Multi-user Instance Relation Extraction Technology For Twitter

As a new type of Internet information exchange platform,social media network has developed rapidly in recent years.More and more netizens are beginning to use social media to share information and pay attention to news events.They build a large and complex social network based on social media.By analyzing and summarizing these social relationships,we can provide an efficient and structured data source for social media data mining research and discover great commercial value.So,the research of user relation extraction in social media is great significance.However,due to the complexity and diversity of social media data,it is difficult to extract the features of relation extraction comparing with the traditional news corpus.To solve this problem,this article takes Twitter which is a representative social media network platform as an object to study multi-user instance relation extraction technology.This article proposes a new deep neural network model which is based on piecewise convolutional neural network which uses characteristics of Residual Network to improve the effect of relation extraction.Besides,the relation extraction datasets of social media are less.To construct sufficient dataset,this article also proposes a multi-channel Twitter data collection framework to achieve the efficient data collection.According to the characteristics of Twitter data,processing data with noise filter and user instance extractor.And then using the word vector technology to represent the processed data and convert it into a data format that can be processed by the neural network.Based on the improvement of convolutional neural network,it can achieve the user relation extraction for Twitter.Finally,on this basis,designing and implementing the user instance relationship extraction system of Twitter.The main work of this article is as follows:Implementing the efficient Twitter data collection.According to the research on Twitter data collection technology and the needs of experiments and application scenarios,this article proposes a multi-channel data collection framework which is the combination of Twitter API and web crawler to achieve data collection.In order to ensure the diversity of data sources and solve the problem of data collecting limited,proposing a data collection strategy for different data scope,including: specific user data collection,keyword searching data collection and real-time data collection.It ensures the efficient of data collection.According to the characteristics of Twitter data,a multi-user instance relation extraction model for Twitter is proposed.The model constructs the sample data for experimental training and testing by using noise filter and user instance extractor on the collected data.On this base,constructing the deep convolutional network model by using the residual network characteristics to optimize the performance of relation extraction.Experiments show that the model has a good effect on Twitter user relation extraction in precision and recall.Designing and implementing the multi-user instance relation extraction system for Twitter.The system consists of data collection module,multi-user instance relation extraction module and display module.The system functions include: Twitter data collection and storage,Twitter data noise filter,Twitter user instance extraction and iteration,user relation extraction and storage,and relation extraction results display.After testing,the system has a good performance,the display interface is simple,convenient and suitable for the actual Twitter user relation extraction.