Web信息检索中搜索结果个性化和多样化算法的融合技术研究

The vast majority of queries to search engines are ambiguous or broad. Different users may have completely different information needs and goals when using precisely the same query for search. Personalized Search and Search Result Diversification are two effective ways to solve this problem. Personalized search provides different search results to different users based upon their interests and preferences; whereas search result diversification aims to provide a list of results that cover as many aspects as possible so that most users can be satisfied by the top results. Both approaches have received much attention in recent years and lots of methods has been proposed for either personalizing or diversifying search results. However, little work has been done on analyzing the relationship between these two types of approaches and no consensus has been reached on whether and how they can be combined together to achieve better search performance. In this project, we will address this problem and aim to propose effective fusion methods for combining personalization and diversification in web search. First, as the basis of the project, we will investigate the methods for analyzing query intent based on knowledge base. We propose first detecting entities within queries, then identifying the head and modifier entities. Based on the understanding result, we will build classification models to identify the ambiguity or broadness of queries then mine subtopics for broad or ambiguous queries; Second, we will analyze how and when personalization or diversification could improve search performance, and build models that can predict performance of the algorithms. Third, based on the study of the first two problems, we will investigate the fusion methods for combining personalization and diversification. We will investigate a weak coupling method that can first personalize then diversify results based on a decision tree. We will also develop deep fusion methods that directly embed personalization and diversification into a single ranking function.

搜索引擎中的查询词往往具有歧义性和模糊性。不同用户即使使用了相同的查询词，也往往具有不同搜索意图。个性化搜索和搜索结果多样化技术是解决这种问题、提高用户满意度的两种有效方法。个性化搜索基于用户的知识背景和兴趣爱好返回给用户个性化的结果，而搜索结果多样化的目标是提高结果列表尤其是顶部的多样性，以保证不同用户都能在搜索结果的靠前位置查找到所需要的信息。目前已经有一系列单独的个性化或多样化算法被提出，但对二者之间的关联关系研究仍然较少。个性化和多样化能否结合，是否可以结合是重要但尚未被深入研究的问题。针对这一问题，本项目以查询意图的理解和分析为基础，充分研究个性化和多样化算法的作用机理及性能预测方法，以此为基础分别研究基于决策树的弱耦合个性化和多样化融合方法及直接对个性化和多样化联合建模的深度融合方法，为搜索引擎进一步提高用户满意度奠定技术基础。

搜索引擎中的查询词往往具有歧义性和模糊性。不同用户即使使用了相同的查询词，也往往具有不同搜索意图。个性化搜索和搜索结果多样化技术是解决这种问题、提高用户满意度的两种有效方法。个性化搜索基于用户的知识背景和兴趣爱好返回给用户个性化的结果，而搜索结果多样化的目标是提高结果列表尤其是顶部的多样性，以保证不同用户都能在搜索结果的靠前位置查找到所需要的信息。目前已经有一系列单独的个性化或多样化算法被提出，但如何进一步改善这些方法，并将二者有效融合，从而达成更高质量的搜索精度并提高用户满意度仍然是值得深入研究的问题。针对这一问题，本项目以查询意图的理解和分析为基础，充分研究个性化和多样化机理及性能预测方法，提出了一系列新的多样化和个性化算法，并以此为基础探索了二者的融合机制，为搜索引擎进一步提高用户满意度奠定了技术基础。项目在研期间发表了一系列国际会议和期刊论文，申请了相关专利和软件著作权。项目研究成果进一步推动了相关领域的研究进展，并在搜索引擎和推荐系统应用方面具有一定的实际应用和推广价值。