基于用户综合兴趣的协同过滤推荐算法研究

发布时间：2018-05-04 01:06

本文选题：推荐系统 + 协同过滤　；参考：《兰州财经大学》2017年硕士论文

【摘要】：大数据时代,互联网信息和资源严重过载,电子商务网站的商品数量和用户数量每天都在增长,人们在商品选择的过程中花费的时间和精力也越来越多。搜索引擎和推荐技术应运而生,推荐系统以其能够提供个性化服务和能够精确匹配用户需求成为当下研究的热点。从20世纪90年代推荐系统诞生以来,协同过滤推荐算法就受到了众多研究学者的青睐,而且也已经在实际应用中发挥作用,比如亚马逊网站就是利用协同过滤技术进行推荐的,而且有35%的销售额和推荐技术有关。随着网络信息越来越复杂,协同过滤技术的不足也显露了出来。本文从用户兴趣的角度分析用户行为,将用户的行为信息融入到协同过滤算法的推荐过程中,首先对用户兴趣进行分类和描述,将用户兴趣分为用户客观兴趣、用户主观兴趣和用户综合兴趣;然后分别基于用户客观兴趣模型和用户主观兴趣模型改进传统的协同过滤的算法:1)提出了基于用户客观兴趣改进的协同过滤推荐算法——OICF算法,采用模糊C均值聚类技术得到用户客观兴趣度,填充评分矩阵来改进基于项目的协同过滤算法,缓解了由于数据稀疏性带来的推荐不准确的问题;2)提出了基于用户主观兴趣改进的协同过滤推荐算法——SICF算法,通过统计的方法,构建基于评级的主观兴趣度矩阵,代替评分矩阵计算相似度,改进基于用户的协同过滤算法,改善由于用户数量和项目数量的增长,评分矩阵越来越大给计算性能造成影响的问题;3)采用两阶段的加权综合方法对改进后的算法进行组合,得到了基于用户综合兴趣改进的协同过滤推荐算法——UICF算法,推荐准确度得到一定程度的提升,并且每种改进效果均在movielens 100k数据集上进行验证;最后根据CRISP-DM过程模型的基本流程用Book-crossing数据集对改进算法进行了进一步的验证。
[Abstract]:In the big data age, the Internet information and resources are overloaded, the number of goods and the number of users of e-commerce websites are increasing every day. People spend more and more time and energy in the process of commodity selection. The search engine and recommendation technology emerge as the times require, and the recommendation system can provide personalized services and accurate matching. User demand has become a hot spot in the present research. Since the birth of the recommendation system in 1990s, collaborative filtering recommendation algorithm has been favored by many researchers, and it has also played a role in practical applications, such as Amazon Web site is recommended by collaborative filtering technology, and there are 35% sales and recommendation technology. As the network information becomes more and more complex, the shortage of collaborative filtering technology is also revealed. This paper analyzes user behavior from the angle of user interest, and integrates user's behavior information into the recommendation process of collaborative filtering algorithm. First, the user interest is classified and described, and user interest is divided into user's objective interest, and user's interest is divided into user interest. Subjective interest and user's comprehensive interest, and then based on the user objective interest model and the user's subjective interest model to improve the traditional collaborative filtering algorithm: 1) a collaborative filtering recommendation algorithm based on the user's objective interest improvement is proposed, OICF algorithm, and the fuzzy C mean clustering technique is used to obtain the user's objective interest degree and fill the scoring moment. A collaborative filtering algorithm based on project is improved to alleviate the problem of inaccurate recommendation caused by data sparsity. 2) a collaborative filtering recommendation algorithm based on user's subjective interest improvement, SICF algorithm, is proposed. By statistical method, the subjective interest matrix based on rating is constructed and the similarity degree is calculated instead of the score matrix. Into the user based collaborative filtering algorithm to improve the increasing number of users and the number of projects, the scoring matrix is becoming more and more influential to the computing performance. 3) the improved algorithm is combined with the two stage weighted synthesis method, and a collaborative filtering recommendation algorithm based on the improvement of the user's comprehensive interest - UICF calculation is obtained. Method, the recommendation accuracy is improved to a certain extent, and each improvement effect is verified on the movielens 100k data set. Finally, the improved algorithm is further verified with the Book-crossing data set based on the basic process of the CRISP-DM process model.

【学位授予单位】：兰州财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】