个性化检索系统中用户兴趣模型构建研究

发布时间：2018-12-07 11:57

【摘要】：随着大数据时代的到来,如何管理、利用互联网中已存在的海量数据成为当下的热点。每天用户对于互联网产品的使用都产生大量的历史数据,但一般的搜索引擎并不会利用这些数据,其检索的结果也通常无法满足用户的个性化需求。因此,个性化检索系统已经成为互联网新技术中的一个研究热点。本文首先介绍了查询-概念二分图理论,通过分析其中对查询权重设计不合理的问题,提出以查询为单位划分用户浏览历史。本文利用从用户历史中抽取的概念描述查询,对于查询本文在计算其权重时参考了查询使用的时间、出现的先后次序等因素。用户兴趣建模过程中本文通过聚类算法将相似的查询组合在一起,降低了用户兴趣模型的维度,使得每个查询在模型中的比例更加合理。在模型存储和更新问题上,本文提出自己了的存储格式、更新算法。最后,利用构建的个性化检索原型系统,实现了用户浏览历史的捕捉,通过模拟用户的浏览过程验证了以查询为单位对用户兴趣的描述更合理。在与其他建模算法的对比中证明了本文提出的建模算法能够获得更为优秀的兴趣模型。系统运行的结果体现了本文设计的更行算法也较为合理。
[Abstract]:With the arrival of big data era, how to manage and utilize the huge amount of data in the Internet has become a hot spot. Every day, users produce a lot of historical data for the use of Internet products, but general search engines do not use these data, and their retrieval results are usually unable to meet the personalized needs of users. Therefore, personalized retrieval system has become a research hotspot in the new Internet technology. This paper first introduces the theory of query-concept bipartite graph. By analyzing the unreasonable design of query weights, this paper proposes to divide the browsing history of users with query as a unit. In this paper, we use the concept extracted from the user history to describe the query. For the query, we refer to the time used in the query, the order of occurrence and other factors in the calculation of the weight of the query. In the process of user interest modeling similar queries are combined by clustering algorithm which reduces the dimension of user interest model and makes the proportion of each query in the model more reasonable. In the problem of model storage and updating, this paper proposes its own storage format and update algorithm. Finally, the user's browsing history is captured by using the personalized retrieval prototype system, and the user's interest is described more reasonably by simulating the user's browsing process. In comparison with other modeling algorithms, it is proved that the proposed modeling algorithm can obtain better interest model. The result of the system operation shows that the algorithm designed in this paper is more reasonable.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【参考文献】