Toxic Comments Classification using LSTM and CNN

Online platforms have grown to be an important medium for expression and communication in the digital era. But the anonymity and absence of in-person communication on these platforms often encourage the spread of offensive remarks, which can be harmful to both people and communities. Before classifying comments as dangerous, natural language processing, or NLP, must first comprehend and assess textual data in order to identify feelings, patterns, and context that may be harmful. Tokenizing the comments, generating features for the classification model, and preprocessing the text input are all done in this instance using natural language processing approaches. Specific NLP methods such as sequence padding, text vectorization, and tokenization may be covered in more detail. This study combines the architectures of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) to present a unique technique for categorizing harmful remarks. This study pre-processes the text data to eliminate noise and unnecessary information. Next, we use word embeddings to embed the comments into numerical vectors. Next, in order to capture the spatial and sequential relationships present in the text data, this study uses a hybrid model that combines CNN and LSTM layers. The CNN component assists in identifying local patterns and characteristics in the comments, while the LSTM component allows the model to comprehend the sequential flow of language. This study uses a publicly accessible dataset of labelled harmful remarks to train and assess the proposed model. The proposed model’s performance is measured using common measures including accuracy, precision, recall, and F1-score.