Emine Yilmaz

Yue Feng, Aldo Lipani, Fanghua Ye et al.

Dialogue State Tracking (DST) aims to keep track of users' intentions during the course of a conversation. In DST, modelling the relations among domains and slots is still an under-studied problem. Existing approaches that have considered such relations generally fall short in: (1) fusing prior slot-domain membership relations and dialogue-aware dynamic slot relations explicitly, and (2) generalizing to unseen domains. To address these issues, we propose a novel \textbf{D}ynamic \textbf{S}chema \textbf{G}raph \textbf{F}usion \textbf{Net}work (\textbf{DSGFNet}), which generates a dynamic schema graph to explicitly fuse the prior slot-domain membership relations and dialogue-aware dynamic slot relations. It also uses the schemata to facilitate knowledge transfer to new domains. DSGFNet consists of a dialogue utterance encoder, a schema graph encoder, a dialogue-aware schema graph evolving network, and a schema graph enhanced dialogue state decoder. Empirical results on benchmark datasets (i.e., SGD, MultiWOZ2.1, and MultiWOZ2.2), show that DSGFNet outperforms existing methods.

Sahan Bulathwela, Meghana Verma, Maria Perez-Ortiz et al.

This work explores how population-based engagement prediction can address cold-start at scale in large learning resource collections. The paper introduces i) VLE, a novel dataset that consists of content and video based features extracted from publicly available scientific video lectures coupled with implicit and explicit signals related to learner engagement, ii) two standard tasks related to predicting and ranking context-agnostic engagement in video lectures with preliminary baselines and iii) a set of experiments that validate the usefulness of the proposed dataset. Our experimental results indicate that the newly proposed VLE dataset leads to building context-agnostic engagement prediction models that are significantly performant than ones based on previous datasets, mainly attributing to the increase of training examples. VLE dataset's suitability in building models towards Computer Science/ Artificial Intelligence education focused on e-learning/ MOOC use-cases is also evidenced. Further experiments in combining the built model with a personalising algorithm show promising improvements in addressing the cold-start problem encountered in educational recommenders. This is the largest and most diverse publicly available dataset to our knowledge that deals with learner engagement prediction tasks. The dataset, helper tools, descriptive statistics and example code snippets are available publicly.

Giorgio Giannone, Serhii Havrylov, Jordan Massiah et al. · mit

Advances in deep learning theory have revealed how average generalization relies on superficial patterns in data. The consequences are brittle models with poor performance with shift in group distribution at test time. When group annotation is available, we can use robust optimization tools to tackle the problem. However, identification and annotation are time-consuming, especially on large datasets. A recent line of work leverages self-supervision and oversampling to improve generalization on minority groups without group annotation. We propose to unify and generalize these approaches using a class-conditional variant of mixup tailored for worst-group generalization. Our approach, Just Mix Once (JM1), interpolates samples during learning, augmenting the training distribution with a continuous mixture of groups. JM1 is domain agnostic and computationally efficient, can be used with any level of group annotation, and performs on par or better than the state-of-the-art on worst-group generalization. Additionally, we provide a simple explanation of why JM1 works.

Rikaz Rameez, Hossein A. Rahmani, Emine Yilmaz

Recently, Twitter has become the social network of choice for sharing and spreading information to a multitude of users through posts called 'tweets'. Users can easily re-share these posts to other users through 'retweets', which allow information to cascade to many more users, increasing its outreach. Clearly, being able to know the extent to which a post can be retweeted has great value in advertising, influencing and other such campaigns. In this paper we propose ViralBERT, which can be used to predict the virality of tweets using content- and user-based features. We employ a method of concatenating numerical features such as hashtags and follower numbers to tweet text, and utilise two BERT modules: one for semantic representation of the combined text and numerical features, and another module purely for sentiment analysis of text, as both the information within text and it's ability to elicit an emotional response play a part in retweet proneness. We collect a dataset of 330k tweets to train ViralBERT and validate the efficacy of our model using baselines from current studies in this field. Our experiments show that our approach outperforms these baselines, with a 13% increase in both F1 Score and Accuracy compared to the best performing baseline method. We then undergo an ablation study to investigate the importance of chosen features, finding that text sentiment and follower counts, and to a lesser extent mentions and following counts, are the strongest features for the model, and that hashtag counts are detrimental to the model.

Bin Wu, Arastun Mammadli, Xiaoyu Zhang et al.

The rapid growth of AI agent ecosystems is transforming how complex tasks are delegated and executed, creating a new challenge of identifying suitable agents for a given task. Unlike traditional tools, agent capabilities are often compositional and execution-dependent, making them difficult to assess from textual descriptions alone. However, existing research and benchmarks typically assume well-specified functionalities, controlled candidate pools, or only executable task queries, leaving realistic agent search scenarios insufficiently studied. We introduce AgentSearchBench, a large-scale benchmark for agent search in the wild, built from nearly 10,000 real-world agents across multiple providers. The benchmark formalizes agent search as retrieval and reranking problems under both executable task queries and high-level task descriptions, and evaluates relevance using execution-grounded performance signals. Experiments reveal a consistent gap between semantic similarity and actual agent performance, exposing the limitations of description-based retrieval and reranking methods. We further show that lightweight behavioral signals, including execution-aware probing, can substantially improve ranking quality, highlighting the importance of incorporating execution signals into agent discovery. Our code is available at https://github.com/Bingo-W/AgentSearchBench.

Yinzhu Chen, Abdine Maiga, Hossein A. Rahmani et al.

Large Language Models (LLMs) are increasingly used for clinical decision support, where hallucinations and unsafe suggestions may pose direct risks to patient safety. These risks are particularly challenging as they often manifest as subtle clinical errors that evade detection by generic metrics, while expert-authored fine-grained rubrics remain costly to construct and difficult to scale. In this paper, we propose a retrieval-augmented multi-agent framework designed to automate the generation of instance-specific evaluation rubrics. Our approach grounds evaluation in authoritative medical evidence by decomposing retrieved content into atomic facts and synthesizing them with user interaction constraints to form verifiable, fine-grained evaluation criteria. Evaluated on HealthBench, our framework achieves a Clinical Intent Alignment (CIA) score of 60.12%, a statistically significant improvement over the GPT-4o baseline (55.16%). In discriminative tests, our rubrics yield a mean score delta ($μ_Δ = 8.658$) and an AUROC of 0.977, nearly doubling the quality separation achieved by GPT-4o baseline (4.972). Beyond evaluation, our rubrics effectively guide response refinement, improving quality by 9.2% (from 59.0% to 68.2%). This provides a scalable and transparent foundation for both evaluating and improving medical LLMs. The code is available at https://anonymous.4open.science/r/Automated-Rubric-Generation-AF3C/.

Fanghua Ye, Meng Fang, Shenghui Li et al.

Query rewriting plays a vital role in enhancing conversational search by transforming context-dependent user queries into standalone forms. Existing approaches primarily leverage human-rewritten queries as labels to train query rewriting models. However, human rewrites may lack sufficient information for optimal retrieval performance. To overcome this limitation, we propose utilizing large language models (LLMs) as query rewriters, enabling the generation of informative query rewrites through well-designed instructions. We define four essential properties for well-formed rewrites and incorporate all of them into the instruction. In addition, we introduce the role of rewrite editors for LLMs when initial query rewrites are available, forming a "rewrite-then-edit" process. Furthermore, we propose distilling the rewriting capabilities of LLMs into smaller models to reduce rewriting latency. Our experimental evaluation on the QReCC dataset demonstrates that informative query rewrites can yield substantially improved retrieval performance compared to human rewrites, especially with sparse retrievers.

Xi Wang, Hossein A. Rahmani, Jiqun Liu et al.

Conversational Recommendation System (CRS) is a rapidly growing research area that has gained significant attention alongside advancements in language modelling techniques. However, the current state of conversational recommendation faces numerous challenges due to its relative novelty and limited existing contributions. In this study, we delve into benchmark datasets for developing CRS models and address potential biases arising from the feedback loop inherent in multi-turn interactions, including selection bias and multiple popularity bias variants. Drawing inspiration from the success of generative data via using language models and data augmentation techniques, we present two novel strategies, 'Once-Aug' and 'PopNudge', to enhance model performance while mitigating biases. Through extensive experiments on ReDial and TG-ReDial benchmark datasets, we show a consistent improvement of CRS techniques with our data augmentation approaches and offer additional insights on addressing multiple newly formulated biases.

Fanghua Ye, Xi Wang, Jie Huang et al.

Existing dialogue datasets contain lots of noise in their state annotations. Such noise can hurt model training and ultimately lead to poor generalization performance. A general framework named ASSIST has recently been proposed to train robust dialogue state tracking (DST) models. It introduces an auxiliary model to generate pseudo labels for the noisy training set. These pseudo labels are combined with vanilla labels by a common fixed weighting parameter to train the primary DST model. Notwithstanding the improvements of ASSIST on DST, tuning the weighting parameter is challenging. Moreover, a single parameter shared by all slots and all instances may be suboptimal. To overcome these limitations, we propose a meta learning-based framework MetaASSIST to adaptively learn the weighting parameter. Specifically, we propose three schemes with varying degrees of flexibility, ranging from slot-wise to both slot-wise and instance-wise, to convert the weighting parameter into learnable functions. These functions are trained in a meta-learning manner by taking the validation set as meta data. Experimental results demonstrate that all three schemes can achieve competitive performance. Most impressively, we achieve a state-of-the-art joint goal accuracy of 80.10% on MultiWOZ 2.4.

Fanghua Ye, Mingming Yang, Jianhui Pang et al.

The proliferation of open-source Large Language Models (LLMs) from various institutions has highlighted the urgent need for comprehensive evaluation methods. However, current evaluation platforms, such as the widely recognized HuggingFace open LLM leaderboard, neglect a crucial aspect -- uncertainty, which is vital for thoroughly assessing LLMs. To bridge this gap, we introduce a new benchmarking approach for LLMs that integrates uncertainty quantification. Our examination involves nine LLMs (LLM series) spanning five representative natural language processing tasks. Our findings reveal that: I) LLMs with higher accuracy may exhibit lower certainty; II) Larger-scale LLMs may display greater uncertainty compared to their smaller counterparts; and III) Instruction-finetuning tends to increase the uncertainty of LLMs. These results underscore the significance of incorporating uncertainty in the evaluation of LLMs.

Xi Wang, Procheta Sen, Ruizhe Li et al.

Despite the success of integrating large language models into the development of conversational systems, many studies have shown the effectiveness of retrieving and augmenting external knowledge for informative responses. Hence, many existing studies commonly assume the always need for Retrieval Augmented Generation (RAG) in a conversational system without explicit control. This raises a research question about such a necessity. In this study, we propose to investigate the need for each turn of system response to be augmented with external knowledge. In particular, by leveraging human judgements on the binary choice of adaptive augmentation, we develop RAGate, a gating model, which models conversation context and relevant inputs to predict if a conversational system requires RAG for improved responses. We conduct extensive experiments on devising and applying RAGate to conversational models and well-rounded analyses of different conversational scenarios. Our experimental results and analysis indicate the effective application of RAGate in RAG-based conversational systems in identifying system responses for appropriate RAG with high-quality responses and a high generation confidence. This study also identifies the correlation between the generation's confidence level and the relevance of the augmented knowledge.

Jiazheng Li, Emine Yilmaz, Bei Chen et al.

Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterized by long interaction traces, inter-agent dependencies, and delayed error manifestation. Existing diagnostic approaches often rely on expensive expert annotation or ''LLM-as-a-judge'' paradigms, which struggle to pinpoint decisive error steps within extended contexts. In this paper, we introduce ErrorProbe, a self-improving framework for semantic failure attribution that identifies responsible agents and the originating error step. The framework operates via a three-stage pipeline: (1) operationalizing the MAS failure taxonomy to detect local anomalies, (2) performing symptom-driven backward tracing to prune irrelevant context, and (3) employing a specialized multi-agent team (Strategist, Investigator, Arbiter) to validate error hypotheses through tool-grounded execution. Crucially, ErrorProbe maintains a verified episodic memory that updates only when error patterns are confirmed by executable evidence, without the need for annotation. Experiments across the TracerTraj and Who&When benchmarks demonstrate that ErrorProbe significantly outperforms baselines, particularly in step-level localization, while the verified memory enables robust cross-domain transfer without retraining.

Hamze Muse, Sahan Bulathwela, Emine Yilmaz

With the boom of digital educational materials and scalable e-learning systems, the potential for realising AI-assisted personalised learning has skyrocketed. In this landscape, the automatic generation of educational questions will play a key role, enabling scalable self-assessment when a global population is manoeuvring their personalised learning journeys. We develop EduQG, a novel educational question generation model built by adapting a large language model. Our initial experiments demonstrate that EduQG can produce superior educational questions by pre-training on scientific text.

Hossein A. Rahmani, Satyapriya Krishna, Xi Wang et al.

Large language models have recently demonstrated remarkable abilities to self-correct their responses through iterative refinement, often referred to as self-consistency or self-reflection. However, the dynamics of this self-correction mechanism may differ substantially depending on whether the model is tasked with open-ended text generation or with selecting the most appropriate response from multiple predefined options. In this paper, we conduct a systematic investigation of these two paradigms by comparing performance trends and error-correction behaviors across various natural language understanding and reasoning tasks, covering language models of different scales and families. Our experimental results reveal distinct patterns of improvement and failure modes: \textit{While open-ended generation often benefits from the flexibility of re-interpretation and compositional refinement, multiple-choice selection can leverage clearer solution boundaries but may be limited by the provided options}. This contrast also reflects the dual demands faced by emerging agentic LLM applications: effective agents must not only generate and refine open-ended plans or explanations, but also make reliable discrete choices when operating within constrained action spaces. Our findings, therefore, highlight that the design of self-correction mechanisms should take into account the interaction between task structure and output space, with implications for both knowledge-intensive reasoning and decision-oriented applications of LLMs.

Shuofei Qiao, Yunxiang Wei, Xuehai Wang et al.

The rapid evolution of Large Language Models has catalyzed a surge in scientific idea production, yet this leap has not been accompanied by a matching advance in idea evaluation. The fundamental nature of scientific evaluation needs knowledgeable grounding, collective deliberation, and multi-criteria decision-making. However, existing idea evaluation methods often suffer from narrow knowledge horizons, flattened evaluation dimensions, and the inherent bias in LLM-as-a-Judge. To address these, we regard idea evaluation as a knowledge-grounded, multi-perspective reasoning problem and introduce InnoEval, a deep innovation evaluation framework designed to emulate human-level idea assessment. We apply a heterogeneous deep knowledge search engine that retrieves and grounds dynamic evidence from diverse online sources. We further achieve review consensus with an innovation review board containing reviewers with distinct academic backgrounds, enabling a multi-dimensional decoupled evaluation across multiple metrics. We construct comprehensive datasets derived from authoritative peer-reviewed submissions to benchmark InnoEval. Experiments demonstrate that InnoEval can consistently outperform baselines in point-wise, pair-wise, and group-wise evaluation tasks, exhibiting judgment patterns and consensus highly aligned with human experts.

Peter Hayes, Mingtian Zhang, Raza Habib et al.

We introduce Integrated Weak Learning, a principled framework that integrates weak supervision into the training process of machine learning models. Our approach jointly trains the end-model and a label model that aggregates multiple sources of weak supervision. We introduce a label model that can learn to aggregate weak supervision sources differently for different datapoints and takes into consideration the performance of the end-model during training. We show that our approach outperforms existing weak learning techniques across a set of 6 benchmark classification datasets. When both a small amount of labeled data and weak supervision are present the increase in performance is both consistent and large, reliably getting a 2-5 point test F1 score gain over non-integrated methods.

Hossein A. Rahmani, Varsha Ramineni, Emine Yilmaz et al.

Test collections are crucial for evaluating Information Retrieval (IR) systems. Creating a diverse set of user queries for these collections can be challenging, and obtaining relevance judgments, which indicate how well retrieved documents match a query, is often costly and resource-intensive. Recently, generating synthetic datasets using Large Language Models (LLMs) has gained attention in various applications. While previous work has used LLMs to generate synthetic queries or documents to improve ranking models, using LLMs to create synthetic test collections is still relatively unexplored. Previous work~\cite{rahmani2024synthetic} showed that synthetic test collections have the potential to be used for system evaluation, however, more analysis is needed to validate this claim. In this paper, we thoroughly investigate the reliability of synthetic test collections constructed using LLMs, where LLMs are used to generate synthetic queries, labels, or both. In particular, we examine the potential biases that might occur when such test collections are used for evaluation. We first empirically show the presence of such bias in evaluation results and analyse the effects it might have on system evaluation. We further validate the presence of such bias using a linear mixed-effects model. Our analysis shows that while the effect of bias present in evaluation results obtained using synthetic test collections could be significant, for e.g.~computing absolute system performance, its effect may not be as significant in comparing relative system performance. Codes and data are available at: https://github.com/rahmanidashti/BiasSyntheticData.

Soyeong Jeong, Aparna Elangovan, Emine Yilmaz et al.

Large Language Models (LLMs) have demonstrated remarkable success in conversational systems by generating human-like responses. However, they can fall short, especially when required to account for personalization or specific knowledge. In real-life settings, it is impractical to rely on users to detect these errors and request a new response. One way to address this problem is to refine the response before returning it to the user. While existing approaches focus on refining responses within a single LLM, this method struggles to consider diverse aspects needed for effective conversations. In this work, we propose refining responses through a multi-agent framework, where each agent is assigned a specific role for each aspect. We focus on three key aspects crucial to conversational quality: factuality, personalization, and coherence. Each agent is responsible for reviewing and refining one of these aspects, and their feedback is then merged to improve the overall response. To enhance collaboration among them, we introduce a dynamic communication strategy. Instead of following a fixed sequence of agents, our approach adaptively selects and coordinates the most relevant agents based on the specific requirements of each query. We validate our framework on challenging conversational datasets, demonstrating that ours significantly outperforms relevant baselines, particularly in tasks involving knowledge or user's persona, or both.

Ruizhe Li, Chen Chen, Yuchen Hu et al.

Retrieval-Augmented Generation (RAG) leverages large language models (LLMs) combined with external contexts to enhance the accuracy and reliability of generated responses. However, reliably attributing generated content to specific context segments, context attribution, remains challenging due to the computationally intensive nature of current methods, which often require extensive fine-tuning or human annotation. In this work, we introduce a novel Jensen-Shannon Divergence driven method to Attribute Response to Context (ARC-JSD), enabling efficient and accurate identification of essential context sentences without additional fine-tuning, gradient-calculation or surrogate modelling. Evaluations on a wide range of RAG benchmarks, such as TyDi QA, Hotpot QA, and Musique, using instruction-tuned LLMs in different scales demonstrate superior accuracy and significant computational efficiency improvements compared to the previous surrogate-based method. Furthermore, our mechanistic analysis reveals specific attention heads and multilayer perceptron (MLP) layers responsible for context attribution, providing valuable insights into the internal workings of RAG models and how they affect RAG behaviours. Our code is available at https://github.com/ruizheliUOA/ARC_JSD.

Yue Feng, Hossein A. Rahmani, Aldo Lipani et al.

Task-oriented dialogue systems aim at providing users with task-specific services. Users of such systems often do not know all the information about the task they are trying to accomplish, requiring them to seek information about the task. To provide accurate and personalized task-oriented information seeking results, task-oriented dialogue systems need to address two potential issues: 1) users' inability to describe their complex information needs in their requests; and 2) ambiguous/missing information the system has about the users. In this paper, we propose a new Multi-Attention Seq2Seq Network, named MAS2S, which can ask questions to clarify the user's information needs and the user's profile in task-oriented information seeking. We also extend an existing dataset for task-oriented information seeking, leading to the \ourdataset which contains about 100k task-oriented information seeking dialogues that are made publicly available\footnote{Dataset and code is available at \href{https://github.com/sweetalyssum/clarit}{https://github.com/sweetalyssum/clarit}.}. Experimental results on \ourdataset show that MAS2S outperforms baselines on both clarification question generation and answer prediction.

Sahan Bulathwela, Maria Perez-Ortiz, Erik Novak et al.

Educational recommenders have received much less attention in comparison to e-commerce and entertainment-related recommenders, even though efficient intelligent tutors have great potential to improve learning gains. One of the main challenges in advancing this research direction is the scarcity of large, publicly available datasets. In this work, we release a large, novel dataset of learners engaging with educational videos in-the-wild. The dataset, named Personalised Educational Engagement with Knowledge Topics PEEK, is the first publicly available dataset of this nature. The video lectures have been associated with Wikipedia concepts related to the material of the lecture, thus providing a humanly intuitive taxonomy. We believe that granular learner engagement signals in unison with rich content representations will pave the way to building powerful personalization algorithms that will revolutionise educational and informational recommendation systems. Towards this goal, we 1) construct a novel dataset from a popular video lecture repository, 2) identify a set of benchmark algorithms to model engagement, and 3) run extensive experimentation on the PEEK dataset to demonstrate its value. Our experiments with the dataset show promise in building powerful informational recommender systems. The dataset and the support code is available publicly.

Sahan Bulathwela, Maria Perez-Ortiz, Emine Yilmaz et al.

With the emergence of e-learning and personalised education, the production and distribution of digital educational resources have boomed. Video lectures have now become one of the primary modalities to impart knowledge to masses in the current digital age. The rapid creation of video lecture content challenges the currently established human-centred moderation and quality assurance pipeline, demanding for more efficient, scalable and automatic solutions for managing learning resources. Although a few datasets related to engagement with educational videos exist, there is still an important need for data and research aimed at understanding learner engagement with scientific video lectures. This paper introduces VLEngagement, a novel dataset that consists of content-based and video-specific features extracted from publicly available scientific video lectures and several metrics related to user engagement. We introduce several novel tasks related to predicting and understanding context-agnostic engagement in video lectures, providing preliminary baselines. This is the largest and most diverse publicly available dataset to our knowledge that deals with such tasks. The extraction of Wikipedia topic-based features also allows associating more sophisticated Wikipedia based features to the dataset to improve the performance in these tasks. The dataset, helper tools and example code snippets are available publicly at https://github.com/sahanbull/context-agnostic-engagement

Nick Craswell, Bhaskar Mitra, Emine Yilmaz et al. · microsoft-research

This is the third year of the TREC Deep Learning track. As in previous years, we leverage the MS MARCO datasets that made hundreds of thousands of human annotated training labels available for both passage and document ranking tasks. In addition, this year we refreshed both the document and the passage collections which also led to a nearly four times increase in the document collection size and nearly $16$ times increase in the size of the passage collection. Deep neural ranking models that employ large scale pretraininig continued to outperform traditional retrieval methods this year. We also found that single stage retrieval can achieve good performance on both tasks although they still do not perform at par with multistage retrieval pipelines. Finally, the increase in the collection size and the general data refresh raised some questions about completeness of NIST judgments and the quality of the training labels that were mapped to the new collections from the old ones which we discuss in this report.

Nick Craswell, Bhaskar Mitra, Emine Yilmaz et al. · microsoft-research

This is the fourth year of the TREC Deep Learning track. As in previous years, we leverage the MS MARCO datasets that made hundreds of thousands of human annotated training labels available for both passage and document ranking tasks. In addition, this year we also leverage both the refreshed passage and document collections that were released last year leading to a nearly $16$ times increase in the size of the passage collection and nearly four times increase in the document collection size. Unlike previous years, in 2022 we mainly focused on constructing a more complete test collection for the passage retrieval task, which has been the primary focus of the track. The document ranking task was kept as a secondary task, where document-level labels were inferred from the passage-level labels. Our analysis shows that similar to previous years, deep neural ranking models that employ large scale pretraining continued to outperform traditional retrieval methods. Due to the focusing our judging resources on passage judging, we are more confident in the quality of this year's queries and judgments, with respect to our ability to distinguish between runs and reuse the dataset in future. We also see some surprises in overall outcomes. Some top-performing runs did not do dense retrieval. Runs that did single-stage dense retrieval were not as competitive this year as they were last year.

Zhengyan Shi, Adam X. Yang, Bin Wu et al.

Instruction tuning plays a crucial role in shaping the outputs of language models (LMs) to desired styles. In this work, we propose a simple yet effective method, Instruction Modelling (IM), which trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. Through experiments across 21 diverse benchmarks, we show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks (e.g., MMLU, TruthfulQA, and HumanEval) and open-ended generation benchmarks (e.g., MT-Bench and AlpacaEval). Remarkably, in the most advantageous case, IM boosts model performance on AlpacaEval 1.0 by over 100%. We identify two key factors influencing the effectiveness of IM: (1) The ratio between instruction length and output length in the training data; and (2) The number of training examples. We observe that IM is especially beneficial when trained on datasets with lengthy instructions paired with brief outputs, or under the Superficial Alignment Hypothesis (SAH) where a small amount of training examples are used for instruction tuning. Further analysis substantiates our hypothesis that our improvement can be attributed to reduced overfitting to instruction tuning datasets. It is worth noting that we are not proposing \ours as a replacement for current fine-tuning processes. Instead, our work aims to provide practical guidance for instruction tuning LMs, especially in low-resource scenarios.

Hossein A. Rahmani, Nick Craswell, Emine Yilmaz et al.

Test collections play a vital role in evaluation of information retrieval (IR) systems. Obtaining a diverse set of user queries for test collection construction can be challenging, and acquiring relevance judgments, which indicate the appropriateness of retrieved documents to a query, is often costly and resource-intensive. Generating synthetic datasets using Large Language Models (LLMs) has recently gained significant attention in various applications. In IR, while previous work exploited the capabilities of LLMs to generate synthetic queries or documents to augment training data and improve the performance of ranking models, using LLMs for constructing synthetic test collections is relatively unexplored. Previous studies demonstrate that LLMs have the potential to generate synthetic relevance judgments for use in the evaluation of IR systems. In this paper, we comprehensively investigate whether it is possible to use LLMs to construct fully synthetic test collections by generating not only synthetic judgments but also synthetic queries. In particular, we analyse whether it is possible to construct reliable synthetic test collections and the potential risks of bias such test collections may exhibit towards LLM-based models. Our experiments indicate that using LLMs it is possible to construct synthetic test collections that can reliably be used for retrieval evaluation.

Nick Craswell, Bhaskar Mitra, Emine Yilmaz et al.

This is the fifth year of the TREC Deep Learning track. As in previous years, we leverage the MS MARCO datasets that made hundreds of thousands of human-annotated training labels available for both passage and document ranking tasks. We mostly repeated last year's design, to get another matching test set, based on the larger, cleaner, less-biased v2 passage and document set, with passage ranking as primary and document ranking as a secondary task (using labels inferred from passage). As we did last year, we sample from MS MARCO queries that were completely held out, unused in corpus construction, unlike the test queries in the first three years. This approach yields a more difficult test with more headroom for improvement. Alongside the usual MS MARCO (human) queries from MS MARCO, this year we generated synthetic queries using a fine-tuned T5 model and using a GPT-4 prompt. The new headline result this year is that runs using Large Language Model (LLM) prompting in some way outperformed runs that use the "nnlm" approach, which was the best approach in the previous four years. Since this is the last year of the track, future iterations of prompt-based ranking can happen in other tracks. Human relevance assessments were applied to all query types, not just human MS MARCO queries. Evaluation using synthetic queries gave similar results to human queries, with system ordering agreement of $τ=0.8487$. However, human effort was needed to select a subset of the synthetic queries that were usable. We did not see clear evidence of bias, where runs using GPT-4 were favored when evaluated using synthetic GPT-4 queries, or where runs using T5 were favored when evaluated on synthetic T5 queries.

George Drayson, Emine Yilmaz, Vasileios Lampos

As Large Language Models (LLMs) become increasingly prevalent, their generated outputs are proliferating across the web, risking a future where machine-generated content dilutes human-authored text. Since online data is the primary resource for LLM pre-training, subsequent models could be trained on an unknown portion of synthetic samples. This could lead to model collapse, a degenerative process whereby LLMs reinforce their own errors, reduce output diversity, and ultimately yield declining performance. In this study, we investigate the impact of decoding strategy on model collapse, analysing the text characteristics at each model generation, the similarity to human references, and the resulting model performance. Using the decoding strategies that lead to the most significant degradation, we evaluate model collapse in a more realistic scenario where the origin of the data (human or synthetic) is unknown. We train a machine-generated text detector and propose an importance resampling approach to prevent model collapse by up-sampling likely human content in the training data. Our method is validated on four LLMs from two model families (GPT-2 and SmolLM2), across a range of model sizes 124M to 1.7B). We demonstrate that it not only prevents model collapse but also improves performance compared to training on purely human data, underscoring the benefit of synthetic samples and the importance of data curation.

Huichi Zhou, Kin-Hei Lee, Zhonghao Zhan et al.

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating external knowledge sources, enabling more accurate and contextually relevant responses tailored to user queries. These systems, however, remain susceptible to corpus poisoning attacks, which can severely impair the performance of LLMs. To address this challenge, we propose TrustRAG, a robust framework that systematically filters malicious and irrelevant content before it is retrieved for generation. Our approach employs a two-stage defense mechanism. The first stage implements a cluster filtering strategy to detect potential attack patterns. The second stage employs a self-assessment process that harnesses the internal capabilities of LLMs to detect malicious documents and resolve inconsistencies. TrustRAG provides a plug-and-play, training-free module that integrates seamlessly with any open- or closed-source language model. Extensive experiments demonstrate that TrustRAG delivers substantial improvements in retrieval accuracy, efficiency, and attack resistance.

Zheng Zhao, Clara Vania, Subhradeep Kayal et al.

Large language models (LLMs) have advanced conversational AI assistants. However, systematically evaluating how well these assistants apply personalization--adapting to individual user preferences while completing tasks--remains challenging. Existing personalization benchmarks focus on chit-chat, non-conversational tasks, or narrow domains, failing to capture the complexities of personalized task-oriented assistance. To address this, we introduce PersonaLens, a comprehensive benchmark for evaluating personalization in task-oriented AI assistants. Our benchmark features diverse user profiles equipped with rich preferences and interaction histories, along with two specialized LLM-based agents: a user agent that engages in realistic task-oriented dialogues with AI assistants, and a judge agent that employs the LLM-as-a-Judge paradigm to assess personalization, response quality, and task success. Through extensive experiments with current LLM assistants across diverse tasks, we reveal significant variability in their personalization capabilities, providing crucial insights for advancing conversational AI systems.

Yuxiang Qiu, Karim Djemili, Denis Elezi et al.

With the advancement and utility of Artificial Intelligence (AI), personalising education to a global population could be a cornerstone of new educational systems in the future. This work presents the PEEKC dataset and the TrueLearn Python library, which contains a dataset and a series of online learner state models that are essential to facilitate research on learner engagement modelling.TrueLearn family of models was designed following the "open learner" concept, using humanly-intuitive user representations. This family of scalable, online models also help end-users visualise the learner models, which may in the future facilitate user interaction with their models/recommenders. The extensive documentation and coding examples make the library highly accessible to both machine learning developers and educational data mining and learning analytics practitioners. The experiments show the utility of both the dataset and the library with predictive performance significantly exceeding comparative baseline models. The dataset contains a large amount of AI-related educational videos, which are of interest for building and validating AI-specific educational recommenders.

Xiao Fu, Hossein A. Rahmani, Bin Wu et al.

Personalised text generation is essential for user-centric information systems, yet most evaluation methods overlook the individuality of users. We introduce \textbf{PREF}, a \textbf{P}ersonalised \textbf{R}eference-free \textbf{E}valuation \textbf{F}ramework that jointly measures general output quality and user-specific alignment without requiring gold personalised references. PREF operates in a three-step pipeline: (1) a coverage stage uses a large language model (LLM) to generate a comprehensive, query-specific guideline covering universal criteria such as factuality, coherence, and completeness; (2) a preference stage re-ranks and selectively augments these factors using the target user's profile, stated or inferred preferences, and context, producing a personalised evaluation rubric; and (3) a scoring stage applies an LLM judge to rate candidate answers against this rubric, ensuring baseline adequacy while capturing subjective priorities. This separation of coverage from preference improves robustness, transparency, and reusability, and allows smaller models to approximate the personalised quality of larger ones. Experiments on the PrefEval benchmark, including implicit preference-following tasks, show that PREF achieves higher accuracy, better calibration, and closer alignment with human judgments than strong baselines. By enabling scalable, interpretable, and user-aligned evaluation, PREF lays the groundwork for more reliable assessment and development of personalised language generation systems.

Mariya Hendriksen, Svitlana Vakulenko, Jordan Massiah et al.

Large Language Models (LLMs) are transforming personalized search, recommendations, and customer interaction in e-commerce. Customers increasingly shop across multiple devices, from voice-only assistants to multimodal displays, each offering different input and output capabilities. A proactive suggestion to switch devices can greatly improve the user experience, but it must be offered with high precision to avoid unnecessary friction. We address the challenge of predicting when a query requires visual augmentation and a cross-device switch to improve product discovery. We introduce Image-Seeking Intent Prediction, a novel task for LLM-driven e-commerce assistants that anticipates when a spoken product query should proactively trigger a visual on a screen-enabled device. Using large-scale production data from a multi-device retail assistant, including 900K voice queries, associated product retrievals, and behavioral signals such as image carousel engagement, we train IRP (Image Request Predictor), a model that leverages user input query and corresponding retrieved product metadata to anticipate visual intent. Our experiments show that combining query semantics with product data, particularly when improved through lightweight summarization, consistently improves prediction accuracy. Incorporating a differentiable precision-oriented loss further reduces false positives. These results highlight the potential of LLMs to power intelligent, cross-device shopping assistants that anticipate and adapt to user needs, enabling more seamless and personalized e-commerce experiences.

Varsha Ramineni, Hossein A. Rahmani, Emine Yilmaz et al.

Ensuring fairness in AI systems is critical, especially in high-stakes domains such as lending, hiring, and healthcare. This urgency is reflected in emerging global regulations that mandate fairness assessments and independent bias audits. However, procuring the necessary complete data for fairness testing remains a significant challenge. In industry settings, legal and privacy concerns restrict the collection of demographic data required to assess group disparities, and auditors face practical and cultural challenges in gaining access to data. In practice, data relevant for fairness testing is often split across separate sources: internal datasets held by institutions with predictive attributes, and external public datasets such as census data containing protected attributes, each providing only partial, marginal information. Our work seeks to leverage such available separate data to estimate model fairness when complete data is inaccessible. We propose utilising the available separate data to estimate a set of feasible joint distributions and then compute the set plausible fairness metrics. Through simulation and real experiments, we demonstrate that we can derive meaningful bounds on fairness metrics and obtain reliable estimates of the true metric. Our results demonstrate that this approach can serve as a practical and effective solution for fairness testing in real-world settings where access to complete data is restricted.

Varsha Ramineni, Hossein A. Rahmani, Emine Yilmaz et al.

As AI becomes prevalent in high-risk domains and decision-making, it is essential to test for potential harms and biases. This urgency is reflected by the global emergence of AI regulations that emphasise fairness and adequate testing, with some mandating independent bias audits. However, procuring the necessary data for fairness testing remains a significant challenge. Particularly in industry settings, legal and privacy concerns restrict the collection of demographic data required to assess group disparities, and auditors face practical and cultural challenges in gaining access to data. Further, internal historical datasets are often insufficiently representative to identify real-world biases. This work focuses on evaluating classifier fairness when complete datasets including demographics are inaccessible. We propose leveraging separate overlapping datasets to construct complete synthetic data that includes demographic information and accurately reflects the underlying relationships between protected attributes and model features. We validate the fidelity of the synthetic data by comparing it to real data, and empirically demonstrate that fairness metrics derived from testing on such synthetic data are consistent with those obtained from real data. This work, therefore, offers a path to overcome real-world data scarcity for fairness testing, enabling independent, model-agnostic evaluation of fairness, and serving as a viable substitute where real data is limited.

Chuan Meng, Francesco Tonolini, Fengran Mo et al.

Proactive search in conversations (PSC) aims to reduce user effort in formulating explicit queries by proactively retrieving useful relevant information given conversational context. Previous work in PSC either directly uses this context as input to off-the-shelf ad-hoc retrievers or further fine-tunes them on PSC data. However, ad-hoc retrievers are pre-trained on short and concise queries, while the PSC input is longer and noisier. This input mismatch between ad-hoc search and PSC limits retrieval quality. While fine-tuning on PSC data helps, its benefits remain constrained by this input gap. In this work, we propose Conv2Query, a novel conversation-to-query framework that adapts ad-hoc retrievers to PSC by bridging the input gap between ad-hoc search and PSC. Conv2Query maps conversational context into ad-hoc queries, which can either be used as input for off-the-shelf ad-hoc retrievers or for further fine-tuning on PSC data. Extensive experiments on two PSC datasets show that Conv2Query significantly improves ad-hoc retrievers' performance, both when used directly and after fine-tuning on PSC.

Bin Wu, Zhengyan Shi, Hossein A. Rahmani et al.

Utilizing user profiles to personalize Large Language Models (LLMs) has been shown to enhance the performance on a wide range of tasks. However, the precise role of user profiles and their effect mechanism on LLMs remains unclear. This study first confirms that the effectiveness of user profiles is primarily due to personalization information rather than semantic information. Furthermore, we investigate how user profiles affect the personalization of LLMs. Within the user profile, we reveal that it is the historical personalized response produced or approved by users that plays a pivotal role in personalizing LLMs. This discovery unlocks the potential of LLMs to incorporate a greater number of user profiles within the constraints of limited input length. As for the position of user profiles, we observe that user profiles integrated into different positions of the input context do not contribute equally to personalization. Instead, where the user profile that is closer to the beginning affects more on the personalization of LLMs. Our findings reveal the role of user profiles for the personalization of LLMs, and showcase how incorporating user profiles impacts performance providing insight to leverage user profiles effectively.

Yue Feng, Yunlong Jiao, Animesh Prasad et al.

User Satisfaction Modeling (USM) is one of the popular choices for task-oriented dialogue systems evaluation, where user satisfaction typically depends on whether the user's task goals were fulfilled by the system. Task-oriented dialogue systems use task schema, which is a set of task attributes, to encode the user's task goals. Existing studies on USM neglect explicitly modeling the user's task goals fulfillment using the task schema. In this paper, we propose SG-USM, a novel schema-guided user satisfaction modeling framework. It explicitly models the degree to which the user's preferences regarding the task attributes are fulfilled by the system for predicting the user's satisfaction level. SG-USM employs a pre-trained language model for encoding dialogue context and task attributes. Further, it employs a fulfillment representation layer for learning how many task attributes have been fulfilled in the dialogue, an importance predictor component for calculating the importance of task attributes. Finally, it predicts the user satisfaction based on task attribute fulfillment and task attribute importance. Experimental results on benchmark datasets (i.e. MWOZ, SGD, ReDial, and JDDC) show that SG-USM consistently outperforms competitive existing methods. Our extensive analysis demonstrates that SG-USM can improve the interpretability of user satisfaction modeling, has good scalability as it can effectively deal with unseen tasks and can also effectively work in low-resource settings by leveraging unlabeled data.

Zhengxiang Shi, Francesco Tonolini, Nikolaos Aletras et al.

Semi-supervised learning (SSL) is a popular setting aiming to effectively utilize unlabelled data to improve model performance in downstream natural language processing (NLP) tasks. Currently, there are two popular approaches to make use of unlabelled data: Self-training (ST) and Task-adaptive pre-training (TAPT). ST uses a teacher model to assign pseudo-labels to the unlabelled data, while TAPT continues pre-training on the unlabelled data before fine-tuning. To the best of our knowledge, the effectiveness of TAPT in SSL tasks has not been systematically studied, and no previous work has directly compared TAPT and ST in terms of their ability to utilize the pool of unlabelled data. In this paper, we provide an extensive empirical study comparing five state-of-the-art ST approaches and TAPT across various NLP tasks and data sizes, including in- and out-of-domain settings. Surprisingly, we find that TAPT is a strong and more robust SSL learner, even when using just a few hundred unlabelled samples or in the presence of domain shifts, compared to more sophisticated ST approaches, and tends to bring greater improvements in SSL than in fully-supervised settings. Our further analysis demonstrates the risks of using ST approaches when the size of labelled or unlabelled data is small or when domain shifts exist. We offer a fresh perspective for future SSL research, suggesting the use of unsupervised pre-training objectives over dependency on pseudo labels.

Fanghua Ye, Zhiyuan Hu, Emine Yilmaz

Dialogue systems have received increasing attention while automatically evaluating their performance remains challenging. User satisfaction estimation (USE) has been proposed as an alternative. It assumes that the performance of a dialogue system can be measured by user satisfaction and uses an estimator to simulate users. The effectiveness of USE depends heavily on the estimator. Existing estimators independently predict user satisfaction at each turn and ignore satisfaction dynamics across turns within a dialogue. In order to fully simulate users, it is crucial to take satisfaction dynamics into account. To fill this gap, we propose a new estimator ASAP (sAtisfaction eStimation via HAwkes Process) that treats user satisfaction across turns as an event sequence and employs a Hawkes process to effectively model the dynamics in this sequence. Experimental results on four benchmark dialogue datasets demonstrate that ASAP can substantially outperform state-of-the-art baseline estimators.

Sahan Bulathwela, Hamze Muse, Emine Yilmaz

The automatic generation of educational questions will play a key role in scaling online education, enabling self-assessment at scale when a global population is manoeuvring their personalised learning journeys. We develop \textit{EduQG}, a novel educational question generation model built by adapting a large language model. Our extensive experiments demonstrate that \textit{EduQG} can produce superior educational questions by further pre-training and fine-tuning a pre-trained language model on the scientific text and science question data.

Fanghua Ye, Yue Feng, Emine Yilmaz

The MultiWOZ 2.0 dataset has greatly boosted the research on dialogue state tracking (DST). However, substantial noise has been discovered in its state annotations. Such noise brings about huge challenges for training DST models robustly. Although several refined versions, including MultiWOZ 2.1-2.4, have been published recently, there are still lots of noisy labels, especially in the training set. Besides, it is costly to rectify all the problematic annotations. In this paper, instead of improving the annotation quality further, we propose a general framework, named ASSIST (lAbel noiSe-robuSt dIalogue State Tracking), to train DST models robustly from noisy labels. ASSIST first generates pseudo labels for each sample in the training set by using an auxiliary model trained on a small clean dataset, then puts the generated pseudo labels and vanilla noisy labels together to train the primary model. We show the validity of ASSIST theoretically. Experimental results also demonstrate that ASSIST improves the joint goal accuracy of DST by up to $28.16\%$ on MultiWOZ 2.0 and $8.41\%$ on MultiWOZ 2.4, compared to using only the vanilla noisy labels.

Maria Perez-Ortiz, Sahan Bulathwela, Claire Dormann et al.

Prior research has shown how 'content preview tools' improve speed and accuracy of user relevance judgements across different information retrieval tasks. This paper describes a novel user interface tool, the Content Flow Bar, designed to allow users to quickly identify relevant fragments within informational videos to facilitate browsing, through a cognitively augmented form of navigation. It achieves this by providing semantic "snippets" that enable the user to rapidly scan through video content. The tool provides visually-appealing pop-ups that appear in a time series bar at the bottom of each video, allowing to see in advance and at a glance how topics evolve in the content. We conducted a user study to evaluate how the tool changes the users search experience in video retrieval, as well as how it supports exploration and information seeking. The user questionnaire revealed that participants found the Content Flow Bar helpful and enjoyable for finding relevant information in videos. The interaction logs of the user study, where participants interacted with the tool for completing two informational tasks, showed that it holds promise for enhancing discoverability of content both across and within videos. This discovered potential could leverage a new generation of navigation tools in search and information retrieval.

Sahan Bulathwela, María Pérez-Ortiz, Emine Yilmaz et al.

In informational recommenders, many challenges arise from the need to handle the semantic and hierarchical structure between knowledge areas. This work aims to advance towards building a state-aware educational recommendation system that incorporates semantic relatedness between knowledge topics, propagating latent information across semantically related topics. We introduce a novel learner model that exploits this semantic relatedness between knowledge components in learning resources using the Wikipedia link graph, with the aim to better predict learner engagement and latent knowledge in a lifelong learning scenario. In this sense, Semantic TrueLearn builds a humanly intuitive knowledge representation while leveraging Bayesian machine learning to improve the predictive performance of the educational engagement. Our experiments with a large dataset demonstrate that this new semantic version of TrueLearn algorithm achieves statistically significant improvements in terms of predictive performance with a simple extension that adds semantic awareness to the model.

Danula Hettiachchi, Mark Sanderson, Jorge Goncalves et al.

This volume contains the position papers presented at CSCW 2021 Workshop - Investigating and Mitigating Biases in Crowdsourced Data, held online on 23rd October 2021, at the 24th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2021). The workshop explored how specific crowdsourcing workflows, worker attributes, and work practices contribute to biases in data. The workshop also included discussions on research directions to mitigate labelling biases, particularly in a crowdsourced context, and the implications of such methods for the workers.

Aldo Lipani, Florina Piroi, Emine Yilmaz

Information availability affects people's behavior and perception of the world. Notably, people rely on search engines to satisfy their need for information. Search engines deliver results relevant to user requests usually without being or making themselves accountable for the information they deliver, which may harm people's lives and, in turn, society. This potential risk urges the development of evaluation mechanisms of bias in order to empower the user in judging the results of search engines. In this paper, we give a possible solution to measuring representation bias with respect to societal features for search engines and apply it to evaluating the gender representation bias for Google's Knowledge Graph Carousel for listing occupations.

Fangyu Liu, Yunlong Jiao, Jordan Massiah et al.

In NLP, a large volume of tasks involve pairwise comparison between two sequences (e.g. sentence similarity and paraphrase identification). Predominantly, two formulations are used for sentence-pair tasks: bi-encoders and cross-encoders. Bi-encoders produce fixed-dimensional sentence representations and are computationally efficient, however, they usually underperform cross-encoders. Cross-encoders can leverage their attention heads to exploit inter-sentence interactions for better performance but they require task fine-tuning and are computationally more expensive. In this paper, we present a completely unsupervised sentence representation model termed as Trans-Encoder that combines the two learning paradigms into an iterative joint framework to simultaneously learn enhanced bi- and cross-encoders. Specifically, on top of a pre-trained Language Model (PLM), we start with converting it to an unsupervised bi-encoder, and then alternate between the bi- and cross-encoder task formulations. In each alternation, one task formulation will produce pseudo-labels which are used as learning signals for the other task formulation. We then propose an extension to conduct such self-distillation approach on multiple PLMs in parallel and use the average of their pseudo-labels for mutual-distillation. Trans-Encoder creates, to the best of our knowledge, the first completely unsupervised cross-encoder and also a state-of-the-art unsupervised bi-encoder for sentence similarity. Both the bi-encoder and cross-encoder formulations of Trans-Encoder outperform recently proposed state-of-the-art unsupervised sentence encoders such as Mirror-BERT and SimCSE by up to 5% on the sentence similarity benchmarks.

Emine Yilmaz, Peter Hayes, Raza Habib et al.

Labelling data is a major practical bottleneck in training and testing classifiers. Given a collection of unlabelled data points, we address how to select which subset to label to best estimate test metrics such as accuracy, $F_1$ score or micro/macro $F_1$. We consider two sampling based approaches, namely the well-known Importance Sampling and we introduce a novel application of Poisson Sampling. For both approaches we derive the minimal error sampling distributions and how to approximate and use them to form estimators and confidence intervals. We show that Poisson Sampling outperforms Importance Sampling both theoretically and experimentally.

Nick Craswell, Bhaskar Mitra, Emine Yilmaz et al.

Evaluation efforts such as TREC, CLEF, NTCIR and FIRE, alongside public leaderboard such as MS MARCO, are intended to encourage research and track our progress, addressing big questions in our field. However, the goal is not simply to identify which run is "best", achieving the top score. The goal is to move the field forward by developing new robust techniques, that work in many different settings, and are adopted in research and practice. This paper uses the MS MARCO and TREC Deep Learning Track as our case study, comparing it to the case of TREC ad hoc ranking in the 1990s. We show how the design of the evaluation effort can encourage or discourage certain outcomes, and raising questions about internal and external validity of results. We provide some analysis of certain pitfalls, and a statement of best practices for avoiding such pitfalls. We summarize the progress of the effort so far, and describe our desired end state of "robust usefulness", along with steps that might be required to get us there.

Nick Craswell, Bhaskar Mitra, Emine Yilmaz et al.

The TREC Deep Learning (DL) Track studies ad hoc search in the large data regime, meaning that a large set of human-labeled training data is available. Results so far indicate that the best models with large data may be deep neural networks. This paper supports the reuse of the TREC DL test collections in three ways. First we describe the data sets in detail, documenting clearly and in one place some details that are otherwise scattered in track guidelines, overview papers and in our associated MS MARCO leaderboard pages. We intend this description to make it easy for newcomers to use the TREC DL data. Second, because there is some risk of iteration and selection bias when reusing a data set, we describe the best practices for writing a paper using TREC DL data, without overfitting. We provide some illustrative analysis. Finally we address a number of issues around the TREC DL data, including an analysis of reusability.