A Study on Highly Accurate Swearing Detection Model Based on Multimodal Data
Jiechao Deng, Yiwen Liu, Ling Peng, Pengju Tang, LU You-min · 2023
The current surge in popularity of social platforms like Twitter, Weibo International, TikTok, and online games has greatly increased users' online information interaction. However, it has also brought about the issue of online profanity speech abuse. Profanity speech, i.e., insulting or offensive remarks about individuals or groups, negatively affects the public environment and user experience, making the development of profanity detection techniques particularly important. In this study, CNN and LSTM are used to extract the sentiment features in the sentence and the dirty word features in the sentence using I-gram, cross-screening, respectively Subsequently, the raw scores of these two features are mapped to the range of probability distribution of [0, 1] by SoftMax function. Then, the probability distributions of these two features are combined together, and a value can be obtained by weighted average or crosstabulation operation, according to which the sentence can be judged whether it contains swear words or not. At the end of the article, the model is compared with other models, which fully demonstrates the advantages of this model.