Research On Matrix Decomposition Recommendation Algorithm Based On Harmonic User Entropy Weight

The traditional recommendation algorithms do not consider the authenticity of the score in the score matrix or the real recommendation contribution ability of the neighbor users to the target users in the selection of the nearest neighbor,which leads to the deviation in the calculation of similarity between users.Based on these problems,proposes time interval weight and user entropy weight similarity are proposed.Firstly,obtains the time information of user rating.Based on the average value of all rating intervals,the time interval of rating which is too long or too short is given a lower time interval weight to obtain a more objective and real rating matrix.Then,calculates the information entropy of user rating to reflect the user rating distribution,and calculates the user entropy weight of reconciliation.Then,combines the user entropy weight with the traditional method of user similarity calculation to obtain a more authentic and accurate user entropy weight similarity of reconciliation.Then,in order to improve the sparsity of the scoring matrix,the Funk SVD algorithm is integrated with the similarity of user entropy weight as a constraint factor to further reduce the error of the recommendation results.Finally,aiming at the problem of over-fitting caused by too many dimensions decomposed by Funk SVD algorithm and the problem of poor expressiveness caused by calculation of prediction score by dot product,a decomposition algorithm based on distance matrix is proposed.The algorithm transforms the scoring matrix into the distance matrix,then decomposes the distance matrix with the harmonic similarity of user entropy weight as the constraint factor to obtain the predicted distance between users and items.In the Movielens data set on the experimental results show that the interval weight gives the original user items after scoring matrix,the error of recommendation results is significantly reduced.In addition,when the decomposition dimension is too large,the accuracy and recall rate of the recommendation results obtained by the matrix decomposition recommendation algorithm based on distance matrix are relatively high compared with the matrix decomposition recommendation algorithm based on the traditional user-item scoring matrix under the same conditions.Finally,compared with other traditional matrix decomposition recommendation algorithms,the errors of the proposed two matrix decomposition algorithms are reduced.