News topic detection based on the principle of minimum entropy
Shuang Hua Yang · 2022
Current topic detection methods generally use different algorithms to aggregate the extracted features to obtain corresponding topics, but do not fully utilize the features of news texts and key elements of news. For this, a news topic detection model based on the principle of minimum entropy (ME-NTD) is proposed. First, we extract news keywords through TextRank, and extract news entity elements with the help of text lexical analysis tools. Then, we map news texts into text vectors through word embedding, calculate text association weights by combining news entity elements, and construct text association graphs through text vectors and text association weights. Finally, we use the idea of hierarchical coding based on the principle of minimum entropy to randomly walk the nodes on the text association graph to achieve hierarchical coding of topics and objects within topics. Experimental results on three news datasets show that the ME-NTD model has better performance and efficiency than the comparative methods.