融入社交关系与信任关系的移动应用推荐方法

发布时间：2018-09-07 09:24

【摘要】：移动互联网环境下,移动应用信息过载是亟待解决的问题。因此,个性化推荐技术成为解决移动应用信息过载的重要途径。传统推荐方法存在数据稀疏、冷启动等问题,采用余弦,Pearson方法计算相似度,当最近邻没有对待预测项目进行评分时,认为该用户对预测结果没有影响,从而影响推荐准确度。社交网络与用户信任关系是目前研究的热点,本文综合考虑了用户的社交关系,偏好及信任关系,提出一种融合用户社交关系与用户信任关系的移动应用推荐方法。该方法融合社交关系,集赞与标签等特征以及用户对应用的偏好计算相似度,利用基于熟人的信任关系与用户声誉计算信任度,并通过合理的将相似关系与信任关系融合进行应用的推荐,提出的方法能有效提高推荐准确度。本文的研究内容主要包括以下几个方面:(1)本文设计实现了移动应用偏好度计算方法。传统的移动应用偏好计算是直接基于用户使用的频次计算,这种方式没有考虑某些用户使用次数很多但是使用时长很少的情况,比如用,户使用次数较多,但是每次点开后使用时间很短,此时并不能认为用户真正喜欢该应用。本文综合考虑了使用频次和使用时长两方面,通过线性加权求和的方式计算用户对每个使用过的应用的偏好度。本文将用户对应用的偏好度代替传统的用户-项目评分,在一定程度上降低了数据稀疏性。(2)本文提出了基于社交关系评分预测模型。利用用户偏好度、用户社交相似度、社交互动行为,结合微信社交网络的特征,综合得到基于社交关系的评分预测模型。(3)本文提出了基于用户信任关系的评分预测模型。通过研究分析社交网络下用户信任传播机制得到基于社交关系的信任;通过微信社交网络的特点,构建微信社交网络下用户特征文档,从而得到用户基于声誉的信任,综合这两方面信任得到最终的基于用户信任关系预测模型。
[Abstract]:Under the environment of mobile Internet, information overload of mobile application is an urgent problem to be solved. Therefore, personalized recommendation technology has become an important way to solve the information overload of mobile applications. The traditional recommendation method has some problems such as sparse data, cold start, etc. When the nearest neighbor does not score the prediction items, it is considered that the user has no influence on the prediction results, thus affecting the recommendation accuracy. The relationship between social network and user trust is a hot topic at present. This paper considers the social relationship preference and trust relationship of users and proposes a mobile application recommendation method which combines user social relationship and user trust relationship. The method combines the features of social relations, likes and tags, and the users' preferences to calculate the similarity. The trust degree is calculated by using the trust relationship based on acquaintance and the reputation of users. The proposed method can effectively improve the accuracy of recommendation by applying the fusion of similarity relationship and trust relationship. The main contents of this paper are as follows: (1) this paper designs and implements a method to calculate the preference degree of mobile applications. The traditional calculation of mobile application preference is directly based on the frequency calculation used by the user. This method does not take into account the situation that some users use a lot of times but use a few hours, for example, the number of times used by the user is more than that of the user. However, after each point of use is very short, this time can not be considered that the user really like the application. In this paper, the frequency and duration of use are considered synthetically, and the user's preference for each used application is calculated by the method of linear weighted summation. In this paper, the user preference degree is replaced by the traditional user-item score, which reduces the data sparsity to a certain extent. (2) this paper proposes a prediction model based on social relationship score. By using user preference, user social similarity, social interaction behavior and the features of WeChat social network, a score prediction model based on social relationship is obtained. (3) this paper proposes a score prediction model based on user trust relationship. Through the research and analysis of user trust communication mechanism under social network to get trust based on social relationship, through the characteristics of WeChat social network, build user characteristic document under social network of WeChat, so as to get user trust based on reputation. Combining these two aspects of trust, the final prediction model based on user trust relationship is obtained.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3;C912.3

【参考文献】