论文检索

Jianqi Gao, Jian Cao, Ranran Bu, Nengjun Zhu, Wei Guan, Hang Yu

Knowledge base question answering (KBQA) refers to the system that produces answers to user queries by reasoning with a large-scale structured knowledge base. Advanced works have achieved great success either by generating logical forms (LF) or directly generating answers. Although the former typically yields better performance, these generated LF could be inaccurate, e.g., non-executable. In this regard, large language models (LLMs) have shown exciting potential for accurate generation. However, it is challenging to fine-tune LLMs to generate LF. This is because the context retrieved for prediction typically leads to an excessive number of reasoning paths. In this context, LLMs can generate numerous LF corresponding to these reasoning paths, but a few LF can result in correct answers. Thus, fine-tuning LLMs to generate answer-relevant LF would conflict with the prior knowledge of the LLMs. In this work, we propose a novel learning framework, FM-KBQA, to fine-tune LLMs using multi-task learning for KBQA. Specifically, we propose to fine-tune LLMs using an additional objective: generating the index of reasoning paths that lead to correct answers. This will direct LLMs to pay attention to answer-relevant paths among numerous reasoning paths by completing a simple task where the selected reasoning paths can be supplementary for non-executable LF. Directly generating answers can make LLMs pay attention to the answer-relevant reasoning paths, but it is much more challenging than generating the index of reasoning paths. To verify FM-KBQA's effectiveness, we conduct experiments on mainstream benchmarks, such as WebQuestionsSP (WQSP) and ComplexWebQuestions (CWQ). Extensive evaluations across two public benchmark datasets underscore the superiority of FM-KBQA over current state-of-the-art methods.

Jian Gao, Xiao Zhang, Miao Li, Ji Wu

Causal language models acquire vast amount of knowledge from general text corpus during pretraining, but the efficiency of knowledge learning is known to be unsatisfactory, especially when learning from knowledge-dense and small-sized corpora. The deficiency can come from long-distance dependencies which are hard to capture by language models, and overfitting to co-occurrence patterns and distracting clues in the training text. To address these issues, the paper proposes a method to enhance knowledge learning during language model pretraining, by enhancing elusive but important clues in text discovered by the language model themselves. We found that larger language models pay more attention to non-obvious but important clues, which are often overlooked by smaller language models. Therefore, we can identify these clues by contrasting the attention weights of large and small language models. We use the identified clues as a guide to perform token-dropout data augmentation on the training text, and observed a significant boost in both small and large models' performance in fact memorization. This shows that the behavior contrast between more and less-performant language models contains important clues for knowledge learning, and it can be "amplified" for a straight-forward improvement in knowledge learning efficiency.

Yifan Feng, Weimin Li, Yue Wang, Jingchao Wang, Fangfang Liu, Zhongming Han

Early detection of fake news is crucial to mitigate its negative impact. Current research in fake news detection often utilizes the difference between real and fake news regarding the support degree from reliable sources. However, it has overlooked their different semantic outlier degrees among unreliable source information during the same period. Since fake news often serves idea propaganda, unreliable sources usually publish a lot of information with the same propaganda idea during the same period, making it less likely to be a semantic outlier. To leverage this difference, we propose the Reliable-Unreliable Source Reference (RUSR) Fake News Early Detection Method. RUSR introduces the publication background for detected news, which consists of related news with common main objects of description and slightly earlier publication from both reliable and unreliable sources. Furthermore, we develop a strongly preference-driven support degree evaluation model and a two-hop semantic outlier degree evaluation model, which respectively mitigate the interference of news with weak validation effectiveness and the tightness degree of semantic cluster. The designed redistribution module and expanding range relative time encoding are adopted by both models, respectively optimizing early checkpoint of training and expressing the relevance of news implied by their release time gap. Finally, we present a multi-model mutual benefit and collaboration framework that enables the multi-model mutual benefit of generalization in training and multi-perspective prediction of news authenticity in inference. Experiments on our newly constructed dataset demonstrate the superiority of RUSR.

Xuan Feng, Tianlong Gu, Xiaoli Liu, Liang Chang

Unnatural text correction aims to automatically detect and correct spelling errors or adversarial perturbation errors in sentences. Existing methods typically rely on fine-tuning or adversarial training to correct errors, which have achieved significant success. However, these methods exhibit poor generalization performance due to the difference in data distribution between training data and real-world scenarios, known as the exposure bias problem. In this paper, we propose a self-correct adversarial training framework for learning from mistakes (LIMIT), which is a task- and model-independent framework to correct unnatural errors or mistakes. Specifically, we fully utilize errors generated by the model that are actively exposed during the inference phase, i.e., predictions that are inconsistent with the target. This training method not only simulates potential errors in real application scenarios, but also mitigates the exposure bias of the traditional training process. Meanwhile, we design a novel decoding intervention strategy to maintain semantic consistency. Extensive experimental results on Chinese unnatural text error correction datasets show that our proposed method can correct multiple forms of errors and outperforms the state-of-the-art text correction methods. In addition, extensive results on Chinese and English datasets validate that LIMIT can serve as a plug-and-play defense module and can extend to new models and datasets without further training.

Yuanheng Fang, Guoqing Chao, Wenqiang Lei, Shaobo Li, Dianhui Chu

Large Language Models (LLMs) have recently achieved impressive results in complex reasoning tasks through Chain of Thought (CoT) prompting. However, most existing CoT methods rely on using the same prompts, whether manually designed or automatically generated, to handle the entire dataset. This one-size-fits-all approach may fail to meet the specific needs arising from the diversities within a single dataset. To solve this problem, we propose the Clustered Distance-Weighted Chain of Thought (CDW-CoT) method, which dynamically constructs prompts tailored to the characteristics of each data instance by integrating clustering and prompt optimization techniques. Our method employs clustering algorithms to categorize the dataset into distinct groups, from which a candidate pool of prompts is selected to reflect the inherent diversity within the dataset. For each cluster, CDW-CoT trains the optimal prompt probability distribution tailored to their specific characteristics. Finally, it dynamically constructs a unique prompt probability distribution for each test instance, based on its proximity to cluster centers, from which prompts are selected for reasoning. CDW-CoT consistently outperforms traditional CoT methods across six datasets, including commonsense, symbolic, and mathematical reasoning tasks. Specifically, when compared to manual CoT, CDW-CoT achieves an average accuracy improvement of 25.34% on LLaMA2 (13B) and 15.72% on LLaMA3 (8B).

Xinyue Fang, Zhen Huang, Zhiliang Tian, Minghui Fang, Ziyi Pan, Quntian Fang, Zhihua Wen, Hengyue Pan, Dongsheng Li

LLMs obtain remarkable performance but suffer from hallucinations. Most research on detecting hallucination focuses on questions with short and concrete correct answers that are easy to check faithfulness. Hallucination detections for text generation with open-ended answers are more hard. Some researchers use external knowledge to detect hallucinations in generated texts, but external resources for specific scenarios are hard to access. Recent studies on detecting hallucinations in long texts without external resources conduct consistency comparison among multiple sampled outputs. To handle long texts, researchers split long texts into multiple facts and individually compare the consistency of each pair of facts. However, these methods (1) hardly achieve alignment among multiple facts; (2) overlook dependencies between multiple contextual facts. In this paper, we propose a graph-based context-aware (GCA) hallucination detection method for text generations, which aligns facts and considers the dependencies between contextual facts in consistency comparison. Particularly, to align multiple facts, we conduct a triple-oriented response segmentation to extract multiple knowledge triples. To model dependencies among contextual triples (facts), we construct contextual triples into a graph and enhance triples’ interactions via message passing and aggregating via RGCN. To avoid the omission of knowledge triples in long texts, we conduct an LLM-based reverse verification by reconstructing the knowledge triples. Experiments show that our model enhances hallucination detection and excels all baselines.

Yuchen Fan, Yuzhong Hong, Qiushi Wang, Junwei Bao, Hongfei Jiang, Yang Song

Alignment, endowing a pre-trained Large language model (LLM) with the ability to follow instructions, is crucial for its real-world applications. Conventional supervised fine-tuning (SFT) methods formalize it as causal language modeling typically with a cross-entropy objective, requiring a large amount of high-quality instruction-response pairs. However, the quality of widely used SFT datasets can not be guaranteed due to the high cost and intensive labor for the creation and maintenance in practice. To overcome the limitations associated with the quality of SFT datasets, we introduce a novel preference-oriented supervised fine-tuning approach, namely PoFT. The intuition is to boost SFT by imposing a particular preference: favoring the target model over aligned LLMs on the same SFT data. This preference encourages the target model to predict a higher likelihood than that predicted by the aligned LLMs, incorporating assessment information on data quality (i.e., predicted likelihood by the aligned LLMs) into the training process. Extensive experiments are conducted, and the results validate the effectiveness of the proposed method. PoFT achieves stable and consistent improvements over the SFT baselines across different training datasets and base models. Moreover, we prove that PoFT can be integrated with existing SFT data filtering methods to achieve better performance, and further improved by following preference optimization procedures, such as DPO.

Cunhang Fan, Enrui Liu, Andong Li, Jianhua Tao, Jian Zhou, Jiahao Li, Chengshi Zheng, Zhao Lv

Although the complex spectrum-based speech enhancement (SE) methods have achieved significant performance, coupling amplitude and phase can lead to a compensation effect, where amplitude information is sacrificed to compensate for the phase that is harmful to SE. In addition, to further improve the performance of SE, many modules are stacked onto SE, resulting in increased model complexity that limits the application of SE. To address these problems, we proposed a dual-path network based on compressed frequency using Mamba. First, we extract amplitude and phase information through parallel dual branches. This approach leverages structured complex spectra to implicitly capture phase information and solves the compensation effect by decoupling amplitude and phase, and the network incorporates an interaction module to suppress unnecessary parts and recover missing components from the other branch. Second, to reduce network complexity, the network introduces a band-split strategy to compress the frequency dimension. To further reduce complexity while maintaining good performance, we designed a Mamba-based module that models the time and frequency dimensions under linear complexity. Finally, compared to baselines, our model achieves an average 8.3 times reduction in computational complexity while maintaining superior performance. Furthermore, it achieves a 25 times reduction in complexity compared to transformer-based models.

Zhichao Duan, Tengyu Pan, Zhenyu Li, Xiuxing Li, Jianyong Wang

Document-level relation extraction (DocRE) is the process of identifying and extracting relations between entities that span multiple sentences within a document. Due to its realistic settings, DocRE has garnered increasing research attention in recent years. Previous research has mostly focused on developing sophisticated encoding models to better capture the intricate patterns between entity pairs. While these advancements are undoubtedly crucial, an even more foundational challenge lies in the data itself. The complexity inherent in DocRE makes the labeling process prone to errors, compounded by the extreme sparsity of positive relation samples, which is driven by both the limited availability of positive instances and the broad diversity of positive relation types. These factors can lead to biased optimization processes, further complicating the task of accurate relation extraction. Recognizing these challenges, we have developed a robust framework called COMM to better solve DocRE. COMM operates by initially employing an instance-aware reasoning method to dynamically capture pertinent information of entity pairs within the document and extract relational features. Following this, COMM takes into account the distribution of relations and the difficulty of samples to dynamically adjust the margins between prediction logits and the decision threshold, a process we call Concentrated Margin Maximization. In this way, COMM not only enhances the extraction of relevant relational features but also boosts DocRE performance by addressing the specific challenges posed by the data. Extensive experiments and analysis demonstrate the versatility and effectiveness of COMM, especially its robustness when trained on low-quality data (achieves >10% performance gains).

Feiyu Duan, Xuemiao Zhang, Sirui Wang, Haoran Que, Yuqi Liu, Wenge Rong, Xunliang Cai

The performance of Large Language Models (LLMs) is intrinsically linked to the quality of its training data. Although several studies have proposed methods for high-quality data selection, they do not consider the importance of knowledge richness in text corpora. In this paper, we propose a novel and gradient-free High-Knowledge Scorer (HKS) to select high-quality data from the dimension of knowledge, to alleviate the problem of knowledge scarcity in the pre-trained corpus. We propose a comprehensive multi-domain knowledge element pool and introduce knowledge density and coverage as metrics to assess the knowledge content of the text. Based on this, we propose a comprehensive knowledge scorer to select data with intensive knowledge, which can also be utilized for domain-specific high-knowledge data selection by restricting knowledge elements to the specific domain. We train models on a high-knowledge bilingual dataset, and experimental results demonstrate that our scorer improves the model's performance in knowledge-intensive and general comprehension tasks, and is effective in enhancing both the generic and domain-specific capabilities of the model.

Yanrui Du, Sendong Zhao, Ming Ma, Yuhan Chen, Bing Qin

Large Language Models (LLMs) have gained significant attention for their exceptional performance across various domains. Despite their advancements, concerns persist regarding their implicit bias, which often leads to negative social impacts. Therefore, it is essential to identify the implicit bias in LLMs and investigate the potential threat posed by it. Our study focused on a specific type of implicit bias, termed the ''Yes-No'' implicit bias, which refers to LLMs' inherent tendency to favor ''Yes'' or ''No'' responses to a single instruction. By comparing the probability of LLMs generating a series of ''Yes'' versus ''No'' responses, we observed different inherent response tendencies exhibited by LLMs when faced with different instructions. To further investigate the impact of such bias, we developed an attack method called Implicit Bias In-Context Manipulation, attempting to manipulate LLMs' behavior. Specifically, we explored whether the ''Yes'' implicit bias could manipulate ''No'' responses into ''Yes'' in LLMs' responses to malicious instructions, leading to harmful outputs. Our findings revealed that the ''Yes'' implicit bias brings a significant security threat, comparable to that of carefully designed attack methods. Moreover, we offered a comprehensive analysis from multiple perspectives to deepen the understanding of this security threat, emphasizing the need for ongoing improvement in LLMs' security.

Xiaohu Du, Fan Mo, Ming Wen, Tu Gu, Huadi Zheng, Hai Jin, Jie Shi

Large Language Models (LLMs) have achieved significant performance in various natural language processing tasks but also pose safety and ethical threats, thus requiring red teaming and alignment processes to bolster their safety. To effectively exploit these aligned LLMs, recent studies have introduced jailbreak attacks based on multi-turn dialogues. These attacks aim to prompt LLMs to generate harmful or biased content by guiding them through contextual content. However, the underlying reasons for the effectiveness of multi-turn jailbreaks remain unclear. Existing attacks often focus on optimizing queries and escalating toxicity to construct dialogues, lacking a thorough analysis of the inherent vulnerabilities of LLMs. In this paper, we first conduct an in-depth analysis of the differences between single-turn and multi-turn jailbreaks and find that successful multi-turn jailbreaks can effectively disperse the attention of LLMs on keywords associated with harmful behaviors, especially in historical responses. Based on this, we propose ASJA, a new multi-turn jailbreak approach by shifting the attention of LLMs, specifically by iteratively fabricating the dialogue history through a genetic algorithm to induce LLMs to generate harmful content. Extensive experiments on three LLMs and two datasets show that our approach surpasses existing approaches in jailbreak effectiveness, the stealth of jailbreak prompts, and attack efficiency. Our work emphasizes the importance of enhancing the robustness of LLMs' attention mechanism in multi-turn dialogue scenarios for a better defense strategy.

Shihan Dou, Yan Liu, Enyu Zhou, Songyang Gao, Tianlong Li, Limao Xiong, Xin Zhao, Haoxiang Jia, Junjie Ye, Rui Zheng 等

The capability of the reward model (RM) is crucial for the success of Reinforcement Learning from Human Feedback (RLHF) in aligning with human preferences. However, as training progresses, the output space distribution of the policy model shifts. The RM, initially trained on responses sampled from the output distribution of the early policy model, gradually loses its ability to distinguish between responses from the newly shifted distribution. This issue is further compounded when the RM, trained on a specific data distribution, struggles to generalize to examples outside of that distribution. These two issues can be united as a challenge posed by the shifted distribution of the environment. To surmount this challenge, we introduce MetaRM, a novel method leveraging meta-learning to adapt the RM to the shifted environment distribution. MetaRM optimizes the RM in an alternating way, by preserving both the preferences of the original preference pairs, as well as maximizing discrimination power over new examples of the shifted distribution. Extensive experiments demonstrate that MetaRM can iteratively enhance the performance of human preference alignment by improving the RM's capacity to identify subtle differences in samples of shifted distributions.

Guanting Dong, Xiaoshuai Song, Yutao Zhu, Runqi Qiao, Zhicheng Dou, Ji-Rong Wen

Following natural instructions is crucial for the effective application of Retrieval-Augmented Generation (RAG) systems. Despite recent advancements in Large Language Models (LLMs), research on assessing and improving instruction-following (IF) alignment within the RAG domain remains limited. To address this issue, we propose VIF-RAG, an automated, scalable, and verifiable synthetic pipeline for instruction-following alignment in RAG systems. We start by manually crafting a minimal set of atomic instructions (100k) through automated processes. To further bridge the gap in instruction-following auto-evaluation for RAG systems, we introduce FollowRAG Benchmark, which includes approximately 3K test samples, covering 22 categories of general instruction constraints and four knowledge-intensive QA datasets. Due to its robust pipeline design, FollowRAG can seamlessly integrate with different RAG benchmarks. Using FollowRAG and eight widely-used IF and foundational abilities benchmarks for LLMs, we demonstrate that VIF-RAG markedly enhances LLM performance across a broad range of general instruction constraints while effectively leveraging its capabilities in RAG scenarios. Further analysis offers practical insights for achieving IF alignment in RAG systems.

Yifan Ding, Matthew Facciani, Ellen Joyce, Amrit Poudel, Sanmitra Bhattacharya, Balaji Veeramani, Sal Aguinaga, Tim Weninger

Question answering systems are rapidly advancing, but their opaque nature may impact user trust. We explored trust through an anti-monitoring framework, where trust is predicted to be correlated with presence of citations and inversely related to checking citations. We tested this hypothesis with a live question-answering experiment that presented text responses generated using a commercial Chatbot along with varying citations (zero, one, or five), both relevant and random, and recorded if participants checked the citations and their self-reported trust in the generated responses. We found a significant increase in trust when citations were present, a result that held true even when the citations were random; we also found a significant decrease in trust when participants checked the citations. These results highlight the importance of citations in enhancing trust in AI-generated content.

Muxi Diao, Rumei Li, Shiyang Liu, Guogang Liao, Jingang Wang, Xunliang Cai, Weiran Xu

As Large Language Models (LLMs) continue to advance in capability and influence, ensuring their security and preventing harmful outputs has become crucial. A promising approach to address these concerns involves training models to automatically generate adversarial prompts for red teaming. However, the evolving subtlety of vulnerabilities in LLMs challenges the effectiveness of current adversarial methods, which struggle to generate diverse, complex prompts and dynamically explore the weaknesses of these models. To tackle these challenges, we introduce the Self-Evolving Adversarial Safety (SEAS) optimization framework, which includes both a SEAS dataset and a SEAS pipeline. The SEAS dataset comprises complex adversarial prompts, while the SEAS pipeline operates through three stages: Initialization, Attack, and Adversarial Optimization. This framework generates a diverse range of adversarial prompts and dynamically explores the model's vulnerabilities to enhance its security. Our contributions include a novel adversarial framework, a comprehensive safety dataset, and empirical evidence demonstrating the effectiveness of SEAS.

Soham Deshmukh, Shuo Han, Hazim Bukhari, Benjamin Elizalde, Hannes Gamper, Rita Singh, Bhiksha Raj

Recent literature uses language to build foundation models for audio. These Audio-Language Models (ALMs) are trained on a vast number of audio-text pairs and show remarkable performance in tasks including Text-to-Audio Retrieval, Captioning, and Question Answering. However, their ability to engage in more complex open-ended tasks, like Interactive Question-Answering, requires proficiency in logical reasoning- a skill not yet benchmarked. We introduce the novel task of Audio Entailment to evaluate an ALM's deductive reasoning ability. This task assesses whether a text description (hypothesis) of audio content can be deduced from an audio recording (premise), with potential conclusions being entailment, neutral, or contradiction, depending on the sufficiency of the evidence. We create two datasets for this task with audio recordings sourced from two audio captioning datasets-AudioCaps and Clotho-and hypotheses generated using Large Language Models (LLMs). We benchmark state-of-the-art ALMs and find deficiencies in logical reasoning with both zero-shot and linear probe evaluations. Finally, we propose "caption-before-reason", an intermediate step of captioning that improves the Zero-Shot and linear-probe performance of ALMs by an absolute 6% and 3%, respectively.

Boyi Deng, Wenjie Wang, Fengbin Zhu, Qifan Wang, Fuli Feng

Retrieval-Augmented Generation (RAG) can alleviate hallucinations of Large Language Models (LLMs) by referencing external documents. However, the misinformation in external documents may mislead LLMs' generation. To address this issue, we explore the task of "credibility-aware RAG", in which LLMs automatically adjust the influence of retrieved documents based on their credibility scores to counteract misinformation. To this end, we introduce a plug-and-play method named Credibility-aware Attention Modification (CrAM). CrAM identifies influential attention heads in LLMs and adjusts their attention weights based on the credibility of the documents, thereby reducing the impact of low-credibility documents. Experiments on Natual Questions and TriviaQA using Llama2-13B, Llama3-8B, and Qwen1.5-7B show that CrAM improves the RAG performance of LLMs against misinformation pollution by over 20%, even surpassing supervised fine-tuning methods.

João Pedro Gandarela De Souza, Danilo Carvalho, André Freitas

This work presents a novel systematic methodology to analyse the capabilities and limitations of Large Language Models (LLMs) with feedback from a formal inference engine, on logic theory induction. The analysis is complexity-graded w.r.t. rule dependency structure, allowing quantification of specific inference challenges on LLM performance. Integrating LLMs with formal methods is a promising frontier in the Natural Language Processing field, as an important avenue for improving model inference control and explainability. In particular, inductive learning over complex sets of facts and rules, poses unique challenges for current autoregressive models, as they lack explicit symbolic grounding. While they can be complemented by formal systems, the properties delivered by LLMs regarding inductive learning, are not well understood and quantified. Empirical results indicate that the largest LLMs can achieve competitive results against a SOTA Inductive Logic Programming (ILP) system baseline, but also that tracking long predicate relationship chains is a more difficult obstacle than theory complexity for LLMs.

Sourish Dasgupta, Harsh Sharma, Devansh Patel, Prarthee Desai, Anil K. Roy

Reviewer Matchmaking (RM) is a pivotal process in academic publishing that aligns manuscripts with appropriate reviewers based on their expertise and prior publications. The demand for an automated RM system has escalated with the significant surge in submissions over the past decade. State-of-the-art (SOTA) RM models are document-representation-based (DR-RM) and match the manuscript and reviewer's past publication using a similarity method defined on a high-dimensional vector space. However, they are far from accurate despite their large-scale usage. In this paper, we establish that conventional RM evaluation measures are unreliable and instead emphasize that standard correlation measures are adequate. For the first time, we compare the performance of six SOTA DR-RM models with those of fourteen SOTA Key-phrase Extraction-based RM (KPE-RM) models - an alternate unexplored approach. We observe that KPE-RM models show comparable results in many cases, with the new best model being PatternRank-RM - a KPE-RM model beating the best DR-RM model SPECTER2-RM (Pearson: 0.004+, Spearman: 0.006+, Kendall: 0.043+). We conclude that KPE-RM models must be contextualized to the RM task and cannot be used as plug-n-play.