A Matrix Decomposition Recommendation Algorithm Introducing Untrusted Information between Users

In response to the problems of poor recommendation performance caused by sparse data in traditional recommendation algorithms and slow convergence speed caused by the fact that the values of trust matrix elements can be any real number, this paper proposes a non negative matrix decomposition algorithm based on trust and non trust information, which comprehensively utilizes user historical information data and trust and non trust information between users. In the algorithm design, non negative constraints were applied to the potential factors of the rating matrix and trust matrix, while negative constraints were applied to the potential factors of the distrust matrix; At the same time, a novel trust regularization and distrust regularization method was also designed for model training. Experiments have shown that compared with SVD++ and TrustSVD, this algorithm not only overcomes the impact of sparse traditional data on recommendation quality to a certain extent, but also effectively enhances the robustness of the algorithm, greatly improving the accuracy of recommendations.