rijke

Conversational interfaces are increasingly popular as a way of connecting people to information. Corpus-based conversational interfaces are able to generate more diverse and natural responses than template-based or retrieval-based agents. With their increased generative capacity of corpusbased conversational agents comes the need to classify and filter out malevolent responses that are inappropriate in terms of content and dialogue acts. Previous studies on the topic of recognizing and classifying inappropriate content are mostly focused on a certain category of malevolence or on single sentences instead of an entire dialogue. In this paper, we define the task of Malevolent Dialogue Response Detection and Classification (MDRDC). We make three contributions to advance research on this task. First, we present a Hierarchical Malevolent Dialogue Taxonomy (HMDT). Second, we create a labelled multi-turn dialogue dataset and formulate the MDRDC task as a hierarchical classification task over this taxonomy. Third, we apply stateof-the-art text classification methods to the MDRDC task and report on extensive experiments aimed at assessing the performance of these approaches.