信息检索中基于用户检索历史挖掘的个性化查询自动补全方法研究

Query auto completion (QAC) is a prominent feature provided by many modern search engines. The goal of QAC is to accurately predict the user’s intended query with only a few keystrokes (i.e., typed prefix), helping the user formulate a satisfactory query with less effort and avoid misspelling it as well. In this project, we would like to focus on studying the following research tasks, i.e., selectively personalize query auto completion, learning-based personalized query auto completion as well as personalizing query auto completion diversification. First of all, in the task of selectively personalizing query auto completion, we will focus on analyzing user’s previous behaviors in current search session, i.e., queries, clicks and dwell time, to detect user’s instant search intent and dynamically inject personalization into a typical query auto completion model. For learning-based query auto completion model, we will concentrate on extracting user’s personal semantic features, which are based on a particular user’s search behaviors in current session. Thus, we will propose a learning-based query auto completion model, which introduces semantic similarity into a QAC model. Finally, we will focus on the task of personalizing query auto completion diversification, which aims at returning users a personalized ranking list of query auto completions with multiple query topics. We correctly recognize a query’s topics based on a built query-document bipartite graph, where each document is labeled by a set of topics as well as their relevance via the Open Directory Project. Based on a greedy algorithm, we propose a personalized query auto completion diversification model to push user’s intended query at the top position of the QAC ranking list as well as to generate a QAC ranking list with multiple topics.

查询自动补全是主流搜索引擎的重要服务之一，即在用户输入少量查询字符时，推荐给用户一组可选的查询短语，帮助用户构造查询，从而减少用户构造查询的时间。本项目将研究选择性个性化查询自动补全方法、基于机器学习的个性化查询自动补全方法和个性化的查询自动补全多样化方法。选择性个性化查询自动补全方法将着重分析用户的查询、网页点击等行为，识别用户实时信息查询意图，动态地将个性化嵌入传统的查询自动补全模型。基于机器学习的个性化查询自动补全方法将着重分析特定用户在当前信息检索会话中的行为数据，提取查询自动补全短语与用户行为数据信息之间的个性化语义特征，构建个性化查询自动补全模型。个性化的查询自动补全多样化方法将根据查询自动补全短语与文档的双向关系图，获取查询短语的主题与相关度标记，建立查询自动补全短语的主题概率分布，提出基于贪婪算法的个性化查询自动补全多样化方法，实现查询自动补全短语的准确排序和主题多样化。

查询自动补全是主流搜索引擎的重要服务之一，即在用户输入少量查询字符时，推荐给用户一组可选的查询短语，帮助用户构造查询，从而减少用户构造查询的时间。本项目主要针对信息检索用户查询意图的实时建模识别问题、基于用户行为分析的语义相似度特征提取问题、以及查询自动补全短语的多样化主题分类识别问题，.研究了选择性个性化查询自动补全方法、基于机器学习的个性化查询自动补全方法和个性化的查询自动补全多样化方法。选择性个性化查询自动补全方法将着重分析用户的查询、网页点击等行为，识别用户实时信息查询意图，动态地将个性化嵌入传统的查询自动补全模型。基于机器学习的个性化查询自动补全方法将着重分析特定用户在当前信息检索会话中的行为数据，提取查询自动补全短语与用户行为数据信息之间的个性化语义特征，构建个性化查询自动补全模型。个性化的查询自动补全多样化方法将根据查询自动补全短语与文档的双向关系图，获取查询短语的主题与相关度标记，建立查询自动补全短语的主题概率分布，提出基于贪婪算法的个性化查询自动补全多样化方法，实现查询自动补全短语的准确排序和主题多样化。通过项目研究，构建了一种基于用户检索历史分析的个性化查询自动补全模型、一种基于机器学习的个性化查询自动补全模型以及一种个性化的查询自动补全多样化模型，同时通过实验分析，验证了项目提出算法模型的有效性和合理性。