arXiv subjects

Suzan Verberne

Publications and source records attributed to Suzan Verberne.

At least 55 records · Page 3Linked to original sources

arXiv2025

Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers

Large language models (LLMs) have been widely integrated into information retrieval to advance traditional techniques. However, effectively enabling LLMs to seek accurate knowledge in complex tasks remains a challenge due to the complexity of multi-hop queries as well as the irrelevant retrieved content. To address these limitations, we propose EXSEARCH, an agentic search framework, where the LLM learns to retrieve useful information as the reasoning unfolds through a self-incentivized process. At each step, the LLM decides what to retrieve (thinking), triggers an external retriever (search), and extracts fine-grained evidence (recording) to support next-step reasoning. To enable LLM with this capability, EXSEARCH adopts a Generalized Expectation-Maximization algorithm. In the E-step, the LLM generates multiple search trajectories and assigns an importance weight to each; the M-step trains the LLM on them with a re-weighted loss function. This creates a self-incentivized loop, where the LLM iteratively learns from its own generated data, progressively improving itself for search. We further theoretically analyze this training process, establishing convergence guarantees. Extensive experiments on four knowledge-intensive benchmarks show that EXSEARCH substantially outperforms baselines, e.g., +7.8% improvement on exact match score. Motivated by these promising results, we introduce EXSEARCH-Zoo, an extension that extends our method to broader scenarios, to facilitate future work.

cs.CL↗

arXiv2025

Improving RAG for Personalization with Author Features and Contrastive Examples

Personalization with retrieval-augmented generation (RAG) often fails to capture fine-grained features of authors, making it hard to identify their unique traits. To enrich the RAG context, we propose providing Large Language Models (LLMs) with author-specific features, such as average sentiment polarity and frequently used words, in addition to past samples from the author's profile. We introduce a new feature called Contrastive Examples: documents from other authors are retrieved to help LLM identify what makes an author's style unique in comparison to others. Our experiments show that adding a couple of sentences about the named entities, dependency patterns, and words a person uses frequently significantly improves personalized text generation. Combining features with contrastive examples boosts the performance further, achieving a relative 15% improvement over baseline RAG while outperforming the benchmarks. Our results show the value of fine-grained features for better personalization, while opening a new research dimension for including contrastive examples as a complement with RAG. We release our code publicly.

cs.IR↗

arXiv2025

Tool Learning in the Wild: Empowering Language Models as Automatic Tool Agents

Augmenting large language models (LLMs) with external tools has emerged as a promising approach to extend their utility, enabling them to solve practical tasks. Previous methods manually parse tool documentation and create in-context demonstrations, transforming tools into structured formats for LLMs to use in their step-by-step reasoning. However, this manual process requires domain expertise and struggles to scale to large toolsets. Additionally, these methods rely heavily on ad-hoc inference techniques or special tokens to integrate free-form LLM generation with tool-calling actions, limiting the LLM's flexibility in handling diverse tool specifications and integrating multiple tools. In this work, we propose AutoTools, a framework that enables LLMs to automate the tool-use workflow. Specifically, the LLM automatically transforms tool documentation into callable functions, verifying syntax and runtime correctness. Then, the LLM integrates these functions into executable programs to solve practical tasks, flexibly grounding tool-use actions into its reasoning processes. Extensive experiments on existing and newly collected, more challenging benchmarks illustrate the superiority of our framework. Inspired by these promising results, we further investigate how to improve the expertise of LLMs, especially open-source LLMs with fewer parameters, within AutoTools. Thus, we propose the AutoTools-learning approach, training the LLMs with three learning tasks on 34k instances of high-quality synthetic data, including documentation understanding, relevance learning, and function programming. Fine-grained results validate the effectiveness of our overall training approach and each individual task. Our methods are an important step towards the use of LLMs for solving real-world tasks with external tools.

cs.CL↗

arXiv2025

Self-seeding and Multi-intent Self-instructing LLMs for Generating Intent-aware Information-Seeking dialogs

Identifying user intents in information-seeking dialogs is crucial for a system to meet user's information needs. Intent prediction (IP) is challenging and demands sufficient dialogs with human-labeled intents for training. However, manually annotating intents is resource-intensive. While large language models (LLMs) have been shown to be effective in generating synthetic data, there is no study on using LLMs to generate intent-aware information-seeking dialogs. In this paper, we focus on leveraging LLMs for zero-shot generation of large-scale, open-domain, and intent-aware information-seeking dialogs. We propose SOLID, which has novel self-seeding and multi-intent self-instructing schemes. The former improves the generation quality by using the LLM's own knowledge scope to initiate dialog generation; the latter prompts the LLM to generate utterances sequentially, and mitigates the need for manual prompt design by asking the LLM to autonomously adapt its prompt instruction when generating complex multi-intent utterances. Furthermore, we propose SOLID-RL, which is further trained to generate a dialog in one step on the data generated by SOLID. We propose a length-based quality estimation mechanism to assign varying weights to SOLID-generated dialogs based on their quality during the training process of SOLID-RL. We use SOLID and SOLID-RL to generate more than 300k intent-aware dialogs, surpassing the size of existing datasets. Experiments show that IP methods trained on dialogs generated by SOLID and SOLID-RL achieve better IP quality than ones trained on human-generated dialogs.

cs.CL↗

arXiv2024

Generate then Refine: Data Augmentation for Zero-shot Intent Detection

In this short paper we propose a data augmentation method for intent detection in zero-resource domains. Existing data augmentation methods rely on few labelled examples for each intent category, which can be expensive in settings with many possible intents. We use a two-stage approach: First, we generate utterances for intent labels using an open-source large language model in a zero-shot setting. Second, we develop a smaller sequence-to-sequence model (the Refiner), to improve the generated utterances. The Refiner is fine-tuned on seen domains and then applied to unseen domains. We evaluate our method by training an intent classifier on the generated data, and evaluating it on real (human) data. We find that the Refiner significantly improves the data utility and diversity over the zero-shot LLM baseline for unseen domains and over common baseline approaches. Our results indicate that a two-step approach of a generative LLM in zero-shot setting and a smaller sequence-to-sequence model can provide high-quality data for intent detection.

cs.CL↗

arXiv2024

What are the limits of cross-lingual dense passage retrieval for low-resource languages?

In this paper, we analyze the capabilities of the multi-lingual Dense Passage Retriever (mDPR) for extremely low-resource languages. In the Cross-lingual Open-Retrieval Answer Generation (CORA) pipeline, mDPR achieves success on multilingual open QA benchmarks across 26 languages, of which 9 were unseen during training. These results are promising for Question Answering (QA) for low-resource languages. We focus on two extremely low-resource languages for which mDPR performs poorly: Amharic and Khmer. We collect and curate datasets to train mDPR models using Translation Language Modeling (TLM) and question--passage alignment. We also investigate the effect of our extension on the language distribution in the retrieval results. Our results on the MKQA and AmQA datasets show that language alignment brings improvements to mDPR for the low-resource languages, but the improvements are modest and the results remain low. We conclude that fulfilling CORA's promise to enable multilingual open QA in extremely low-resource settings is challenging because the model, the data, and the evaluation approach are intertwined. Hence, all three need attention in follow-up work. We release our code for reproducibility and future work: https://anonymous.4open.science/r/Question-Answering-for-Low-Resource-Languages-B13C/

cs.IR↗

arXiv2024

CAUSE: Counterfactual Assessment of User Satisfaction Estimation in Task-Oriented Dialogue Systems

An important unexplored aspect in previous work on user satisfaction estimation for Task-Oriented Dialogue (TOD) systems is their evaluation in terms of robustness for the identification of user dissatisfaction: current benchmarks for user satisfaction estimation in TOD systems are highly skewed towards dialogues for which the user is satisfied. The effect of having a more balanced set of satisfaction labels on performance is unknown. However, balancing the data with more dissatisfactory dialogue samples requires further data collection and human annotation, which is costly and time-consuming. In this work, we leverage large language models (LLMs) and unlock their ability to generate satisfaction-aware counterfactual dialogues to augment the set of original dialogues of a test collection. We gather human annotations to ensure the reliability of the generated samples. We evaluate two open-source LLMs as user satisfaction estimators on our augmented collection against state-of-the-art fine-tuned models. Our experiments show that when used as few-shot user satisfaction estimators, open-source LLMs show higher robustness to the increase in the number of dissatisfaction labels in the test collection than the fine-tuned state-of-the-art models. Our results shed light on the need for data augmentation approaches for user satisfaction estimation in TOD systems. We release our aligned counterfactual dialogues, which are curated by human annotation, to facilitate further research on this topic.

cs.CL↗

arXiv2024

The Impact of Quantization on Retrieval-Augmented Generation: An Analysis of Small LLMs

Post-training quantization reduces the computational demand of Large Language Models (LLMs) but can weaken some of their capabilities. Since LLM abilities emerge with scale, smaller LLMs are more sensitive to quantization. In this paper, we explore how quantization affects smaller LLMs' ability to perform retrieval-augmented generation (RAG), specifically in longer contexts. We chose personalization for evaluation because it is a challenging domain to perform using RAG as it requires long-context reasoning over multiple documents. We compare the original FP16 and the quantized INT4 performance of multiple 7B and 8B LLMs on two tasks while progressively increasing the number of retrieved documents to test how quantized models fare against longer contexts. To better understand the effect of retrieval, we evaluate three retrieval models in our experiments. Our findings reveal that if a 7B LLM performs the task well, quantization does not impair its performance and long-context reasoning capabilities. We conclude that it is possible to utilize RAG with quantized smaller LLMs.

cs.CL↗

arXiv2024

Efficient course recommendations with T5-based ranking and summarization

In this paper, we implement and evaluate a two-stage retrieval pipeline for a course recommender system that ranks courses for skill-occupation pairs. The in-production recommender system BrightFit provides course recommendations from multiple sources. Some of the course descriptions are long and noisy, while retrieval and ranking in an online system have to be highly efficient. We developed a two-step retrieval pipeline with RankT5 finetuned on MSMARCO as re-ranker. We compare two summarizers for course descriptions: a LongT5 model that we finetuned for the task, and a generative LLM (Vicuna) with in-context learning. We experiment with quantization to reduce the size of the ranking model and increase inference speed. We evaluate our rankers on two newly labelled datasets, with an A/B test, and with a user questionnaire. On the two labelled datasets, our proposed two-stage ranking with automatic summarization achieves a substantial improvement over the in-production (BM25) ranker: nDCG@10 scores improve from 0.482 to 0.684 and from 0.447 to 0.844 on the two datasets. We also achieve a 40% speed-up by using a quantized version of RankT5. The improved quality of the ranking was confirmed by the questionnaire completed by 29 respondents, but not by the A/B test. In the A/B test, a higher clickthrough rate was observed for the BM25-ranking than for the proposed two-stage retrieval. We conclude that T5-based re-ranking and summarization for online course recommendation can obtain much better effectiveness than single-step lexical retrieval, and that quantization has a large effect on RankT5. In the online evaluation, however, other factors than relevance play a role (such as speed and interpretability of the retrieval results), as well as individual preferences.

cs.IR↗

arXiv2024

Learning to Use Tools via Cooperative and Interactive Agents

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).

cs.CL↗

arXiv2024

Which topics are best represented by science maps? An analysis of clustering effectiveness for citation and text similarity networks

A science map of topics is a visualization that shows topics identified algorithmically based on the bibliographic metadata of scientific publications. In practice not all topics are well represented in a science map. We analyzed how effectively different topics are represented in science maps created by clustering biomedical publications. To achieve this, we investigated which topic categories, obtained from MeSH terms, are better represented in science maps based on citation or text similarity networks. To evaluate the clustering effectiveness of topics, we determined the extent to which documents belonging to the same topic are grouped together in the same cluster. We found that the best and worst represented topic categories are the same for citation and text similarity networks. The best represented topic categories are diseases, psychology, anatomy, organisms and the techniques and equipment used for diagnostics and therapy, while the worst represented topic categories are natural science fields, geographical entities, information sciences and health care and occupations. Furthermore, for the diseases and organisms topic categories and for science maps with smaller clusters, we found that topics tend to be better represented in citation similarity networks than in text similarity networks.

cs.DL↗

arXiv2024

Biomedical Entity Linking for Dutch: Fine-tuning a Self-alignment BERT Model on an Automatically Generated Wikipedia Corpus

Biomedical entity linking, a main component in automatic information extraction from health-related texts, plays a pivotal role in connecting textual entities (such as diseases, drugs and body parts mentioned by patients) to their corresponding concepts in a structured biomedical knowledge base. The task remains challenging despite recent developments in natural language processing. This paper presents the first evaluated biomedical entity linking model for the Dutch language. We use MedRoBERTa.nl as base model and perform second-phase pretraining through self-alignment on a Dutch biomedical ontology extracted from the UMLS and Dutch SNOMED. We derive a corpus from Wikipedia of ontology-linked Dutch biomedical entities in context and fine-tune our model on this dataset. We evaluate our model on the Dutch portion of the Mantra GSC-corpus and achieve 54.7% classification accuracy and 69.8% 1-distance accuracy. We then perform a case study on a collection of unlabeled, patient-support forum data and show that our model is hampered by the limited quality of the preceding entity recognition step. Manual evaluation of small sample indicates that of the correctly extracted entities, around 65% is linked to the correct concept in the ontology. Our results indicate that biomedical entity linking in a language other than English remains challenging, but our Dutch model can be used to for high-level analysis of patient-generated text.

cs.CL↗

arXiv2024

Investigating the Robustness of Modelling Decisions for Few-Shot Cross-Topic Stance Detection: A Preregistered Study

For a viewpoint-diverse news recommender, identifying whether two news articles express the same viewpoint is essential. One way to determine "same or different" viewpoint is stance detection. In this paper, we investigate the robustness of operationalization choices for few-shot stance detection, with special attention to modelling stance across different topics. Our experiments test pre-registered hypotheses on stance detection. Specifically, we compare two stance task definitions (Pro/Con versus Same Side Stance), two LLM architectures (bi-encoding versus cross-encoding), and adding Natural Language Inference knowledge, with pre-trained RoBERTa models trained with shots of 100 examples from 7 different stance detection datasets. Some of our hypotheses and claims from earlier work can be confirmed, while others give more inconsistent results. The effect of the Same Side Stance definition on performance differs per dataset and is influenced by other modelling choices. We found no relationship between the number of training topics in the training shots and performance. In general, cross-encoding out-performs bi-encoding, and adding NLI training to our models gives considerable improvement, but these results are not consistent across all datasets. Our results indicate that it is essential to include multiple datasets and systematic modelling experiments when aiming to find robust modelling choices for the concept `stance'.

cs.CL↗

arXiv2024

High Recall, Small Data: The Challenges of Within-System Evaluation in a Live Legal Search System

This paper illustrates some challenges of common ranking evaluation methods for legal information retrieval (IR). We show these challenges with log data from a live legal search system and two user studies. We provide an overview of aspects of legal IR, and the implications of these aspects for the expected challenges of common evaluation methods: test collections based on explicit and implicit feedback, user surveys, and A/B testing. Next, we illustrate the challenges of common evaluation methods using data from a live, commercial, legal search engine. We specifically focus on methods for monitoring the effectiveness of (continuous) changes to document ranking by a single IR system over time. We show how the combination of characteristics in legal IR systems and limited user data can lead to challenges that cause the common evaluation methods discussed to be sub-optimal. In our future work we will therefore focus on less common evaluation methods, such as cost-based evaluation models.

cs.IR↗

arXiv2024

Measuring Bias in a Ranked List using Term-based Representations

In most recent studies, gender bias in document ranking is evaluated with the NFaiRR metric, which measures bias in a ranked list based on an aggregation over the unbiasedness scores of each ranked document. This perspective in measuring the bias of a ranked list has a key limitation: individual documents of a ranked list might be biased while the ranked list as a whole balances the groups' representations. To address this issue, we propose a novel metric called TExFAIR (term exposure-based fairness), which is based on two new extensions to a generic fairness evaluation framework, attention-weighted ranking fairness (AWRF). TExFAIR assesses fairness based on the term-based representation of groups in a ranked list: (i) an explicit definition of associating documents to groups based on probabilistic term-level associations, and (ii) a rank-biased discounting factor (RBDF) for counting non-representative documents towards the measurement of the fairness of a ranked list. We assess TExFAIR on the task of measuring gender bias in passage ranking, and study the relationship between TExFAIR and NFaiRR. Our experiments show that there is no strong correlation between TExFAIR and NFaiRR, which indicates that TExFAIR measures a different dimension of fairness than NFaiRR. With TExFAIR, we extend the AWRF framework to allow for the evaluation of fairness in settings with term-based representations of groups in documents in a ranked list.

cs.CL↗

arXiv2024

A Multi-Agent Conversational Recommender System

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.

cs.IR↗

arXiv2024

Answer Retrieval in Legal Community Question Answering

The task of answer retrieval in the legal domain aims to help users to seek relevant legal advice from massive amounts of professional responses. Two main challenges hinder applying existing answer retrieval approaches in other domains to the legal domain: (1) a huge knowledge gap between lawyers and non-professionals; and (2) a mix of informal and formal content on legal QA websites. To tackle these challenges, we propose CE_FS, a novel cross-encoder (CE) re-ranker based on the fine-grained structured inputs. CE_FS uses additional structured information in the CQA data to improve the effectiveness of cross-encoder re-rankers. Furthermore, we propose LegalQA: a real-world benchmark dataset for evaluating answer retrieval in the legal domain. Experiments conducted on LegalQA show that our proposed method significantly outperforms strong cross-encoder re-rankers fine-tuned on MS MARCO. Our novel finding is that adding the question tags of each question besides the question description and title into the input of cross-encoder re-rankers structurally boosts the rankers' effectiveness. While we study our proposed method in the legal domain, we believe that our method can be applied in similar applications in other domains.

cs.IR↗

arXiv2023

ChatGPT as a commenter to the news: can LLMs generate human-like opinions?

ChatGPT, GPT-3.5, and other large language models (LLMs) have drawn significant attention since their release, and the abilities of these models have been investigated for a wide variety of tasks. In this research we investigate to what extent GPT-3.5 can generate human-like comments on Dutch news articles. We define human likeness as `not distinguishable from human comments', approximated by the difficulty of automatic classification between human and GPT comments. We analyze human likeness across multiple prompting techniques. In particular, we utilize zero-shot, few-shot and context prompts, for two generated personas. We found that our fine-tuned BERT models can easily distinguish human-written comments from GPT-3.5 generated comments, with none of the used prompting methods performing noticeably better. We further analyzed that human comments consistently showed higher lexical diversity than GPT-generated comments. This indicates that although generative LLMs can generate fluent text, their capability to create human-like opinionated comments is still limited.

cs.CL↗