Study On A Community Question Answering Based Exploratory Search Query Expansion Method

While Web search engines have significantly progressed in effectiveness and efficiency, there still exist certain user needs that cannot be satisfied. These needs cannot be generally achieved by a traditional one-shot search, and thus need more complex search processes which just present the aspect of exploratory search. Besides, a large number of studies have shown that, most of users whose information needs cannot be achieved, can address their needs via asking questions and waiting for answers in Community Question Answering (CQA) systems. These features of CQA can guarantee the promising question-answering resources for exploratory search.Based on the above observations, this thesis proposes a CQA based exploratory search query expansion method. This thesis explores how to select questions and answers related to queries submitted by users, how to extract key concepts from these selected questions and answers, and then how to utilize these concepts as query expansions. By these means, valuable information which can help users achieve exploratory search can be discovered.Specifically, semantic relations between questions and answers are utilized to select questions and answers related to exploratory queries. Notice that there really exist two interesting phenomena of user questioning and answering in CQA that can be leveraged to effectively identify subtopics of questions and answers, referred to as 'a same sense between question and answer' and 'subtopics overlap between two questions with different length', and then this thesis leverages them to cluster words in questions and answers, thus to generate label sets and word sets of subtopics. Subsequently, based on the latent relations of questions and answers information in subtopics, the semantic similarities between words in subtopics and exploratory queries submitted by users are calculated to generate candidate expansion concepts. Finally, based on the relations between label sets and word sets of subtopics and the similarities between candidate expansion concepts, this thesis structures concept hierarchy models to achieve the ranking and selection of expansion concepts, and then utilizes the concept hierarchy models structured to expand queries as multiple sets of concepts sequences so as to help users to achieve exploratory search.Experiments are conducted for the above methods proposed. This thesis mainly sets contrast experiments between baseline methods and methods of generating candidate expansion concepts and methods of ranking and selecting expansion concepts. Experimental results show that based on subtopics of questions and answers mined, candidate expansion concepts generating method and expansion concepts ranking and selection method can be effectively leveraged to discover more valuable information for users.