arXiv subjects

Zhumin Chu

Publications and source records attributed to Zhumin Chu.

6 recordsLinked to original sources

arXiv2025

Overview of the NTCIR-18 Automatic Evaluation of LLMs (AEOLLM) Task

In this paper, we provide an overview of the NTCIR-18 Automatic Evaluation of LLMs (AEOLLM) task. As large language models (LLMs) grow popular in both academia and industry, how to effectively evaluate the capacity of LLMs becomes an increasingly critical but still challenging issue. Existing methods can be divided into two types: manual evaluation, which is expensive, and automatic evaluation, which faces many limitations including task format (the majority belong to multiple-choice questions) and evaluation criteria (occupied by reference-based metrics). To advance the innovation of automatic evaluation, we propose the AEOLLM task which focuses on generative tasks and encourages reference-free methods. Besides, we set up diverse subtasks such as dialogue generation, text expansion, summary generation and non-factoid question answering to comprehensively test different methods. This year, we received 48 runs from 4 teams in total. This paper will describe the background of the task, the data set, the evaluation measures and the evaluation results, respectively.

cs.CL↗

arXiv2024

CalibraEval: Calibrating Prediction Distribution to Mitigate Selection Bias in LLMs-as-Judges

The use of large language models (LLMs) as automated evaluation tools to assess the quality of generated natural language, known as LLMs-as-Judges, has demonstrated promising capabilities and is rapidly gaining widespread attention. However, when applied to pairwise comparisons of candidate responses, LLM-based evaluators often exhibit selection bias. Specifically, their judgments may become inconsistent when the option positions or ID tokens are swapped, compromising the effectiveness and fairness of the evaluation result. To address this challenge, we introduce CalibraEval, a novel label-free method for mitigating selection bias during inference. Specifically, CalibraEval reformulates debiasing as an optimization task aimed at adjusting observed prediction distributions to align with unbiased prediction distributions. To solve this optimization problem, we propose a non-parametric order-preserving algorithm (NOA). This algorithm leverages the partial order relationships between model prediction distributions, thereby eliminating the need for explicit labels and precise mathematical function modeling.Empirical evaluations of LLMs in multiple representative benchmarks demonstrate that CalibraEval effectively mitigates selection bias and improves performance compared to existing debiasing methods. This work marks a step toward building more robust and unbiased automated evaluation frameworks, paving the way for improved reliability in AI-driven assessments

cs.CL↗

arXiv2024

Auto-PRE: An Automatic and Cost-Efficient Peer-Review Framework for Language Generation Evaluation

The rapid development of large language models (LLMs) has highlighted the need for efficient and reliable methods to evaluate their performance. Traditional evaluation methods often face challenges like high costs, limited task formats, dependence on human references, and systematic biases. To address these limitations, we propose Auto-PRE, an automatic LLM evaluation framework inspired by the peer review process. Unlike previous approaches that rely on human annotations, Auto-PRE automatically selects evaluator LLMs based on three core traits: consistency, pertinence, and self-confidence, which correspond to the instruction, content, and response stages, respectively, and collectively cover the entire evaluation process. Experiments on three representative tasks, including summarization, non-factoid QA, and dialogue generation, demonstrate that Auto-PRE achieves state-of-the-art performance while significantly reducing evaluation costs. Furthermore, the structured and scalable design of our automatic qualification exam framework provides valuable insights into automating the evaluation of LLMs-as-judges, paving the way for more advanced LLM-based evaluation frameworks.

cs.CL↗

arXiv2024

PRE: A Peer Review Based Large Language Model Evaluator

The impressive performance of large language models (LLMs) has attracted considerable attention from the academic and industrial communities. Besides how to construct and train LLMs, how to effectively evaluate and compare the capacity of LLMs has also been well recognized as an important yet difficult problem. Existing paradigms rely on either human annotators or model-based evaluators to evaluate the performance of LLMs on different tasks. However, these paradigms often suffer from high cost, low generalizability, and inherited biases in practice, which make them incapable of supporting the sustainable development of LLMs in long term. In order to address these issues, inspired by the peer review systems widely used in academic publication process, we propose a novel framework that can automatically evaluate LLMs through a peer-review process. Specifically, for the evaluation of a specific task, we first construct a small qualification exam to select "reviewers" from a couple of powerful LLMs. Then, to actually evaluate the "submissions" written by different candidate LLMs, i.e., the evaluatees, we use the reviewer LLMs to rate or compare the submissions. The final ranking of evaluatee LLMs is generated based on the results provided by all reviewers. We conducted extensive experiments on text summarization tasks with eleven LLMs including GPT-4. The results demonstrate the existence of biasness when evaluating using a single LLM. Also, our PRE model outperforms all the baselines, illustrating the effectiveness of the peer review mechanism.

cs.IR↗

arXiv2022

Corrected Evaluation Results of the NTCIR WWW-2, WWW-3, and WWW-4 English Subtasks

Unfortunately, the official English (sub)task results reported in the NTCIR-14 WWW-2, NTCIR-15 WWW-3, and NTCIR-16 WWW-4 overview papers are incorrect due to noise in the official qrels files; this paper reports results based on the corrected qrels files. The noise is due to a fatal bug in the backend of our relevance assessment interface. More specifically, at WWW-2, WWW-3, and WWW-4, two versions of pool files were created for each English topic: a PRI ("prioritised") file, which uses the NTCIRPOOL script to prioritise likely relevant documents, and a RND ("randomised") file, which randomises the pooled documents. This was done for the purpose of studying the effect of document ordering for relevance assessors. However, the programmer who wrote the interface backend assumed that a combination of a topic ID and a document rank in the pool file uniquely determines a document ID; this is obviously incorrect as we have two versions of pool files. The outcome is that all the PRI-based relevance labels for the WWW-2 test collection are incorrect (while all the RND-based relevance labels are correct), and all the RND-based relevance labels for the WWW-3 and WWW-4 test collections are incorrect (while all the PRI-based relevance labels are correct). This bug was finally discovered at the NTCIR-16 WWW-4 task when the first seven authors of this paper served as Gold assessors (i.e., topic creators who define what is relevant) and closely examined the disagreements with Bronze assessors (i.e., non-topic-creators; non-experts). We would like to apologise to the WWW participants and the NTCIR chairs for the inconvenience and confusion caused due to this bug.

cs.IR↗

arXiv2022

Towards Better Understanding of User Satisfaction in Open-Domain Conversational Search

With the increasing popularity of conversational search, how to evaluate the performance of conversational search systems has become an important question in the IR community. Existing works on conversational search evaluation can mainly be categorized into two streams: (1) constructing metrics based on semantic similarity (e.g. BLUE, METEOR and BERTScore), or (2) directly evaluating the response ranking performance of the system using traditional search methods (e.g. nDCG, RBP and nERR). However, these methods either ignore the information need of the user or ignore the mixed-initiative property of conversational search. This raises the question of how to accurately model user satisfaction in conversational search scenarios. Since explicitly asking users to provide satisfaction feedback is difficult, traditional IR studies often rely on the Cranfield paradigm (i.e., third-party annotation) and user behavior modeling to estimate user satisfaction in search. However, the feasibility and effectiveness of these two approaches have not been fully explored in conversational search. In this paper, we dive into the evaluation of conversational search from the perspective of user satisfaction. We build a novel conversational search experimental platform and construct a Chinese open-domain conversational search behavior dataset containing rich annotations and search behavior data. We also collect third-party satisfaction annotation at the session-level and turn-level, to investigate the feasibility of the Cranfield paradigm in the conversational search scenario. Experimental results show both some consistency and considerable differences between the user satisfaction annotations and third-party annotations. We also propose dialog continuation or ending behavior models (DCEBM) to capture session-level user satisfaction based on turn-level information.

cs.IR↗