Publications

Published papers

• SAFE-QAQ: End-to-End Slow-Thinking Audio-Text Fraud Detection via Reinforcement Learning

  Peidong Wang^*, Zhiming Ma^*, Xin Dai, YongKang Liu, Shi Feng, Xiaocui Yang, Daling Wang, Wenxing Hu, Zhihao Wang

  Author ranking: First Author

  Contribution:

  • Proposed SAFE-QAQ, the first end-to-end fraud detection framework combining slow-thinking reasoning.
  • Introduced an optimization process, reducing reasoning lengths by 48.87\% without performance loss.
  • Developed a real-time detection system enabling 81.4\% faster detection (8.98s vs. 48.31s) while maintaining high accuracy.

  Submitted to ACM MM 2025 (under review)

• TeleAntiFraud-28k: An Audio-Text Slow-Thinking Dataset for Telecom Fraud Detection (Submitted to ACM MM 2025)

  Zhiming Ma^*, Peidong Wang^*, Minhua Huang, Jingpeng Wang, Kai Wu, Xiangzhao Lv, Yachun Pang, Yin Yang, Wenjie Tang, Yuchen Kang

  Author ranking: Co-first Author

  Contribution:

  • Proposed TeleAntiFraud-28k, the first multi-task slow-thinking audio-language dataset for telecommunication fraud prevention.
  • Designed a pipeline using real-call ASR, LLM simulation, and multi-agent generation to maximize fraud coverage.
  • Established TeleAntiFraud-Bench, a benchmark for evaluating telecom fraud models with slow-thinking assessments.

  Submitted to ACM MM 2025 (under review)

• Language Models as Continuous Self-Evolving Data Engineers

  Peidong Wang, Ming Wang, Zhiming Ma, Xiaocui Yang, Shi Feng, Daling Wang, Yifei Zhang, Kaisong Song

  Author ranking: First Author

  Contribution:

  • Proposed LANCE, a novel paradigm enabling LLMs to autonomously generate, clean, review, and annotate data for training themselves.
  • LANCE automates post-training data construction, improving efficiency, quality, and model performance across tasks by iterative self-training.
  • LANCE enhances mathematical reasoning, improving both basic and advanced tasks with general-purpose data.

  Submitted to ACL 2025 (under review)

• STICKERCONV: Generating Multimodal Empathetic Responses from Scratch

  Yiqun Zhang^*, Fanheng Kong^*, Peidong Wang^*, Shuang Sun, Lingshuai Wang, Shi Feng, Daling Wang, Yifei Zhang, Kaisong Song

  Author ranking: Co-first Author

  Contribution:

  • Introduced Agent4SC, a multi-agent framework that creates S{\small TICKER}C{\small ONV}, a multimodal empathetic dialogue dataset.
  • Designed PEGS📌, a framework generating empathetic text and sticker responses based on emotional and contextual dynamics.
  • Proposed a method for evaluating multimodal empathetic responses, leveraging LLMs to assess empathy, consistency, and ranking quality.

  Annual Meeting of the Association for Computational Linguistics(ACL 2024) [CCF A] [Home page] [PDF] [GitHub] [arxiv] [🤗]

• TIGER: A Unified Generative Model Framework for Multimodal Dialogue Response Generation

  Fanheng Kong, Peidong Wang, Shi Feng, Daling Wang, Yifei Zhang

  Author ranking: Second Author

  Contribution:

  • Proposed TIGER, a unified generative framework for multimodal dialogue response generation with text and images.
  • Achieved state-of-the-art results on automatic and human evaluations, validating TIGER’s effectiveness in multimodal conversations.

  The 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) [CCF B] [PDF] [GitHub] [Demo]