Abstract: The traditional query recommendation algorithms generate query suggestions by mining query logs of search engines. However, existing models focus more on the relationships (e.g., semantic similarity or relevance etc.) between original query and candidate suggestions without considering users’ search state during the search process. To address this challenge, this paper proposes a basic assumption for query suggestion inspired by the fact that users have different satisfaction states when searching for information, then verifies the assumption by conducting a large scale online user questionnaire. According to the verified assumption, this paper presents an auto-adaptive query recommendation model which is able to provide different categories of queries to users intelligently. In the proposed model, the system will provide new query suggestions to users when they are satisfied with current search results; on the contrary, the system will tend to recommend those query suggestions which are relevant and have more powerful representing ability. Large scale experiments on query logs show that the proposed model outperforms the traditional query flow graph model significantly and demonstrate the effectiveness of the proposed model.

Key words: query recommendation, query flow graph, search state, satisfaction

摘要： 传统的查询推荐算法通过挖掘查询日志为用户推荐查询词。通常现存模型只考虑原始查询词与推荐词之间的关系（例如语义相似性或相关性等），没有考虑用户在搜索过程中的满意度情况。针对用户在搜索过程中表现出的不同满意度状态，提出了一个查询推荐基本假设，并通过开展在线用户问卷调查，验证了这一假设。基于相应的假设，提出了一种基于用户搜索满意度状态的自适应查询推荐模型，该模型可以为用户智能推荐不同种类的查询词。当用户对搜索结果满意时，模型将为用户提供更加新颖的推荐词；当用户对搜索结果不满意时，模型将为用户提供一些增强信息表示能力的查询词。大规模日志实验表明，提出的推荐模型显著优于传统的查询流图模型，证明了所提模型的有效性。

关键词: 查询推荐, 查询流图, 搜索状态, 满意度