基于用户上网数据的电影个性化推荐系统研究

发布时间：2018-08-21 12:49

【摘要】：近十年间,随着互联网逐渐在社会普及,信息爆炸现象日益明显,各行业各领域用户不断为互联网提供信息输送,使得互联网成为一个无所不包、无所不含的信息汇聚体,而互联网用户很难从其中快速地找到契合自己兴趣的信息;每个用户在使用搜索引擎检索信息时,同一个关键字得到的结果是相同的,而用户对信息的需求是多元化和个性化的,因此,传统的同质化的搜索引擎为代表的信息检索系统已不能满足用户千人千面的需求,个性化推荐系统在这种背景下走到了台前。个性化推荐系统通过挖掘用户的历史行为数据,从中提取出与兴趣相关的记录,根据一定的规则算法计算用户的兴趣点,然后主动向用户进行信息的个性化推送,从而解决了信息量大和信息选择困难之间的矛盾。推荐系统通过长期追踪用户历史行为,不断更新与迭代用户兴趣,使推荐的信息始终与用户的兴趣点相贴合,让用户更加方便的获取到自己感兴趣的信息,最终目的是实现针对用户全方位的个性化定制推送。本文讲述了如何构建出完整电影知识图谱的方法,从而可以结构化描述用户行为;同时根据用户的观影行为特征和电影本身的属性,将电影分为独立电影和系列电影,更细粒度的构建电影知识图谱,同时提出了一套发掘电影系列的实现算法。基础数据为用户的上网请求,在不需要用户参与的情况下获得用户的电影兴趣行为,避免了用户主观选择存在的不全面、不方便等问题,通过分析和处理用户的原始上网请求,从中提取出与电影相关的上网数据,然后根据电影知识图谱将用户上网行为一一映射为用户兴趣行为,达到提取用户兴趣的目的。基于TF-IDF算法计算用户各维度元素兴趣度,构建向量形式的用户兴趣模型,然后根据用户对某电影各要素的兴趣度,计算出用户对该电影的总兴趣度。最后,通过实验分析,证明了本文所提方案的高召回率和准确率。
[Abstract]:In the past ten years, with the gradual popularization of the Internet in society, the phenomenon of information explosion is becoming more and more obvious. Users in various fields provide information for the Internet, which makes the Internet an all-encompassing and all-inclusive information aggregator. Internet users find it hard to quickly find information that suits their interests; each user uses a search engine to retrieve information with the same keyword and the same result. However, the information demand of users is diversified and individualized. Therefore, the traditional information retrieval system represented by homogeneous search engines can not meet the needs of thousands of users, and the personalized recommendation system has come to the front of the stage under this background. By mining the historical behavior data of users, the personalized recommendation system extracts the records related to interest, calculates the points of interest of users according to certain rules algorithm, and then proactively pushes the information to the users. Thus, the contradiction between the large amount of information and the difficulty of information selection is solved. The recommendation system continuously updates and iterates the user's interest by tracking the user's historical behavior for a long time, so that the recommended information always matches the user's point of interest, so that the user can obtain the information of his interest more conveniently. The ultimate goal is to achieve the user-oriented personalized customization push. This paper describes how to construct a complete film knowledge map so as to structurally describe user behavior, and divide films into independent films and series films according to the characteristics of the user's viewing behavior and the properties of the film itself. A finer-grained film knowledge map is constructed, and a set of algorithms for discovering film series is proposed. The basic data is the user's online request, which can obtain the user's interest in movies without the user's participation, and avoid the problems such as incompleteness and inconvenient of the user's subjective choice. Through analyzing and processing the user's original Internet request, The online data related to movies are extracted, and then, according to the movie knowledge map, the user's online behavior is mapped to user's interest behavior, and the purpose of extracting user's interest is achieved. Based on the TF-IDF algorithm, the user interest degree of each dimension is calculated, and the user interest model in vector form is constructed, and then the total interest degree of the user to the movie is calculated according to the interest degree of the user to each element of a movie. Finally, the high recall rate and accuracy of the proposed scheme are proved by experimental analysis.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【相似文献】