| Personalized recommendation system was originally used in Internet shopping in Amazon, thesite’s customers’ loyalty improved and enterprise’s sales increased, by now in Internet it is used inother areas by more and more, and its attention increasingly higher. Based on the analysis of theuser’s interests the personalized recommendation system recommends items to users. Collaborativefiltering algorithm is one of the most widely used recommendation technology. However，with theincreasing amount of data in recommender systems, collaborative filtering algorithm has exposedsome bottleneck problem, such as data sparse, cold start, extensibility and real-time problem.Based on the study of the collaborative filtering algorithm this thesis proposes an improvedcollaborative filtering algorithm with the interest changes of the user-project. The improvedalgorithm considers forgotten laws of human beings, introduces a nonlinear time forgetting function,gives users’ evaluation in different periods in different weights, and highlights the importance of therecent evaluation in recommended results. On the basis of the concept hierarchy, it sets project scorethreshold, reduces the user-project score matrix, and establishes a candidate data set in order toimprove the scalability of recommendation algorithm. According to the user’s interests to fill the data,it balances the user interest preferences, score matrix and project objective attributes. Predictionscore values can match the user’s real interests, effectively reduce the data sparseness and cold startproblems.This thesis uses the MovieLens data set to take an experiment that the improved algorithm iscompared with the traditional collaborative filtering algorithm. The experimental results show thatthe improved collaborative filtering algorithm better than the traditional collaborative filteringalgorithm in the accuracy and completeness, it achieves the purpose of improvement, which caneffectively improve the quality of the recommendation results. |
PDF Full Text Request