Cross-media Big Data Evolution Mining,Search And Visualization Of Social Networks

Social networks act as intermediaries in the process of people’s communication and sharing.Not only can they provide users with high-quality services,but also social network platforms can collect user-generated data such as text,images,and videos.Although social network cross-media big data has great monetization value for information flow companies,the data edited by users often has the disadvantages of semantic sparseness,noise and ambiguity.This thesis is based on deep learning,pre-trained technologies,clustering algorithms,generative adversarial learning and other knowledge,it uses the social network big data semantic expansion algorithm to extract features of social network big data,and maps social network cross-media big data features to a unified public semantic space through an adversarial generation network and parameter sharing mechanism,then introduce the association relationship between social network cross-media big data to assist the acceleration of social network cross-media big data search,and realize the social network cross-media big data evolution mining,search and visualization system.The main work completed in this article is as follows:(1)Aiming at the problems of semantic sparseness,ambiguity,and noise in social network big data,short text semantic expansion algorithm(BBLA)and social network image semantic expansion algorithm(VMA)of social network are proposed.In terms of social network short text semantic expansion,a deep neural network suitable for social network short text data was designed.By combining the pre-trained model Bert,long and shortterm recurrent network and attention mechanism,the text semantic expansion vector with context semantic information is obtained.Then BBLA uses supervised methods to train and learn text data,and implements the semantic expansion of social network text.In terms of social network image semantic expansion,the VGG-16 model is used to obtain social network image data features.By introducing a multi-head attention mechanism in the process of extracting image features,the image semantic expansion vector with rich semantics is obtained,and implements the semantic expansion of social network image.(2)In order to accelerate the search efficiency of social network cross-media big data,a cross-media big data association relationship discovery algorithm(ADSTR)based on time and forwarding relationship is proposed.Through the cluster analysis of social network cross-media big data,the association relationship between social network data is calculated,and the association matrix of data within and between media is constructed to realize the discovery of the association relationship between social network cross-media big data.For the mining of social media cross-media big data evolution rules,the social network cross-media big data association relationship discovery algorithm based on time and forwarding relationship was used to discover the time-slice association relationship of social network cross-media big data,and obtained each topic.The changes in the evolution process have enabled the mining of social media cross-media big data evolution rules.(3)Aiming at the problems of the semantic gap and low search efficiency of social networks represented by Weibo,a social media cross-media big data search algorithm(CPRT-IR)based on public representation and association relations is proposed.CPRT-IR uses the adversarial generation network to map the social network text features and corresponding image features to the common semantic space,and uses the difference between the reconstructed features of the cross-media data and the original features and the discriminant loss of the data between the media as a part of the network loss.The mutual game between the generative model and the discriminant model obtains the public representation of social network cross-media big data,and solves the problem of the semantic gap between social network cross-media big data.In order to further speed up the search efficiency,by combining the public representation and association relationship of social network cross-media big data,and implements the efficient search of social media cross-media big data.Through the methods of dimensionality reduction and clustering analysis,the public representation and search results of social network cross-media big data are visualized respectively.The experimental results further verify the effectiveness of the social network cross-media big data search algorithm.(4)Implementing a social network cross-media big data evolution mining,search and visualization system.The system consists of three modules:semantic expansion module,association discovery and evolution mining module,and search and visualization module.