Aiming at the fact that the traditional collaborative filtering algorithm cannot deeply mine user relationship and recommend new items to users, a Trust and Item Preference Collaborative Filtering (TIPCF) recommendation algorithm was proposed. Firstly, in order to mine the latent trust relationship of the users, the user reliability was gotten and the trust degree between users was quantified by analyzing user ratings. Secondly, by considering that the difference of users' preference for different target items has an effect on user similarity, user preference was added to the traditional user similarity algorithm to improve the similarity algorithm. Thirdly, the choice of nearest neighbor set was more accurate by incorporating user reliability and improved similarity. Finally, the users' preference on item attribute was used to recommend new items. Experimental results show that, compared with traditional collaborative algorithm, the Mean Absolute Error (MAE) of TIPCF was decreased by 6.7%, and the MAE of TIPCF was decreased by 10.7% when recommending new items on the Movielens dataset. TIPCF not only improves the accuracy of recommendation, but also increases the recommended probablity of new items.

In order to overcome the disadvantages of the K-Means Clustering (KMC) algorithm, such as the poor global search ability, being sensitive to initial cluster centroid, as well as the initial random, being vulnerable to trap in local optima and the slow convergence velocity in later period of the original Artificial Bee Colony (ABC) algorithm, an Improved ABC (IABC) algorithm was proposed. IABC algorithm adopted the max-min distance product algorithm for initial bee colony to form a fitness function, which is adapted to the KMC algorithm, and a position updating method based on the global leading to enhance the efficiency of the iterative optimization process. The combination of the IABC and KMC (IABC-Kmeans) would improve the efficiency of clustering. The simulation experiments were conducted on the four standard test functions including Sphere, Rastrigin, Rosenbrock and Griewank and the UCI standard data sets. The experimental results indicate that IABC algorithm has a fast convergence speed, and overcomes the defect of the original algorithm being easily falling into local optimal solution; IABC-Kmeans has better clustering quality and general performance.