基于标签聚类和兴趣划分的个性化推荐算法研究

发布时间：2018-10-29 16:36

【摘要】：随着互联网的发展,大量信息出现在人们的视野中。信息爆炸使人们能更方便地接收多方面的信息。但与此同时,有价值信息的快速获取也变得更加困难。为了解决这种情况,人们通常在获取信息时先对其进行检索和过滤。搜索引擎作为信息检索技术的代表可以很好地帮助人们从海量的信息中检索出有用的信息。但当搜索的关键词不能恰当的反应出搜索需求时,查询的结果就会令人失望。而个性化推荐作为信息过滤中典型的应用正好可以弥补这方面的不足。目前主流的推荐算法包括基于内容的推荐、协同过滤推荐、基于规则的推荐、混合推荐等。在这些推荐算法中,协同过滤技术是实际应用中最为广泛的推荐技术。它根据产品评分和相似性算法选出与目标用户有着相似兴趣偏好的用户集合,再从这些相似用户评价高的产品中选出那些目标用户尚未评价过的产品推荐给用户。但传统的协同过滤没有考虑到标签对推荐结果的影响,只根据用户对资源的评分单方面挖掘用户兴趣,未能对用户兴趣进行有效划分,同时也忽略了用户兴趣随着时间推移发生的变化。为了解决以上问题,本文进行了如下研究:1.针对传统的协同过滤忽略了用户喜好因时间推移而发生的改变,本文提出了一种融合时间因子的协同过滤推荐算法。该算法考虑了产品评分时间和不同时段产品受关注的程度对用户兴趣偏好的影响,分别建立了时间遗忘模型和时间窗口模型,并把这两种模型融合,生成时间因子。之后,在用户相似度的计算中通过时间因子对产品评分进行时间上的过滤,从而能够更加准确地计算出目标用户的相似用户,减小因时间因素造成的推荐质量的下降。实验表明该法能有效地适应用户兴趣变化,提高智能Web系统在推荐中的准确率。2.考虑到用户与标签之间的关系,本文提出了一种基于标签聚类和兴趣划分的协同过滤推荐算法。该算法考虑了标签和用户评分对推荐结果的影响,通过标签聚类划分用户兴趣,并分别在标签和产品评分上对目标用户的相似用户进行选择。同时,在计算标签和产品评分权重时融入了时间因子,以适应用户的兴趣变化。实验部分,在Movielens数据集上通过交叉验证和与其它推荐算法的对比说明了该算法能有效的划分用户兴趣,减少时间因素对推荐质量的影响,提高推荐的准确度。
[Abstract]:With the development of the Internet, a lot of information appears in people's vision. Information explosion makes it easier for people to receive many kinds of information. But at the same time, rapid access to valuable information has become more difficult. In order to solve this problem, information is usually retrieved and filtered. As the representative of information retrieval technology, search engine can help people to retrieve useful information from a large amount of information. However, when the search keywords do not reflect the search requirements properly, the results of the query will be disappointing. Personalized recommendation as a typical application of information filtering can make up for this deficiency. The current mainstream recommendation algorithms include content-based recommendation, collaborative filtering recommendation, rule-based recommendation, mixed recommendation and so on. Among these recommendation algorithms, collaborative filtering is the most widely used recommendation technology. According to the product score and similarity algorithm, the users with similar interests and preferences are selected, and those products that have not been evaluated by the target users are selected from the products with high evaluation. However, the traditional collaborative filtering does not take into account the impact of labels on the recommended results, only according to the user's score of resources unilaterally mining user interest, failed to effectively divide user interest. It also ignores the changes in user interest over time. In order to solve the above problems, this paper has carried out the following research: 1. In view of the fact that the traditional collaborative filtering neglects the change of user preferences due to the passage of time, a collaborative filtering recommendation algorithm combining time factors is proposed in this paper. Taking into account the influence of product scoring time and the degree of product attention in different time periods on user interest preference, the time forgetting model and time window model are established, and the two models are combined to generate time factors. After that, in the calculation of user similarity, time factor is used to filter the product score, so that the similar users of target users can be calculated more accurately, and the quality of recommendation caused by time factors can be reduced. Experiments show that this method can effectively adapt to the change of user interest and improve the accuracy of intelligent Web system in recommendation. 2. Considering the relationship between users and tags, this paper proposes a collaborative filtering recommendation algorithm based on tag clustering and interest partition. The algorithm takes into account the influence of labels and user ratings on the recommended results, classifies user interests by label clustering, and selects similar users of target users in terms of labels and product ratings. At the same time, time factor is incorporated in the calculation of label and product rating weight to adapt to the change of user's interest. Experimental results show that the proposed algorithm can effectively divide user interest reduce the influence of time factors on recommendation quality and improve recommendation accuracy through cross-validation and comparison with other recommendation algorithms on Movielens data set.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】