Multilingual Toxic Text Classification Model Based On Deep Learning

The nature of comments usually has an important impact on the network environment. Polite and gentle comments can not only promote communication between users, but also maintain the stability of the network platform. On the contrary, rude and toxic comments will make the communication environment unacceptable. Therefore, we need to impose certain restrictions on comments. This article is based on the XLM-RoBERTa model to achieve the classification of multilingual toxic comments. We first use training and verification data to train and optimize the model, and then use the test data to get the final classification results. In addition, our model is compared with models such as LSTM and RNN. Experiments show that the model proposed in this paper has better classification performance.

I. Introduction

The nature of comments plays a particularly important role in the network environment. Reasonable and polite comments can not only bring convenience to users, but also maintain the peace of the communication platform. On the contrary, some impolite comments may make online discussions unpleasant. Therefore, we need a machine learning model that can actively identify the toxic content in sentences, where toxicity refers to any rudeness, rudeness, or other things that may cause people to leave the discussion. When users post comments, use the model to identify toxic comments. Then such comments will be blocked, so that the network environment can become more secure and comfortable.