Abstract. Businesses are increasingly facing complaints from customers online, leading to the need for proper classification of these messages to improve transaction processing efficiency. In spite of the importance that has to be given to the issue only a little research was conducted in the categorization of complaint texts, particularly in regard to the utilization of negative emotions and character details. In addition, clients' erratic and aggressive behaviours present further grammatical and semantic difficulties. To tackle these issues, a complaint categorization model has been developed based on Word Embedding and GRU & LSTM, which creates word-level text vectors and extracts features from complaint messages. The relative importance of each word feature is then determined using an LSTM and GRU-based approach, allowing the model to prioritize the most important features for categorization. Experimental results have shown that this model outperforms several text classifications baselines. Keywords: Recurrent Neural Network ,LSTM and GRU,Skip-Gram,Continous Bag of Words
Bindhu et al. (Sun,) studied this question.