基于用户行为的个性化推荐优化方法研究

发布时间：2018-04-17 04:13

本文选题：个性化推荐 + 用户行为　；参考：《哈尔滨商业大学》2017年硕士论文

【摘要】：随着电子商务和互联网的发展普及,使得面向用户的个性化推荐越来越被重视,准确的推荐不仅仅能帮助用户节省大量时间而且可以帮助电子商务网站赢得用户的关注,进而提高电子商务网站的销量。然而不同用户的背景不同,对推荐结果的期望也不同,因此需要从多个角度挖掘用户隐藏的信息,更好的为用户给出个性化的推荐。本文研究基于用户的潜在行为以及社交关系发掘用户的偏好,从以下方面优化用户个性化推荐:研究用户行为时效特征以及用户行为和偏好的关系,并提升预测用户偏好的效率;研究网页拓扑与网页权重的关系,进而优化网页排序,为用户给出准确的推荐;研究基于负向社交关系以及主题被推荐给用户的概率特征为历史信息非常稀少的非活跃用户给出推荐,优化基于负向社交关系和推荐主题概率特征的融合,为非活跃用户给出准确的推荐。提出以下方法优化用户的个性化推荐:(1)提出一种优化基于行为感知的用户个性化推荐方法,通过分析用户的历史访问行为,建立映射用户行为和偏好关系的隐马尔可夫模型,并通过聚类减少优化用户参数的时间,得到一种平衡准确度和时间复杂度的个性化推荐方法。(2)提出基于网页拓扑识别网页异常排名的用户推荐优化方法,基于网页拓扑识别异常提升网页排名的行为,研究网页拓扑对网页权重值的提升效果,并通过网页的链入权重值和链入网页数量的关系识别网页异常排名现象,为个性化推荐创造公正的网页排名环境,进而提高对用户的个性化推荐质量。(3)提出基于负向社交关系和泊松过程融合的非活跃用户推荐优化方法,由用户间的初始负向社交关系和衰减传递系数矩阵扩展用户间的负向社交关系,基于负向社交关系约束为非活跃用户给出推荐主题的期望,然后基于泊松过程预测用户对推荐主题满意的概率,将概率高的推荐主题作为最终的推荐发送给用户,为非活跃用户给出准确的推荐。
[Abstract]:With the development of electronic commerce and Internet, more and more attention has been paid to personalized recommendation for users. Accurate recommendation can not only help users save a lot of time, but also help e-commerce websites win the attention of users.And then improve the sales of e-commerce websites.However, different users have different backgrounds and different expectations for the recommended results. Therefore, it is necessary to mine hidden information from multiple angles to provide personalized recommendations for users.In this paper, we study the potential behaviors and social relationships of users to explore their preferences, and optimize their personalized recommendations from the following aspects: the characteristics of user behavior and the relationship between user behaviors and preferences.It also improves the efficiency of predicting user preferences, studies the relationship between web topology and web page weights, and then optimizes the ranking of web pages to give users accurate recommendations.The probabilistic features based on negative social relationships and topics recommended to users are proposed for inactive users with very little historical information, and the fusion of probability features based on negative social relationships and recommended themes is optimized.Give accurate recommendations for inactive users.This paper proposes the following methods to optimize the personalized recommendation of users: 1) A user personalized recommendation method based on behavioral awareness is proposed. By analyzing the historical access behavior of users, a hidden Markov model is established to map the relationship between user behavior and preference.Through clustering to reduce the time of optimizing user parameters, a personalized recommendation method, which balances accuracy and time complexity, is obtained.) based on the abnormal ranking of web page topology, a user recommendation optimization method is proposed.Based on the behavior of page topology recognition anomaly to enhance the ranking of web pages, this paper studies the effect of web topology on the weight of web pages, and identifies the abnormal ranking phenomenon of web pages by the relationship between the value of link weight and the number of web pages.In order to create a fair website ranking environment for personalized recommendation, and then improve the quality of personalized recommendation to users, this paper proposes an optimization method of inactive user recommendation based on negative social relationship and Poisson process fusion.The initial negative social relationship between users and the attenuation transfer coefficient matrix are used to expand the negative social relationship among users. Based on the constraint of negative social relationship, the expectation of recommending topics for inactive users is obtained.Then, based on the Poisson process, the probability of users' satisfaction with the recommendation topic is forecasted, and the recommendation topic with high probability is sent to the user as the final recommendation to give the accurate recommendation for inactive users.
【学位授予单位】：哈尔滨商业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3;F713.55

【参考文献】