Multimodal Learning Approach for Multi-topic Twitter Summarization
Yangyang Cao, Yanmin Wu, JinLi Qi, Zizhong Chen · 2022
Twitter summary is designed to filter out the content summary from lots of noisy tweets, which can be used for Twitter search, public opinion analysis, hot topic discovery, etc. But the existing Twitter summary methods mostly focus on individual topics or hot events, and there are few studies on Twitter summary with multi-topic posts. For the sake of resolve this problem, we propose a Twitter summary method based on social network information and manifold learning, which can extract tweets covering multiple topics, and better integrates the topic information and structural information of the tweets. First, the multi-modal manifold learning method is improved, and the social network information is integrated. Second, in order to test the relevance of the tweet summary and the topic, a test method of topic membership is designed. What's more, due to the lack of evaluation data sets, the author built a manual evaluation dataset. Our experiments on large-scale real tweets demonstrate the effectiveness of our framework.