Topic identification and sentiment trends in Weibo and WeChat content related to intellectual property in China

Intense frictions in global trade have made intellectual property (IP) an important topic of public concern. Meanwhile, new media and online communities have become important platforms for the public to discuss IP issues. Mining the core topics and judging their sentiment status from the public's massive online IP data are important means for the government to formulate and evaluate IP policies, for enterprises to carry out R&D and identify business opportunities. Hence, this study aims to conduct topic identification and sentiment trends in Weibo and WeChat content related to IPs in China by employing a novel ensemble method combining the term frequency inverse document frequency (TF-IDF), TextRank, latent Dirichlet allocation (LDA), the Word2vec model, and attention-based bidirectional long short-term memory (BiLSTM). To be more specific, the text information on IPs in Weibo and WeChat is extracted using the TF-IDF and TextRank algorithms. Then, the probability of keywords in text and their IP topics are obtained based on the LDA and t-SNE models. Sentiment polarity and topic trends are analyzed by the Word2vec model and BiLSTM. The results show that 16 topics related to IP were identified, and most topics presented high levels of positive sentiment; the development trend lines of the two emotions are easily affected by abnormal events, and thus, show obvious fluctuation.