基于用户浏览模式的新闻推荐系统设计

发布时间：2018-01-26 01:04

本文关键词： 新闻混合算法推荐系统　出处：《云南财经大学》2017年硕士论文　论文类型：学位论文

【摘要】：如今的因特网行业正在快速发展,在这个时代,信息数量巨大,更新速度飞快,使得网络浏览者在众多资讯中无法找到自己真正所需。为了解决这一问题,在推荐系统出现之前,人们运用搜索引擎通过关键词找到自己对信息的需求,然而某些场景下用户无法很精确地明确自己所需要的关键词,使得搜索引擎的效果大打折扣。个性化推荐经过对用户数据的分析,从而发现他们的相应特征与偏好,及时提供最符合用户的推荐结果。作为有效解决用户没有明确需求下的信息过载问题的工具之一,它已经变为许多领域的研究热点。个性化推荐系统可以智能地为因特网用户推荐他们所感兴趣的内容,让人们从海量数据的迷茫中解脱出来。在因特网新闻方面,个性化推荐也极其重要,今日头条网(http://www.toutiao.com/)、新浪新闻网(http://news.sina.com.cn/)等网站每天都在发布各行各业的时事新闻,随着新闻信息量与信息更新速度的不断增大,网页新闻浏览者难以看到自身所感兴趣的新闻内容,常常让自己丢失在海量级别的新闻资讯中。当遇到这一类问题时,新闻推荐系统可以根据浏览者个性化的浏览记录,发掘出他们的潜在浏览偏好,形成相应的推荐结果。从而节约了大量浏览者的新闻探寻时间,提高了浏览者的满意度,同时降低网页新闻资源浪费程度。利用用户的显式反馈信息进行推荐的推荐方法是目前比较常见的方法,然而相对于显式反馈,由于隐式反馈信息更容易获取,具有普遍性,因此根据隐式反馈信息所设计的推荐系统具有更加广泛的适用性,本文所设计的推荐系统是根据隐式反馈信息所设计的。本文主要对网页新闻浏览者的隐式反馈数据进行处理,对推荐模型以及推荐算法、用户模型的构建、推荐的混合方案和策略等内容开展研究,将浏览者群体按照浏览频率进行划分,对不同浏览者群体采用不同推荐算法混合,对于经常浏览用户,综合用户协作型过滤算法、内容推荐算法进行结果上的混合,对于不常浏览用户,综合了物品协作过滤算法的相似度计算以及内容推荐算法的相似度计算法则,进行相应算法上的混合,并将得出的相应推荐结果与基于随机漫步的PersonalRank算法进行混合。使得推荐系统中单一算法存在的问题如新加入物品的推荐、数据的稀疏性等不足得以降低。根据上述设计思路以及相应算法的实现完成了整个新闻推荐系统的设计,同时本文所使用的混合策略的有效性在后续实验中根据相应评价指标的对比得以验证。
[Abstract]:Nowadays, the Internet industry is developing rapidly. In this era, the amount of information is huge and the update speed is very fast. In order to solve this problem, the users of the Internet can't find what they really need in a lot of information. Before the emergence of recommendation systems, people use search engines to find their needs for information through keywords. However, in some scenarios, users can not identify the keywords they need very accurately. Personalized recommendation through the analysis of user data to find their corresponding characteristics and preferences. Timely provide the most consistent with the user's recommended results. As an effective solution to the problem of information overload without clear requirements of the user one of the tools. It has become a research hotspot in many fields. Personalized recommendation systems can intelligently recommend content of interest to Internet users. Personalised recommendation is also extremely important in Internet news, and today's headline is http: / / www.toutiao.com.com.com.com. (http: / / www.toutiao.com / www.toutiao.com / www.toutiao.com / www.toutiao.com / www.toutiao.com / /. Websites such as http: / / news.sina.com.cn.cn.cn.com., etc., publish news about current affairs in various industries every day, as the amount of news and the rate of update increases. Web news viewers find it difficult to see the news content they are interested in and often lose themselves in the mass of news information. When it comes to this kind of problems. The news recommendation system can discover their potential browsing preferences according to their personalized browsing records, and form the corresponding recommendation results, thus saving a large number of visitors' news search time. It can improve the satisfaction of visitors and reduce the waste of web news resources. Using explicit feedback from users to recommend is a relatively common method at present, but relative to explicit feedback. Because implicit feedback information is easier to obtain and universal, the recommendation system designed based on implicit feedback information has more extensive applicability. The recommendation system designed in this paper is based on the implicit feedback information. This paper mainly deals with the implicit feedback data of the page news viewer, and constructs the recommendation model, recommendation algorithm and user model. The content of the recommended mix scheme and strategy is studied, the viewer group is divided according to the browsing frequency, the different recommendation algorithm is used to the different visitors group, and the frequent browsing user is used. Integrated user collaborative filtering algorithm, content recommendation algorithm for the results of the hybrid, for the less frequent browsing users, the integration of articles collaborative filtering algorithm similarity calculation and content recommendation algorithm similarity calculation rules. The corresponding algorithm is mixed. The corresponding recommendation results are mixed with the PersonalRank algorithm based on random walk, which makes the problem of single algorithm in recommendation system such as the recommendation of newly added items. The lack of data sparsity can be reduced. According to the above design ideas and the corresponding algorithm to complete the design of the entire news recommendation system. At the same time, the effectiveness of the hybrid strategy used in this paper is verified by comparison of corresponding evaluation indexes in subsequent experiments.
【学位授予单位】：云南财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3;G252

【参考文献】