一种关系数据库对象级别检索结果的聚类方法

摘要

Tuple-level keyword retrieval in relational database has the problems of the querying sentence ambiguity and the presenting results redundancy and so on.Aiming at the aforesaid problems, we propose an object-level retrieval results clustering method for relational databases.This method takes into account the pertinence and the diversity of retrieval results comprehensively from the perspective of the object, and clusters them from two levels of structure and content.Based on cover tree it makes the isomorphic judgement on retrieval results to realise the first-level clustering;It uses the kernel function to calculate the similarity of the contents contained between the retrieval results in isomorphic classes to realise the second-level clustering.Simultaneously, it dynamically updates the clustered result set.This clustering method reduces the redundancy of the presenting results effectively, increases the available results categories for user to select, and improves the performance of the retrieval system.%针对关系数据库元组级别关键词检索中存在查询语句多义性及展现结果冗余性等问题，提出一种关系数据库对象级别检索结果的聚类方法。以对象的观点，综合考虑检索结果的相关性和多样性，从结构和内容两个层面对其聚类。基于覆盖树对检索结果进行同构判断，实现第一级聚类；利用核函数计算同构类别中检索结果间所包含内容的相似性，实现第二级聚类；同时对聚类后的结果集进行动态更新。该聚类方法有效降低了展现结果的冗余性，增加了用户可选择的结果类别，提高了检索系统的性能。