Recommendation Algorithm Based On Mean-segmentation And Word2vec

With the diversification of the recommended content,explicit feedback is no longer able to meet increasingly complex recommendations,and it is necessary to use implicit feedback to enhance the applicability of the recommendation system.However,implicit feedback has problems such as difficult to understand user preferences and lack of negative feedback,using the matrix factorization recommendation algorithm directly will result in prediction results don't have negative samples leading to the recommendation system doesn't have the ability to judge.In addition,the matrix factorization recommendation algorithm faces the problem of data sparseness.This paper has carried on the thorough research to these problems,on this basis raises own improved method.First,for the problem of lack of negative feedback in implicit feedback,the mean-segmentation matrix factorization recommendation algorithm(MSMF)is proposed to solve this problem.Considering that each user will have likes and dislikes,this paper proposes a mean-segmentation method to solve the problem of lack of negative feedback and uses this method to improve the matrix factorization recommendation algorithm.The improved matrix factorization recommendation algorithm can obtain the prediction results including positive samples and negative samples through a training set including only positive feedback,the recommendation system can determine from the prediction results what the user likes and dislikes.Secondly,for the problem that data sparsity leads to inaccurate prediction results,this paper introduces the word2 vec matrix factorization recommendation algorithm(wMF).This paper introduces word2 vec technology and applies it in implicit feedback to calculate the similarity to improve the efficiency of the similarity calculation method.And use the obtained similarity to fill the scoring matrix,so as to achieve the purpose of reducing the sparsity of the scoring matrix.Finally,using two real-world data sets to experiment with the two recommendation algorithms proposed in this paper.Experimental results show that the two recommendation algorithms proposed in this paper can solve the problem of lack of negative feedback in implicit feedback,reduce the sparsity of data,and improve the accuracy of the prediction results.