Article Preview
Top1. Introduction
With the continuous popularization and development of information superhighway, information technology has penetrated into every corner of our social living (He 2021, Camero 2019). It has changed people's life and work style with unprecedented speed and ability. We are in an era of information explosion (Kumari 2017). On one side, the Internet contains a vast amount of information which is far beyond people’s imagination. On the other hand, people often feel helpless when they face the vast ocean of information. It is called as information overload (Schmitt 2018, Swar 2017). It is a challenging task to help people effectively to manage massive information and quickly select useful information that they are interested in.
The web information is increasing, including online news, e-magazines, online technical reports, online documents, e-mail, BBS, online announcements (Yamamoto 2018). The traditional method to handle daily huge amount of information is time-consuming and laborious. The automatic text classification can directly filter and classify the document information (Kadhim 2019, Nguyen 2018). The user can only receive minor part which they are interested in. Then, users can be liberated from tedious document processing work and can easily understand and distinguish different document contents. A large number of complicated documents can be regulated and systematized, and the utilization rate of information can be greatly improved.
One the other hand, while people can easily send and obtain web information, they also face with many harmful or illegal information, such as pornography, violence, and superstition. Unhealthy content such as gambling can be seen everywhere on the Internet. Even criminal text may be exist in BBS, blog, e-mail etc. to carry out reactionary propaganda, fraud, extortion, terrorist threats, drug sales and other illegal and criminal activities. Installing illegal information filtering software cannot effectively prevent the occurrence of illegal web information. Through legislative means, investigating the criminal responsibility of criminals can effectively crack down on this kind of criminal behavior. However, due to the lack of effective evidence, the criminals may be free from the evasion of legal sanction.
Text classification refers to the process of marking a free document with one or more predefined category labels according to its content information (Kowsari 2019, Mirończuk 2018). In order to correctly perform the task of text classification, it must input the useful information of the text into the computer to scientifically abstract the text and establish mathematical model to describe the text. The document expression is a key part of the text classification. The text representation refers to many representation methods and techniques of text retrieval (Wang 2020, Luo 2019). The common used text retrieval methods include: Boolean model (Lashkari 2009), vector space model (Raghavan 1986) and probabilistic model (Feng 2018). These models deal with feature weighting, category learning, and similarity calculation from different perspectives.