基于用户行为的协同过滤算法研究

发布时间：2018-05-28 07:00

本文选题：协同过滤 + 用户行为　；参考：《西北大学》2017年硕士论文

【摘要】：互联网中与日俱增的信息在给用户提供诸多方便的同时也悄然带来了“信息过载”的烦恼。如何更有效地获取信息,应对信息过载问题,个性化推荐系统已成为公认最有前途的技术发展方向之一。其中,协同过滤算法作为近年来的推荐系统热点研究方向,在web2.0应用中取得了成功。但协同过滤算法中存在的数据稀疏性、系统扩展性和推荐多样性等问题,尚有待进一步研究。本文针对上述问题,开展相关研究工作,提出了三种改进算法。主要研究工作包括:提出了基于行为相似度的算法。针对传统算法使用用户评分信息作为输入数据存在数据稀疏性问题,选取数据量更为丰富的用户行为数据作为输入,将行为相似度作为用户或项目近邻的选择依据,提出了基于行为相似度的协同过滤算法,算法中设计了基于行为序列的行为相似度计算策略和基于图的行为相似度计算策略。仿真实验结果表明该算法可以缓解数据稀疏性的负面影响,提高推荐结果的质量,体现了算法的有效性。研究了基于行为影响力的算法。在实际应用中,用户并不总是希望系统推荐相似的项目,有时更需要一些有关联性的推荐项目。挖掘用户行为间的影响力,利用影响力来度量用户间或项目间的关联关系,以此作为最终预测和推荐的基础具有可行性,并在仿真实验结果中得到了验证。提出了具有时序多样性的协同过滤优化算法。推荐列表的多样性对提升用户体验有着积极和重要的影响。在前两个算法的基础上,引入用户情境,对推荐结果进行时序多样性优化。仿真实验结果表明该算法在保持推荐质量的同时,显著提高推荐结果的多样性。
[Abstract]:The increasing information in the Internet not only provides users with many convenience, but also quietly brings "information overload" trouble. How to obtain information more effectively and deal with information overload, personalized recommendation system has become one of the most promising technology development direction. Among them, collaborative filtering algorithm, as a hot research direction of recommendation system in recent years, has been successful in web2.0 application. However, the problems of data sparsity, system expansibility and recommendation diversity in collaborative filtering algorithms need to be further studied. In this paper, three improved algorithms are proposed to solve the above problems. The main research work includes: the algorithm based on behavior similarity is proposed. In order to solve the problem of data sparsity in traditional algorithms which use user scoring information as input data, the user behavior data with more abundant data is selected as input, and the behavior similarity is used as the basis for the selection of users or project neighbors. A collaborative filtering algorithm based on behavioral similarity is proposed. A behavioral similarity calculation strategy based on behavior sequence and a behavior similarity calculation strategy based on graph are designed in the algorithm. Simulation results show that the algorithm can alleviate the negative impact of data sparsity, improve the quality of the recommended results, and reflect the effectiveness of the algorithm. The algorithm based on behavioral influence is studied. In practical applications, users do not always want the system to recommend similar items, sometimes they need some related items. It is feasible to mine the influence of user behavior and use it to measure the relationship between users or projects. It is feasible to use it as the basis of final prediction and recommendation, and it is verified in the simulation results. A collaborative filtering optimization algorithm with time series diversity is proposed. The diversity of recommendation lists has a positive and important impact on improving the user experience. On the basis of the first two algorithms, the user situation is introduced to optimize the timing diversity of the recommended results. Simulation results show that the proposed algorithm can significantly improve the diversity of recommendation results while maintaining the quality of recommendations.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】