基于数据稀疏性的协同过滤推荐算法改进研究
Research on Improvement of Collaborative Filtering Recommendation Algorithm Based on Data Sparseness
作者：岳希(成都信息工程大学 软件工程学院，四川 成都 610225;软件自动生成与智能服务四川省重点实验室，四川 成都 610225)；唐聃(成都信息工程大学 软件工程学院，四川 成都 610225;软件自动生成与智能服务四川省重点实验室，四川 成都 610225)；舒红平(成都信息工程大学 软件工程学院，四川 成都 610225;软件自动生成与智能服务四川省重点实验室，四川 成都 610225)；安义文(成都信息工程大学 软件工程学院，四川 成都 610225)
Author：YUE Xi(College of Software Eng.,Chengdu Univ. of Info. Technol., Chengdu 610225, China;Automatic Software Generation and Intelligence Service Key Lab. of Sichuan Province, Chengdu 610225, China)；TANG Dan(College of Software Eng.,Chengdu Univ. of Info. Technol., Chengdu 610225, China;Automatic Software Generation and Intelligence Service Key Lab. of Sichuan Province, Chengdu 610225, China)；SHU Hongping(College of Software Eng.,Chengdu Univ. of Info. Technol., Chengdu 610225, China;Automatic Software Generation and Intelligence Service Key Lab. of Sichuan Province, Chengdu 610225, China)；AN Yiwen(College of Software Eng.,Chengdu Univ. of Info. Technol., Chengdu 610225, China)
收稿日期：2019-03-11 年卷（期）页码：2020,52(1):198-202
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：稀疏性;推荐算法;相似度;优化
Key words：sparsity;recommendation algorithm;similarity;optimization
基金项目：四川省科技厅人工智能重大专项（2019YFG0398）
中文摘要
针对根据用户的活动行为向其推荐感兴趣项目的协同过滤推荐算法，随着用户数量和项目数量增多，用户在单一项目上的活动行为减少，导致推荐质量不佳的问题，本文提出了在数据稀疏的情况下提高推荐质量的优化算法。将基于项目和基于用户的推荐方法相结合，根据用户之间的相似度初步预测用户对项目的评分，再基于项目之间的相似度产生推荐；在填补未评分的空缺值时，将平均值与预测值相结合；在计算相似度时，考虑用户之间共同评分的项目数权重和项目之间被用户共同评分的用户数权重。实验首先对比了几种基本推荐算法的推荐效果以选取较佳的基本算法进行研究，接着将本文提出的优化算法与其他算法进行了对比，最后不同程度地增加数据稀疏性进一步进行对比。结果表明：在优化算法的实验中，本文提出的优化算法一直具有较好的推荐效果；在数据稀疏性改变的实验中，随着数据稀疏度的增大，本文提出的优化算法推荐效果更具有明显优势。
英文摘要
According to the users’ activity-behaviors, the collaborative filtering recommendation algorithms recommend the interest items to users. However, as the number of users and items increase, the activity of users on a single item decreases, resulting in poor recommendation quality. To solve this problem, an optimization algorithm to improve the recommendation quality in the case of sparse data was proposed in this paper, which combines the item-based and user-based recommendation algorithm. Firstly, the score of the items was predicted according to the similarity between users preliminary; secondly, the recommendations are got based on similarities between items. The average value was combined with the predicted value when filling unrated vacancy values. The weight of the items that commonly scored by users and the weight of users that commonly scored the same items were considered. In experiment, the recommended effects of several basic recommendations algorithms were firstly compared to select a better basic for deep research. Secondly, the proposed algorithm was compared with other algorithms. Finally, for further comparison the data sparsity to several extents was increased. The experiment results showed that the optimization algorithm proposed in this paper had better recommendation. The experiment of data sparsity changes showed that as the data sparsity increases, the optimization algorithm proposed in this paper has obvious advantages.
上一条：一种Spark作业配置参数智能优化方法
【关闭】
