The Research On Key Technologies Of Online Social Recommendation System

Social network makes communication easier than before, in which recommendation system makes the relation between items develop better. The extraction and effective use of feature vector is still major problem in recommendation system, especially in the context of online network. The large sum of data makes the feature matrix very huge, costs more computation. How to develop a recommendation algorithm using the large data effectively has become one of the forefront topics in data mining field.Recommendation algorithm based on Factorization Machines model is suitable for online network scenarios, because it is capable of handling large data problem in recommendation system. By making use of Block Structure, FM model can reduce redundancy of feature matrix. However, it lack of uniform method for constructing Block Structure and mapping function. In this paper, the proposed "Index Mapping Method" can get both Block Structure and its mapping function quickly with the help of database technology, therefore it save more computation resource and consume less time.Main effort has been taken in those four respects:1. Study Block Structure systematically. Using Block Structure in FM model can greatly reduce algorithm space complexity, in other words, we spend less time in computation. In order to research on the degree of optimization on recommendation algorithm, several comparison experiments between FM model and FM-BS model were designed.2. Extract item feature with the aid of database technology. When design database structure it can get the feature block structure for FM model at the same time. This procedure also lays the foundation for role analysis of different feature vectors in FM models.3. Propose Index Mapping Method. This paper replace repeating model in feature matrix of corresponding index which is similar to concept of ’index’ in database system. This method should construct block structure and mapping function quickly.4. Gives the general design structure for recommendation algorithm. This paper shows the general design structure for recommendation algorithm firstly, evaluate the experiment based on the actual situations later.