A Bert-based Model with Tsk Fuzzy Neural Network for Ai Comments Detection

With the advancement of large language models (LLMs) and chatbot technologies, the application of AI-generated text on social media has become increasingly popular. Therefore, the detection of AI comments has become important. Traditional AI text detection methods are mainly applied to long texts like academic writings. To make up for this deficiency, our study focuses on comments, which are relatively short texts. We propose two BERT-based models with Takagi-Sugeno-Kang fuzzy neural network (TSK FNN) to detect AI-generated comments on social media platforms. The first model is just a TSK FNN (BERT-TSK for short), which uses BERT features extracted from the hidden layers and the sentiment value of each comment (obtained by the sentiment analysis model from Hugging Face) as inputs, and the second is a BERT-TSK hybrid model (BERT-TSK-H for short), which embeds TSK FNN into BERT hidden layers for jointly training. The study compares the performance of our proposed two models with the fine-tuned BERT model on Chinese and English comment datasets. The experimental results show that the proposed hybrid model BERT-TSK-H outperforms BERT-TSK and the fine-tuned BERT model in terms of most of the metrics, Accuracy, Precision, Recall and F1-score on the two datasets.