A Link Prediction Method Toward Cooperative Relationship In Heterogeneous Information Network

The structure of the real network has been much more complex and more information with the arrival of big data era.As an important research direction of data mining,the research object of link prediction has begun to turn from homogeneity information network to heterogeneous information network.Link prediction method toward cooperative relationship in heterogeneous information network can improve the accuracy and prediction efficiency of the cooperative relationship link prediction by using the rich information in the network.It plays a valuable guiding role for the cooperative relations in the academic,literary and commercial aspects in the reality and has great significance for research.Most of the existing cooperative relation prediction methods are applicable for specific networks,which fail to make full use of the rich semantics of heterogeneous networks and the timeliness of cooperative relations.This paper proposes a link prediction method toward cooperative relationship in heterogeneous information network,based on the meta-path of heterogeneous information network,improving the existing supervision learning prediction algorithm from two angles of feature extraction and classification algorithm.Firstly,according to the schema of cooperative network,all possible meta-paths and all instances of meta-paths between user pairs can be generated.Secondly take the attribute of the meta-path of user pair in the heterogeneous information network as the attribute of the machine learning sample data.Link entropy based link probability fraction is added into consideration on the basis of the feature of meta-path topology to improve the meta-path synthesis feature.Meanwhile,in view of the timeliness of the cooperative relationship,also take the user's timeliness characteristics,user's stage activity,as one of sample properties.Finally,combined with the idea of ensemble learning,improved randomized forest classification algorithm is used to build prediction model.And under-sampling method based on fast clustering is chosen to solve the class imbalance problem of cooperative relationship.The parameters of random forest need to be Adjusted to improve the prediction accuracy and reduce the running time at the same time.To evaluate the performance of the prediction method,three real cooperative heterogeneous information network data sets DBLP,Movielens,and the video data of Douban are selected.Compared with the prediction results of the traditional heterogeneous information network link prediction method,the improved method proposed in this paper guarantees the accuracy of the forecast while improving the recall rate.The results showed that the proposed link prediction method toward cooperative relationship in heterogeneous information network is more accurate.