An indicative opinion generation model for short texts on social networks

Opinion generation is of great value since it provides main opinions for users within a short period of time. The last decade has witnessed a rapid development of social networks and massive data, and it is challenging for people to get the main opinions of short texts within a short period of time. Many studies of pinion generation have used feature weights based methods to summarize these texts sharing the same topic. However, these techniques fail to just regard the original text as the generation without considering the simplicity of language. To overcome the drawback, in this paper, we develop an indicative opinion generation model utilizing BM25 to identify the important text and using syntactic parsing to obtain the brief opinion representation. We first create a vector space model for clustering the short texts using the K-means algorithm. Then, by ranking the short texts sharing the same topic, we obtain the top-ranked representative short texts. Finally, we develop an indicative opinion generation model to obtain the main ideas by using syntactic parsing. We conduct extensive experiments on real datasets and evaluate the results by objective and subjective assessments. The experimental results show that our proposed model is effective and outperforms state-of-the-art methods.