Research On Personalized Search Of Social Network Cross Media Big Data Based On User Portrait

Social networks contain a large amount of cross-media data,which is extremely valuable in the business and security fields.Cross-media search needs have emerged as the times require.However,traditional single text and image analysis methods cannot complete cross-media search tasks.At the same time,in recent years,the demand for personalized search has been increasing,and users hope to obtain search results that they are interested in.Therefore,using deep learning methods to extract cross-media big data features of social networks,constructing a multi-modal public semantic space,and constructing user portraits by mining user social network information are of great significance for personalized search of cross-media data.Four tasks completed in this article are as follows:(1)Group intelligence perception of cross-media big data content in online social networks.Aiming at the cross-media big data of social networks,we use web crawlers for crawling and design a perceptual recognition model to filter the captured text and image data.Elimination of stop punctuation,simplification conversion,and word segmentation processing are performed on text data,and image data is cropped,scaled and extracted feature by VGG19 network operations are performed.The processed data is stored in a relational database or a non-relational database according to characteristics.(2)The cross-media big data semantic analysis based on adversarial generative network.A social media cross-modal search method(SSACR)based on semantic similarity and adversarial learning is proposed.Adversarial learning is used to train feature projection network and modal discrimination network,making the data of the same modal with different semantics farther in the common semantic space,while the data of the same semantics with different modals closer.And we introduce the concept of semantic similarity to eliminate the impact of modalities on the data.Experimental results show that cross-media search indicators are superior to other comparison algorithms.(3)The establishment of user portraits and the search of cross-media big data.A user relationships based multi-dimensional user portrait construction algorithm(RMDUP)is proposed.The user's Weibo text is analyzed to predict the user's basic attributes,and the basic attributes are modified according to the user's following and followed relationship.At the same time,a user portraits based cross-modal retrieval algorithm(UPCMR)is proposed,which utilizes the characteristics of users with different attributes to pay different degree of attention to the same Weibo to achieve personalized search results ranking.The experimental results show that the use of user portraits has a positive impact on search results,and greatly improves the map value of search.(4)Design and implementation of cross-media big data personalized search system of social media based on user portraits.The system includes three major modules:user attribute prediction result,text search image and image search text.An interactive system with friendly interface and convenient operation was developed.While verifying the various algorithms proposed in this thesis,the correctness,real-time and validity of the system are verified as well.This thesis proposed a cross-media search algorithm based on semantic similarity and adversarial learning,combined with user portraits to achieve personalized search,which has certain practical value.