A Collaborative Filtering Algorithm Based On Recommendation Weight And Dynamic Reliable Neighbors

This paper first discusses the background and significance of the recommendation system,and introduces the the structure and logic module of recommendation system,then compares the characteristics of the several popular recommended techniques. To sovle the common problems of recommendation system, we put forward some improvement ideas,and design some experiments to verify the algorithm.Our main work is as follows:1)The traditional collaborative filtering algorithm is based on K-nearest neighbors. However, in order to meet the requirements on the number, K nearest neighbors will also select some items with low similarity,which will affect the recommendation accuracy directly. Therefore, we need to define two similarity thresholds to select high quality neighbors. One is for similarity calculation between users and another is for the similarity computation of items. while a similarity between the object and the target object is greater than the threshold,the object can be selected as neighbors.2)Since the Pearson calculation method has several disadvantages when the intersection of two sets is very small. So using Pearson calculation method to calculated similarity will be a great error in this case, so we propose the concept called recommendation weight, the more the intersection of the two projects, the higher the weight.3)We set a intersection threshold for the neighbors of items and users, when the intersection of an object with the target object is less than this threshold, then the similarity calculation is considered unreliable. By calculating the proportion of unreliable neighbors of both items and users, we can select a more reliable list of neighbors to predict the scores.4)In order to verify the validity of the algorithms this paper proposed, severaltests are designed base on the Movie Lens dataset.