基于共享知识模型的跨领域推荐算法

Cross-Domain Recommendation Algorithm Based on Sharing Knowledge Pattern

互联网的普及使得大量信息不断累积,推荐系统作为解决信息过载的有效手段,能够帮助人们迅速准确地筛选出感兴趣的内容.但是由于用户项目评分数据过于稀疏,新用户或新商品存在"冷启动"问题,使得传统的推荐算法计算复杂性过高、准确性较低.考虑到用户会在互联网不同领域使用各类应用,在不同领域积累了大量行为数据和评价信息.而从用户群体的角度来说,在不同领域间存在着用户群体的偏好相似性,因此如果通过在不同领域中共享代表偏好的知识模型,将有助于提升在新领域推荐的准确性,解决冷启动问题.本文提出了基于共享知识模型的跨领域推荐算法SKP(Sharing Knowledge Pattern),通过对各个领域中用户-项目的评分矩阵分解,得到用户的潜在特征矩阵和项目的潜在特征矩阵,对用户和项目的潜在特征分别聚类,得到了用户分组对项目分组的评分知识模型,最终利用目标领域的个性知识模型和各个领域的共性知识模型来得出推荐结果.本文对三个不同领域的数据集进行了分析和划分,并在物理集群环境下进行了实验.结果表明,通过利用数据稠密的辅助领域数据,本文提出的SKP算法与已有的单领域算法、跨领域算法相比,具有更高的准确率和更低的RMSE值.

With the popularity of the Internet and the accumulation of large amounts of data,recommendation system,as an effective means to solve the problem of information overload,can help people quickly select what they are interested in.Because of the sparse user-item rating data,and the cold start problem of new users or new items,traditional recommendation algorithm has the shortcoming of high complexity,low accuracy.Considering the accumulated users behavior or rating data across different domains can have the same preferences,we can share the knowledge pattern among different domains.Based on the matrix factorizationof user-itemrating data in different domains,we can obtain the latent feature matrix of users and items respectively.Considering the user group preference,the latent features of users and items are clustered separately as the domainknowledge pattern.Moreover,By clustering the cross-domain knowledge patterns,we can get shared common knowledge pattern.With the domain knowledge pattern and the shared common knowledge pattern,we can make the finalrecommendation.Based on the above consideration,this paper proposes the SKP(Sharing Knowledge Pattern)algorithm.And the SKP is realized in a parallel manner.Experimentsare carried out in the physical cluster environment.By exploiting three different datasets,the results show that the SKP algorithm has better recommendation accuracy and lower RMSE values compared with the existing single-domain algorithm and other cross-domainalgorithms.