Harmful Information Detection of Web pages with Attentional Deep Neural Networks and Multi-level Feature Fusion

With the rapid development of the Internet in our country, our country has gradually realized the informatization of society and formed a digital society through a huge population of netizens. In this digital society, websites are an important means for people to obtain information services. Our country has a large number of websites, and each website provides a large number of web pages. However, many of these websites are not effectively maintained. Every year, a large number of websites are hacked, causing web pages to be tampered with, and the tampered web pages basically spread harmful information. In addition, outside of government supervision, there are a large number of overseas websites that specialize in spreading harmful information. This large amount of harmful information that floods the Internet continues to erode and endanger normal social life. Therefore, relevant government departments are required to supervise and manage this problem. The means of supervision is to identify and detect harmful information content on the web pages published by each website. This paper focuses on the problem of classifying web pages for harmful information content in web pages, conducts stop word mining for web page text content, reduces the noise in the data by removing stop words unique to web page text, and proposes a method based on the classification model of deep attention learning and multi-level feature fusion achieves better classification and prediction effects on web page text. This paper conducts a large number of comparative experiments on real data sets, and the experimental results demonstrate the effectiveness of this method.