A Short Text Classification Approach with Event Detection and Conceptual Information

Text classification is an elementary task in Natural Language Processing (NLP). Existing methods, such as Long Short-Term Memory Networks (LSTM) and Attention Mechanism have recently achieved strong performance on multiple NLP related tasks. However, in the field of text classification, their results are often limited by the quality of feature extraction. This phenomenon is particularly prominent in short text classification tasks, since short text does not have enough contextual information compared to paragraphs and documents. To address this challenge, in this article, we propose a method to enhance the semantic information of short text with two aspects: event-level information extracted from text and conceptual information retrieved from external knowledge base. We take event and conceptual information as a type of supplementary knowledge and incorporate it into deep neural networks. Attention mechanism is utilized to measure the importance of the supplementary knowledge. Meanwhile, we have discussed the granularity selection for Chinese word segmentation, and select char-based models. Finally, we classify a short text with the help of event and conceptual information. The experimental results show that the proposed method outperforms the state-of-the-art methods.