Deep Learning for Reddit Text Classification: TextCNN and TextRNN Approaches
Qiyu Long, Z. Wang, Hao Yu · 2024
This research aims to enhance the subreddit recommendation system currently reliant on popularity metrics. It focuses on matching users with content relevant to their interests rather than just trending posts. Utilizing Word2Vec for word embedding and feature extraction, our model is based on a dataset comprising 20 subreddits, each with a total of 200,000 to 500,000 posts. Then we applied TextCNN, a simplified convolutional neural network, and TextRNN, a recurrent convolutional neural network for text classification to our model and compared their accuracy. The accuracies for identifying the top 1 most relevant subreddit were 65.1 % and 55.3 % respectively, and the top 3 accuracies were 80.5% and 72.9%, showing TextCNN's strength in classifying content on Reddit.