Tweet Irony Detection Using Ensembles of Word Level Attentive Long Short-term Memory and Convolutional Neural Network
| Autor: | Xiaoni Zhao, Sheng Li, Dequan Zheng, Bing Xu |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
business.industry
Computer science media_common.quotation_subject Sentiment analysis 02 engineering and technology Semantics computer.software_genre Convolutional neural network Irony 020204 information systems Test set 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence business computer Natural language Natural language processing Word (computer architecture) media_common |
| Zdroj: | ICNC-FSKD |
| DOI: | 10.1109/fskd.2018.8687128 |
| Popis: | Irony detection is a key subtask of many natural language processing tasks. For example, irony can flip the polarity of “apparently positive” sentences and have a negative impact on polarity detection performance in sentiment analysis. Up to now, most of the irony detection methods are simple to deal with the task as text classification. However, irony expresses semantics in a very subtle way, so it requires a deeper understanding of natural language, which is beyond the grasp of standard text categorization techniques. In this paper, we design and ensemble two independent models, based on Bidirectional Long Short-term Memory(BiLSTM) strengthened with self-attention mechanism and a pre-trained Convolutional Neural Network(CNN) to capture both the semantic and sentiment information in tweets. In the test set of irony detection, our model is superior to the traditional approach in terms of F1-score. More importantly, the model training takes less time because we don't need to manually find the rules and prepare a lot of time-consuming and labor-intensive other language knowledge, so our model is more developable. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|