Sentiment and topic analysis on social media

摘要: Both sentiment analysis and topic classification are frequently used in customer care marketing. They can help people understand the brand perception opinions from social media, such as online posts, tweets, forums, blogs. As such, recent years, many solutions have been proposed for both tasks. However, we believe that following two problems not addressed adequately: (1) Conventional usually treat tasks isolation. When closely related (e.g., posts about "customer care" often a "negative" tone), exploring their correlation may yield better accuracy; (2) Each post is assigned with only one label label. Since media is, compared to traditional document corpus, more noisy, ambiguous, sparser, single be able capture classes accurately. To address these problems, this paper, propose multi-task multi-label (MTML) model performs of sentiments topics concurrently. It incorporates results each task prior steps promote reinforce other iteratively. For task, trained multiple labels so they class ambiguity. In empirical validation, compare accuracy MTML against four competing methods different settings. Results show produces much higher classifications.