Research On Recommendation Algorithm Based On Comprehensive Feature Clustering And Time Factor

With the rapid development of web2.0,the amount of information on the Internet is growing rapidly today,the problem of information overload is difficult to avoid,and it bothers users for their quick access to their valuable information.Personalized recommendation technology has emerged and gradually become an effective tool to alleviate this problem;it can help users quickly get the information they need from the massive information resources.As the most important part of the implementation of personalized recommendation,the recommendation algorithm has many different types according to the specific characteristics;the collaborative filtering algorithm has become one of the most widely used recommended algorithms because of its many excellent characteristics.Influenced by the sparse of the rating data and the user's interest migration problem,the accuracy of the collaborative filtering algorithm is not high.In order to mitigate the effects of the above problems,the work of this paper is:(1)Because the user's rating data is limited,the result of the traditional similarity calculation method is not accurate enough,in this paper;an improved similarity calculation method is proposed to improve the accuracy of similarity calculation between items.Considering the amount of item characteristics and the age of rating users data is pretty sufficient and stable,combines the two aspects of the data with the score data,avoiding the only use of sparse rating data for similarity calculation leads to inaccuracy of the calculation results.(2)Through further analysis have found,the weight value inaccurate caused by traditional weight function which simulates the attenuation of user interest with time that set the same interest attenuation rate for different users.Aim at this problem,in this paper the user's age factor is considered into the weight function,Setting interest attenuation rates that conform to user's age feature for users of different age,to give user rating a more reasonable weight value.The algorithm proposed in this paper first clustering the items according to the improved similarity calculation formula and search the nearest neighbors,then in the rating prediction stage,lead the weight function that take user's age factor into consideration into the rating prediction formula.Improving the accuracy of the final recommendation results.The experiment section is carrying out the contrast experiment between the algorithms that is proposed in this paper and traditional algorithm.The results show the improved algorithm in this paper can alleviate the problem of sparse of rating data and the influence of user interest migration,the prediction accuracy of the algorithm has been improved.