Multi-Information Model And Recommendation Technology Research

Society structure characterize the partial aggregation of the edge connection relationship in the network. The society in the network is usually consist of functional similar or property similar node. Detecting community structure in the network is helpful for uncover the structure and functional of network relationship.The early research of detecting community in the network was mainly concentrated in these media such as individuals’ blogs or emails. With the web2.0’s flourishing, the social network service such as Facebook, Twitter, Weibo and RenRen network have cumulate many users’ personal data. Detecting the community structure in the social network service is of significant commercial value thus have attracted lots of researchers who devote their energies in researching the network community structure. However, most of these research are based on the single data source to construct models that numerous data was not used. In this paper based on the modularity maximization algorithm, the friends relationship, location information of user’s profile and user’s tags information was used which was crawled through the Weibo’s API to find community structure of Weibo user network by apply the spectral clustering method. All these user information dimensionality was reduced by means of tensor decomposition. The clustering coefficient is used to evaluate cluster quality and used a single source information of user to find community structure to compare the experiment result which is conducted by multi-information source.Currently, the majority collaborative filter technology are mainly based on user, item and the combination of user and item. With the development of network and user’s interest drifting, the interest tag of user and the location data is becoming a new vital factor in the recommendation technology research. Apply the user’s friend relationship, interesting tags and location data for personalized recommendation is of significant value. In this paper, user’s friend relationship, interesting tags and location data are used to construct an original tensor, with the higher order singular value decomposition technology fitting the result tensor which could be used to recommendation. Finally, the recall rate, the accurate rate and the value of F are used to evaluate the recommendation quality.