【作者】 宋巍； 张宇； 刘挺； 李生；

【Author】 SONG Wei,ZHANG Yu,LIU Ting,LI Sheng Information Retrieval Lab,Harbin Institute of Technology,Harbin,150001

【机构】 哈尔滨工业大学信息检索研究室；

【摘要】 基于检索历史隐式地学习用户偏好是个性化检索研究的热点,而根据用户检索历史重构新的查询输入是其中主要的研究内容。已有的研究在利用检索历史进行查询重构时,通常不区分检索历史中的内容是否与当前查询相关,而是将全部检索历史视为整体,因而使重构后的查询含有较多噪声。本文基于相关词语在上下文中大量共现的特征,将用户历史检索结果的网页摘要作为上下文语境,结合用户点击,选择检索历史中与当前查询共现程度最高的词语重构查询模型。对初始检索结果重排序的实验表明,该方法可以有效地选择相关词语,减少噪声。用p@5和NDCG两种指标评价,比最好的基准系统分别相对提高12.8%和7.2%,比初始排序结果相对提高26.0%和11.4%。更多还原

【Abstract】 Learning user preference implicitly is a hot research topic for personalized search and query model reformulation based on user search history is a key problem.Existing work consider the search history as a whole without distinguishing whether it is relevant to current query.Much noisy may be brought in due to that.In this paper,we assumed the relevant terms pretend to co-occurrence in context,viewed each past snippet as a context and modeled user clicks to select the terms most relevant to the whole query to reformulate query model for search results reranking.The experiment results show that the algorithm can select relevant terms and reduce noisy.With the evaluating metric p@5 and NDCG,the system achieves a relative improvement by 12.8%and 7.2%respectively against the best baseline system,26.0%and 11.4%compared to the original ranking.更多还原