Research On Microblog Rumor Detection Based On Hierarchical Attention Mechanism

Sina Weibo allows any user with an Internet-connected device to share their thoughts in real time and publish in real time what they might have witnessed.The diversity of Sina Weibo has stimulated people's enthusiasm for expressing opinions,which has made Sina Weibo more and more used as a tool for gathering information.Although it brings great convenience,there are almost no restrictions on the content that can be shared on Sina Weibo.Its openness and timeliness lead to the rapid generation and dissemination of rumor information,and the dissemination of rumor information is devastating.Therefore,the openness of the Sina Weibo platform provides an opportunity for how to use natural language processing and data mining technology for rumor detection.Rumor detection is a subtask of text classification.The purpose of this research is to excavate important features from event-level microblogs composed of massive original microblogs and their comment information,and use deep learning models to detect them.The main work of this paper is as follows:(1)In response to the problem that the Weibo data set disclosed by the previous research is too long,this article crawled and screened the rumor event information from January 2019 to September 2020 from the official community management center of Sina Weibo.After screening,there are a total of 1906.Weibo information about a rumor event.We also collected non-rumorous events from the headline section of the Weibo homepage.After screening,there were 2017 non-rumorous events in Weibo.Finally,combine with the public data set to construct a new data set for experimentation.(2)Aiming at the problem of information loss caused by artificial division of event time periods in the previous research on rumor detection,a variable length time series comment division algorithm is proposed.The algorithm makes full use of the likes and time information in the comments to dynamically divide the event time series.On the basis of this algorithm,a hierarchical attention mechanism with dynamic division(HAD)model based on dynamic division is proposed to study the problem of rumor detection in event-level microblogs.The model uses a hierarchical attention mechanism to pay attention to the microblog events at the microblog level and time period level at the same time,and then classify them.The model is compared with the classic baseline model to verify the effectiveness of the proposed model.(3)Aiming at the problem of the lack of semantics of a single model and the problem of key feature selection,a new model is proposed by combining the Dynamic Convolutional Neural Network(DCNN)with the HAD model.The fused model not only uses the attention weight to retain effective information,but also solves the problem of key text information loss.