Research On Key Technologies Of Homology Analysis On Large Scale Video Data

In recent years,with people's increasingly dependent on the network video,social networking way gradually shift from words and images to video.If we use homologous analysis techniques to find records in different video social networks corresponding to the same real world entity.At the commercial level,merchants can make accurate video recommendations,advertising recommendations and friend recommendations to users through users' behaviors in different video social networks.At the same time,it can avoid the repeated push of commercial advertisements to the same users;At the user level,user can find people who have unauthorized access to publish user's vedio in the same or different video social network through the homology analysis technology,and protect legal rights;At the security level,some illegal user publish spam videos through different account to earn profit,the operator can use homologous analysis techniques to locate all the related accounts and quickly delete them.In view of the particularity and data characteristics of video social network,this paper makes research as follow:(1)This thesis propose video social networking homology analysis model.Video social networking homology analysis model can be divided into four steps: build video social network diagram,video similarity preprocessing,user similarity calculation and results pruning.(2)The construction method of video social network is proposed,and the algorithm of each step in the video homology analysis model was optimized.It mainly includes the distributed video similarity preprocessing algorithm based on the graph structure,which can obtains the dynamic video similarity,so as to better reflect the similarity between different videos.The user similarity calculation method based on video weight was introduced to judge the similarity of users and improve the accuracy of user similarity calculation by weighting video similarity.Using the result pruning based on the user relationship,pruning is done to retrieved high relation-similarity users to remove the noise users.(3)In this thesis,we implements a distributed video download system to crawl information from two video social networks that have similar personas.Then we analyses and preprocess the video social network data information.Finally,we make experiments on the data set we crawled to verify the performance of the homologyanalysis method we propose.