Abstract:

With the development of retrieval technology, interactive retrieval becomes more and more important in the field of information retrieval. Interactive retrieval adds the ability to capture the fine-grained behavior of users in the traditional retrieval mode in order to improve the performance of the retrieval engine through the user simulator. Rule-based user simulators lack of personalized user features and cannot learn new behavior, poor adaptability. Model-based user simulators can learn more user personalized behavior characteristics, which can effectively improve the performance of interactive search engine. This paper describes the interactive process of user simulator and retrieval engine, summarizes the construction methods of rule-based user simulator and model-based user simulator and the evaluation methods of user simulator in recent years, and focuses on the introduction of model-based user simulator. Finally, the paper compares the differences between the user simulator for interactive retrieval and the traditional user simulator, and takes the interactive dissertation retrieval scenario as an example to forecast the application of the user simulator.

Key words: user simulator, interactive retrieval, reinforcement learning, artificial intelligence, personalized characteristics of users

摘要：

随着检索技术的发展，交互式检索在信息检索领域中变得尤为重要。交互式检索在传统检索模式下增加了捕捉用户细粒度行为的功能，以便通过用户模拟器提升检索引擎性能。基于规则的用户模拟器缺乏个性化用户特征，适应性较差。基于模型的用户模拟器能够学习到更多的用户个性化行为特征，可以有效提升交互式检索引擎的性能。阐述了用户模拟器与检索引擎的交互过程，对基于规则的用户模拟器和基于模型的用户模拟器的构建方法以及近年来用户模拟器的评价方法进行了归纳总结，并重点介绍了基于模型的用户模拟器。最后对比了面向交互式检索的用户模拟器和传统的用户模拟器的差异，并以交互式学位论文检索场景为例，通过此检索场景对用户模拟器的应用进行了展望。

关键词: 用户模拟器, 交互式检索, 强化学习, 人工智能, 用户个性化特征