Internet Forum Texual Information Filtering Mechanism for Emergency based on Bi-LSTM Neural Network

The internet forums are an important way of emergency information dissemination. However, with the development of internet and the vast amounts of users, there is plenty of useless information in internet forum data, especially during emergency. This not only increases the load for the subsequent data processing, but also influences the efficiency and accuracy of public opinion analysis. As a result, the useless data is harmful for the internet public opinion guidance for emergencies. Therefore, in view of internet forum text data of emergency events, in this work, we propose an information filtering mechanism of internet forum based on bilayer long short-term memory (Bi-LSTM) neural network. First, we propose a text information classification model based on LSTM neural network and related mechanism. Then, we evaluate the proposed mechanism by using the practical internet forums. The experimental results and analysis reveal that the proposed algorithm significantly improves the textual information quality as well as the accuracy.

I. Introduction

The network media is eliminating the monopoly of traditional media. It is re-defining the information dissemination methods in digital times [1]. In addition, the network media has gradually transformed itself into an important platform for public opinion development of emergency [2]. However, the network media is a combination of multiple forms, such as post bars, forums, teacher and student blogs, and microblogs [3]. Moreover, the internet forum has become an effective tool for emergency public opinion dissemination. Meanwhile, textual information takes a great position in internet forum, especially during emergency.