Matrix factorization （MF） has been proved to be a very effective technique for collaborative filtering （ CF）, and hence has been widely adopted in today＇s recommender systems, Yet due to its lack of consideration...Matrix factorization （MF） has been proved to be a very effective technique for collaborative filtering （ CF）, and hence has been widely adopted in today＇s recommender systems, Yet due to its lack of consideration of the users＇ and items＇ local structures, the recommendation accuracy is not fully satisfied. By taking the trusts among users＇ and between items＇ effect on rating information into consideration, trust-aware recommendation systems （TARS） made a relatively good performance. In this paper, a method of incorporating trust into MF was proposed by building user-based and item-based implicit trust network under different contexts and implementing two implicit trust-based context-aware MF （]TMF） models. Experimental results proved the effectiveness of the methods.展开更多
Aiming at the problem that the data in the user rating matrix is missing and the importance of implicit trust between users is ignored when using the TrustSVD model to fill it,this paper proposes a recommendation algo...Aiming at the problem that the data in the user rating matrix is missing and the importance of implicit trust between users is ignored when using the TrustSVD model to fill it,this paper proposes a recommendation algorithm based on TrustSVD++and XGBoost.Firstly,the explicit trust and implicit trust were introduced into the SVD++model to construct the TrustSVD++model.Secondly,considering that there is much data in the interaction matrix after filling,which may lead to a rather complex calculation process,the K-means algorithm is introduced to cluster and extract user and item features at the same time.Then,in order to improve the accuracy of rating prediction for target users,an XGBoost model is proposed to train user and item features,and finally,it is verified on the data sets MovieLens-1M and MovieLens-100k.Experiments show that compared with the SVD++model and the recommendation algorithm without XGBoost model training,the proposed algorithm has the RMSE value reduced by 2.9%and the MAE value reduced by 3%.展开更多
文摘Matrix factorization （MF） has been proved to be a very effective technique for collaborative filtering （ CF）, and hence has been widely adopted in today＇s recommender systems, Yet due to its lack of consideration of the users＇ and items＇ local structures, the recommendation accuracy is not fully satisfied. By taking the trusts among users＇ and between items＇ effect on rating information into consideration, trust-aware recommendation systems （TARS） made a relatively good performance. In this paper, a method of incorporating trust into MF was proposed by building user-based and item-based implicit trust network under different contexts and implementing two implicit trust-based context-aware MF （]TMF） models. Experimental results proved the effectiveness of the methods.
基金Guangdong Science and Technology University Young Projects(GKY-2023KYQNK-1 and GKY-2023KYQNK-10)Guangdong Provincial Key Discipline Research Capacity Improvement Project(2022ZDJS147)。
文摘Aiming at the problem that the data in the user rating matrix is missing and the importance of implicit trust between users is ignored when using the TrustSVD model to fill it,this paper proposes a recommendation algorithm based on TrustSVD++and XGBoost.Firstly,the explicit trust and implicit trust were introduced into the SVD++model to construct the TrustSVD++model.Secondly,considering that there is much data in the interaction matrix after filling,which may lead to a rather complex calculation process,the K-means algorithm is introduced to cluster and extract user and item features at the same time.Then,in order to improve the accuracy of rating prediction for target users,an XGBoost model is proposed to train user and item features,and finally,it is verified on the data sets MovieLens-1M and MovieLens-100k.Experiments show that compared with the SVD++model and the recommendation algorithm without XGBoost model training,the proposed algorithm has the RMSE value reduced by 2.9%and the MAE value reduced by 3%.
文摘为解决传统协同过滤算法中存在的数据稀疏与冷启动问题,社会化信任推荐机制被引入推荐系统,通过加入用户的显式信任信息,可有效地缓解上述问题。但是显式信任较难获取,并且数据较为稀疏,为了更好地提高推荐效率,在基于显式信任的Trust SVD算法的基础上,加入用户的隐式信任信息,提出了一种基于双信任机制的奇异值分解(singular value decomposition,SVD)算法EITrust SVD。在利用显式信任获得可靠推荐的同时,通过隐式信任的影响获得与用户喜好相关的推荐。通过实验证明,该方法可以较好地解决冷启动问题,且能提高推荐的准确率。
