Jointly Recommendation Algorithm of KNN Matrix Factorization with Weights

Two improved algorithms based on k Nearest Neighbor Matrix Factorization algorithm were proposed to solve the problem of predicting negative score in k-nearest neighbor matrix Factorization algorithm. First, KMF + algorithm constructs the Nearest Neighbor matrix and dissolves it to obtain the corresponding user's factor matrix and item's factor matrix. Secondly, the score prediction model is established by user's factor matrix and item's factor matrix, and the factor matrix is optimized by Matrix Factorization algorithm. Finally, the predicted score value of the target users to the target project is calculated. KMFwS algorithm is improved on the basis of KMF + algorithm, and the influence of KMFwS algorithm on the predicted score value through weight constraint when the nearest neighbor matrix is zero matrix. The simulation results on data sets and a real data set show that KMF + algorithm effectively solves the problem that the score value is negative and keeps the score value well constrained between 0 and 5. Meanwhile, KMFwS algorithm obtains more accurate score results than KMF + algorithm by avoiding the error caused by zero neighbor matrix to the score value.