Abstract:
Aiming at the important part of user behavior analysis in personalized recommendation system, this paper takes user value and preference habits as the starting point. By extracting the implicit information in user history records, a new analysis method of TV user viewing behavior is proposed. The user's overall average operating frequency, the recent average operating frequency, the operating frequency change ratio, and the most recent operating time interval are extracted as the value model indicators, and the classification results of eight different user value groups including loyalty, new life, and loss are proposed according to the characteristic indicators; For 12 categories, a user preference habit feature matrix based on activity and stability is constructed, and user preference habits are divided into 4 cluster classes to quantify and analyze user preferences. Experiments on 361 459 pieces of recorded data generated by 1 025 users show that the method can effectively analyze the viewing behavior of any TV users and improve the accuracy of the preference analysis results.