Hybrid words representation for airlines sentiment analysis

Social media sentimental analysis is interesting field with the aim to analyze social conservation and determine deeper context as they apply to a topic or theme. However, it is challenging as tweets are unstructured, informal and noisy in nature. Also, it involves natural language complexities like words with same meanings (Polysemy). Most of the existing approaches mainly rely on clean textual data, however Twitter data is quite noisy in real life. Aiming to improve the performance, in this paper, we present hybrid words representation and Bi-directional Long Short Term Memory (BiLSTM) with attention modeling resulting in improvement in tweet quality by not only treating the noise within the textual context but also considers polysemy, semantics, syntax, out of vocabulary (OOV) words as well as words sentiments within a tweet. The proposed model overcomes the current limitations and improves the accuracy for tweets classification as showed by the evaluation of the model performed on real-world airline related datasets.