Xia Zeng

Xia Zeng

Publications

Teaching LLM to be Persuasive: Reward-Enhanced Policy Optimization for Alignment frm Heterogeneous Rewards

Published in arxiv/under review, 2025

We propose Reward-Enhanced Policy Optimization (REPO), a reinforcement learning post-training framework that aligns an LLM with heterogeneous rewards: a preference-trained reward model (RM) for dense human alignment, an LLM-as-a-judge (RJ) for high-level persuasive behavior and SOP compliance, and programmatic reward functions (RF) for deterministic checks on numerics, formatting, and guardrails..

Recommended citation: tbc https://arxiv.org/abs/2510.04214

Combining Large Language Models and Crowdsourcing for Hybrid Human-AI Misinformation Detection

Published in SIGIR 2024, 2024

We study hybrid human-AI approaches jointly leveraging the potential of large language models and crowdsourcing.

Recommended citation: Xia Zeng, David La Barbera, Kevin Roitero, Arkaitz Zubiaga, and Stefano Mizzaro. 2024. Combining Large Language Models and Crowdsourcing for Hybrid Human-AI Misinformation Detection. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '24). Association for Computing Machinery, New York, NY, USA, 2332–2336. https://doi.org/10.1145/3626772.3657965 https://dl.acm.org/doi/pdf/10.1145/3626772.3657965

MAPLE: Micro Analysis of Pairwise Language Evolution for Few-Shot Claim Verification

Published in EACL Findings 2024, 2024

We propose MAPLE (Micro Analysis of Pairwise Language Evolution), a pioneering approach that explores the alignment between a claim and its evidence with a small seq2seq model and a novel semantic measure.

Recommended citation: Xia Zeng and Arkaitz Zubiaga. 2024. MAPLE: Micro Analysis of Pairwise Language Evolution for Few-Shot Claim Verification. In Findings of the Association for Computational Linguistics: EACL 2024, pages 1177–1196, St. Julian’s, Malta. Association for Computational Linguistics. https://aclanthology.org/2024.findings-eacl.79/

Active PETs: Active Data Annotation Prioritisation for Few-Shot Claim Verification with Pattern Exploiting Training

Published in EACL Findings 2023, 2023

We propose Active PETs, a novel weighted approach that utilises an ensemble of Pattern Exploiting Training (PET) models based on various language models, to actively select unlabelled data as candidates for annotation.

Recommended citation: Xia Zeng and Arkaitz Zubiaga. 2023. Active PETs: Active Data Annotation Prioritisation for Few-Shot Claim Verification with Pattern Exploiting Training. In Findings of the Association for Computational Linguistics: EACL 2023, pages 190–204, Dubrovnik, Croatia. Association for Computational Linguistics. https://aclanthology.org/2023.findings-eacl.14/

Aggregating pairwise semantic differences for few-shot claim verification

Published in PeerJ Computer Science, 2022

This article introduces Semantic Embedding Element-wise Difference (SEED), a novel vector-based method to few-shot claim verification that aggregates pairwise semantic differences for claim-evidence pairs.

Recommended citation: Zeng X, Zubiaga A. 2022. Aggregating pairwise semantic differences for few-shot claim verification. PeerJ Computer Science 8:e1137 https://doi.org/10.7717/peerj-cs.1137 https://peerj.com/articles/cs-1137/

QMUL-SDS at SCIVER: Step-by-Step Binary Classification for Scientific Claim Verification

Published in Proceedings of the Second Workshop on Scholarly Document Processing, 2021

In this paper, we propose an approach that performs scientific claim verification by doing binary classifications step-by-step.

Recommended citation: Zeng, X., & Zubiaga, A. (2021). QMUL-SDS at SCIVER: Step-by-step binary classification for scientific claim ver-ification. In Proceedings of the Second Workshop on Scientific Document Processing (pp. 116–123). Association for Computational Linguistics. https://aclanthology.org/2021.sdp-1.15.pdf