Social applications such as Weibo have provided a quick platform for information propagation, which have led to an explosive propagation for hot topic. User sentiments about propagation information play an important role in propagation speed, which receive more and more attention from data mining field. In this paper, we propose an sentiment-based hot topics prediction model called PHT-US. PHT-US firstly classifies a large amount of text data in Weibo into different topics, then converts user sentiments and time factors into embedding vectors that are input into recurrent neural networks (both LSTM and GRU), and predicts whether the target topic could be a hot spot. Experiments on Sina Weibo show that PHT-US can effectively predict the hot topics in the future. Social applications such as Weibo provide a platform for quick information propagation, which leads to an explosive propagation for hot topics. User sentiments about propagation information play an important role in propagation speed, and thus receive more attention from data mining field. In this paper, a sentiment-based hot topics prediction model called PHT-US is proposed. Firstly a large amount of text data in Weibo was classified into different topics, and then user sentiments and time factors were converted into embedding vectors that are input into recurrent neural networks (both LSTM and GRU), and future hotspots were predicted. Experiments on Sina Weibo show that PHT-US can effectively predict hot topics in the future.
展开▼
