大语言模型对检索公平性与无偏性的挑战.pdf

《大语言模型对检索公平性与无偏性的挑战.pdf》由会员分享，可在线阅读，更多相关《大语言模型对检索公平性与无偏性的挑战.pdf（23页珍藏版）》请在三个皮匠报告上搜索。

1、大大语语言言模模型型对对检检索索公公平平性性与与无无偏偏性性的的挑挑战战徐君 中国人民大学高瓴人工智能学院YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP20242OutlineBias and Unfairness in IR+LLMsSource Bias in Document RetrievalYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP

2、2024Information Retrieval are Everywhere3Product SearchMusicAppsNew BingVideoYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024IR Meets LLM41 Unifying Bias and Unfairness in Information Retrieval:A Survey of Challenges and Opportunities with Large Language Mod

3、els,KDD 2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024A Unified View of Distribution Alignment5 Bias:Target ground truth distribution represents objective and factual realities Unfairness:Target ground truth distribution reflects human values and socia

4、l contracts and evolves with the progress of timePredicted Distribution Target Ground Truth Distribution1 Unifying Bias and Unfairness in Information Retrieval:A Survey of Challenges and Opportunities with Large Language Models,KDD 2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSN

5、LP2024YSSNLP2024YSSNLP2024A Unified View of Distribution Alignment6Taxonomies of Mitigation Strategies1 Unifying Bias and Unfairness in Information Retrieval:A Survey of Challenges and Opportunities with Large Language Models,KDD 2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP

6、2024YSSNLP2024YSSNLP20247OutlineBias and Unfairness in IR+LLMsSource Bias in Document RetrievalYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024Evolution of Information Retrieval ParadigmCorpus in Pre-LLM Era:Human-Written ContentCorpus in LLM Era:Human-Writt