Personalized Recommendation Of Mobile Data Services Bansed On Social Relationship

Telecommunication-based personalized recommendations emerged because of the diversity of mobile data services and the needs of telecom users. However, the consumption behaviors of telecom users can be influenced by social ties between them, at the same time, due to the larger number of mobile data services, the user-product matrix will be very sparse. The traditional recommendation algorithm needs to be improved targeted on these areas. Based on all above, the main research contents and results of this paper are as follows:1. The relationship between consumers'behaviors and their contacts between each other has been proved by data experiment. Two hypotheses are verified:(1) Two users with a certain social ties will have similar data services consumer preferences; (2) the stronger their social ties are, the higher their similarity of behaviors will be.Two hypotheses are firstly proposed through the summary and analysis of previous studies and then this paper design series of statistical experiment to verify them. Main idea of the experiment is to compare the similarity of consumption behaviors between the user-pair with and without social ties using method of Hypothesis Testing. The final results indicate that there is significant effect between users'social connections and their consumer behaviors and the two users with social contacts are more likely to have similar data-services-consumption behaviors. Correlation analysis results between social contact strength (calling frequency) and the similarity of their consumption behaviors showed that the two variables have a strong positive correlation. As a result, two assumptions have been confirmed. 2. This study proposed a method to measure strength of social ties based on the call records of telecom users and then designed a vacancy filling method for user-product matrix.To take advantage of social connections between users for their consumption behavior to achieve the recommendation method design for data services, this paper exploit the attributes of users' calling record (such as call start time, call duration, call type, etc.) and their statistics information to construct a more accurate method of measuring strength of social ties between telecom users. Based on this method and the research results from the first part, a vacancy filling method for user-product matrix is designed and in our experimental projects the sparsity of the user-product matrix was reduced from the 91.87% to 80.21%, which achieves a good filling effect.3. A new recommendation algorithm based on user'social ties (Social-CF) have been proposed. The data experiment proves that the new algorithm is better than the classic one (collaborative filtering) in predicting accuracy and recommendation effect.Based on the vacancies filling method, this research proposes integrated similarity calculation method, which improves the filtering rules of user-neighbors. In order to ensure the integrity of the recommendation sets, the concept of major set and subsidiary set is given to meet the potential preferences and needs of users. In this experiment, 1464 users' consumption behaviors data have been used and the value of parameterαin integrated similarity is confirmed through experiment. After that, this paper compares Social-CF with best parameter value and classic CF and the results shows that the proposed algorithm has certain advantages in predicting accuracy and recommendation effects (maximum of recall rate:78.37%, higher than 69.21% of classic CF; accuracy:33.99%, classic CF:30.46%; minimum of MAE is 0.136 slightly lower than 0.146 of classic CF).