Research On User Characteristics Based Collaborative Filtering Recommendation Algorithm

The rapid development of the Internet makes the massive growth of Internet information. Userneeds to coordinate information according to their own characteristics accurately, so thepersonalized recommendation technology appeared as the circumstances required. This paperconducts an intensive research on collaborative filtering algorithm and proposes some improvedschemes based on traditional collaborative filtering, which aims to explore the solutions to improvethe recommendation accuracy and alleviate the existing problems of traditional algorithm. Themain work of this paper is as follows:(1) Adding user feature factor and time scene factor into traditional collaborative filteringrecommendation algorithm. Firstly, we pick up the hours data from the dataset time stamp andclassify the dataset according to the daily time. Then we calculate similarity of users using theclassified data with the factors including historical rating data and user features. Adding userfeature factor can alleviate the cold start and data sparsity problems of traditional algorithm, andadding time scene can meet the demands of users.(2) Adding user trust factors into collaborative filtering recommendation algorithm. Traditionalcollaborative filtering’s prediction method is based on weighted average, which the weight is usersimilarity. This paper considers the user trust as weight at the same time, which the trust is thequantity of each user’s ratings. Adding the user trust factor makes the prediction more accurate.(3) Collaborative filtering by adding user feature factor is based on using historical rating dataand user features to calculate similarity, which are combined by a certain weight. Reasonableweight value guarantees optimal recommendation results. we researched the law of weight valuethrough experiment.(4) By doing some experiments on the Movielens dataset, the experimental results show thatthe improved recommendation algorithms is more accurate than traditional method. We also gavesome suggestions of factor and weight value selection.