Popularity Prediction of Micro-Blog Hot Topics Based on Time-Series Data
Résumé
Topic popularity prediction has always been a popular research field. In previous studies, traditional methods are mostly used to predict the popularity of topics characterized by long text, long period, and large amounts of data. These models do not apply to hot topics generated on social network platforms. Hot topics on social platforms are characterized by rapid generation, development, explosion, and vanishing, resulting in topics often disappearing within a few days or even hours, significantly reducing the amount of data available. According to these features, this paper collects blogs related to the hot topic “2023 Hangzhou Asian Games” on Sina Micro-blog, and designs a hotness formula to measure the topic's popularity within a certain period. Subsequently, two baseline models were trained by combining convolutional neural networks (CNN) and long short-term memory (LSTM). The prediction results are compared using two evaluation metrics, RMSE and MAE. The model proposed in this paper outperforms the two baseline models in prediction.