The Design And Implementation Of Multi-Source And Multimodal Data Analysis Platform

With the rapid development of the Internet,a large number of network platforms are emerging for different user needs.The life of the users is closely linked with the Internet.This phenomenon leads users to generate large amounts of multi-source and multimodal data containing texts,pictures and videos on different network platforms.By analyzing multi-source and multi-modal data generated by users,the user can be provided with better services.Multi-source and multi-modal data analysis includes text data analysis,image data analysis,and multi-source and multi-modal data fusion.For multi-source and multi-modal data analysis,the traditional big data analysis platform cannot effectively support.Building multi-source and multi-modal data analysis platform can effectively analyze multi-source and multi-modal data to create better economic and social benefits.The project derives from Beijing Municipal Education Commission project "The research of cross-media data analysis and mining based on social perception".This project introduces the social perception into the multi-source and multi-modal cross-media data mining,mining the hidden multi-granularity and multi-angle correlation between the cross-media data and accurately understanding the cross-media data content.This paper mainly designs and implements a big data platform that can capture,store,analyze and visualize multi-source and multi-modal cross-media data.This paper firstly introduces the big data technology which deals with massive data and the deep learning technology which is good at processing images.Then it analyzes the related technologies of multi-source and multi-modal data analysis and analyzes the current development status of multi-source and multi-modal data analysis platform at home and abroad.To solves the difficulty that the single data analysis platform cannot analyze multi-network and multi-modal data such as texts,images and videos,the multi-source and multi-modal data based on distributed computing framework Spark and deep learning framework TensorFlow are designed and implemented.Analysis platform.From the perspective of platform performance,reliability,scalability and ease of use,the overall architecture of multi-source and multi-modal data analysis platform is designed,more over,multi-source multi-modal data acquisition,storage,analysis and visualization of four modules is designed and implemented.So the platform can meet the need for mass multi-source multimodal data analysis.Finally,I use Sina Weibo text data analysis,image content identification and multi-source multi-modal picture recommendation based on Twitter and Flickr data respectively to verify the data acquisition,storage,analysis and visualization capabilities of the designed and implemented platform.Through the analysis of Sina Weibo text data,we verify the platform's ability to collect data from Sina Weibo and visualize the storage,analysis and text analysis of text data.Through the identification of the image content,it verifies the platform's ability to collect,store,analyze and visualize the image data.The multi-source and multi-modal image recommendation based on Twitter and Flickr data validates the platform's ability to collect Flickr and Twitter data,as well as the storage capabilities,fusion analysis capabilities and analysis results of multi-source and multi-modal data.