基于GLSLIM模型的混合推荐算法研究

Research on Hybrid Recommendation Algorithm based on GLSLIM Model

GLSLIM模型(Global and Local SLIM)是基于SLIM模型(Sparse Linear Methods)并优于SLIM模型的推荐算法。它将GLOBAL模型和LOCAL模型结合,GLOBAL模型用来捕获物品在所有用户之间的相似性,LOCAL模型用来获取物品在某个用户子集中的相似性,通过两种模型的结合可以进一步优化用户的个性化推荐。但该算法存在天然缺陷,就是被用户评价或购买过的物品之间的相似度才可以被学习到,没有被购买过的物品之间的相似度为0。这将导致用户购买过的相似物品才有机会被推荐,相似度为0的物品无法推荐给用户。为了改善这种情况,利用一种PLSA模型解决这个问题,基于两种模型的组合进行协同过滤推荐。实验结果表明,虽然推荐结果的准确性略微降低,但是能挖掘用户的潜在兴趣。

The GLSLIM model(Global and Local SLIM)is based on the SLIM model(Sparse Linear Methods)and better than the SLIM model.It combines the GLOBAL model with the LOCAL model that the GLOBAL model is used to capture the similarity between items in all users and the LOCAL model is used to obtain the similarity of items in a subset of users.The combination of the two models can further improve the user personalized recommendation.However,the natural defect of the algorithm is that the similarity between the items evaluated by the user or purchased can be learned,and the similarity between the items that have not been purchased is zero.In order to improve this situation,the new recommendation algorithm that based on the combination of the two models(GLSLIM and PLSA)for collaborative filtering recommendations.Experimental results show that although the accuracy of the recommended results is slightly reduced,but it can tap the user's potential interest.