基于标签信息的跨领域推荐算法研究

发布时间：2018-08-09 06:58

【摘要】：伴随信息技术和互联网应用的发展,网络上的信息发生了爆炸式的增长。然而面对海量信息,个人用户所能接触到的不过沧海一粟。在这样需求的推动下,个性化推荐技术(Personal Recommendation Technology)应运而生。传统的推荐技术仅仅是依靠单一领域的信息为本领域的用户进行推荐。然而随着互联网信息的发展,越来越多的信息平台交互连接,用户也越来越不满足于单一领域的信息来源,传统的单一领域推荐技术一直存在着数据稀疏、冷启动等问题,为了提高个性化推荐系统的准确性和多样性,跨领域信息推荐技术成为了当前的一个研究热点。跨领域推荐的优势在于能够综合分析来自多个领域的数据,对用户或者预测对象进行更加充分的建模,提高推荐结果的准确性;还能够为用户提供来自不同领域的预测对象的建议,提高推荐结果的多样性。基于以上种种优势,跨领域的推荐技术研究成为工业界和学术界的研究热点。一般的推荐算法,无论是单领域或是跨领域,主要是基于用户的评分数据来实现的,大部分情况下推荐算法被简化为了评分预测问题。然而这种形式使得推荐算法一直受制于评分数据稀疏的问题。因此,在推荐算法的发展过程中,其他类型的数据源也被纳入到考虑之中,期望以此来提高推荐算法的表现。其中,基于标签信息的推荐算法一直是研究的热点之一。标签是一种帮助用户描述和分类信息的关键字。用户可以自由的选择和描述最符合自己情况的标签,因此标签是一种可以强烈反应用户兴趣的信息。当前在各个网站和平台中充满了丰富的标签信息,这也为结合标签的推荐系统提供了可能。本文充分利用了多领域的标签信息,从而有效挖掘用户在不同领域中对信息对象的评价方式,提高了跨领域信息推荐的准确性,在一定程度上扩展了标签信息利用的新方式。最后,在真实数据集上验证了本文提出的跨领域推荐算法的有效性。
[Abstract]:With the development of information technology and Internet application, the information on the network has explosive growth. However, in the face of massive information, personal users can access but a drop in the ocean. Driven by this demand, personalized recommendation technology (Personal Recommendation Technology) came into being. Traditional recommendation technology only relies on single domain information to recommend users in this field. However, with the development of Internet information, more and more information platforms are connected, and users are not satisfied with the information source in a single field. The traditional single-domain recommendation technology has many problems, such as sparse data, cold start and so on. In order to improve the accuracy and diversity of personalized recommendation system, cross-domain information recommendation technology has become a research hotspot. The advantage of cross-domain recommendation is that it can analyze the data from many fields synthetically, model users or forecast objects more fully, and improve the accuracy of recommendation results. It can also provide users with suggestions from different areas of prediction objects, and improve the diversity of recommended results. Based on the above advantages, cross-domain recommendation technology research has become a research hotspot in industry and academia. General recommendation algorithms, whether single-domain or cross-domain, are mainly implemented on the basis of the user's rating data. In most cases, the recommendation algorithm is simplified to the problem of score prediction. However, this form makes the recommendation algorithm always subject to the problem of sparse rating data. Therefore, in the development of recommendation algorithms, other types of data sources are also taken into account in order to improve the performance of recommendation algorithms. Among them, the recommendation algorithm based on label information has been one of the hot research topics. Tags are keywords that help users describe and classify information. Users are free to choose and describe labels that best suit their needs, so tags are information that strongly reflects the user's interest. At present, various websites and platforms are full of rich tag information, which also provides the possibility for tag-based recommendation system. In this paper, we make full use of multi-domain tag information, so as to effectively mine users' evaluation methods of information objects in different fields, improve the accuracy of cross-domain information recommendation, and extend the new way of label information utilization to a certain extent. Finally, the effectiveness of the proposed cross-domain recommendation algorithm is verified on real data sets.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】