Yao Chen, Zixi Wang, Kevin Xuan
Over the past decade, Sentiment Analysis has been getting extremely popular with the arrival of deep learning neural networks and has always been one of the frequent tasks to do in Natural Language Processing (NLP). Recently, the idea of ensemble method and multi-modal learning have also been widely used in different applications, and in this paper we also exploit model ensemble to derive the emotion that is conveyed in the text of a review. To extrapolate the sentiment of a paragraph of text, we start with a baseline model and then construct three separate models which individually focus on the following: polarity, stress, and sarcasm. We believe that these sentiments have high correlation to whether the text is conveying a positive or a negative sentiment, and learning from a combination of the outputs of each model will allow us to earn higher performance in sentiment analysis.
Due to the size limit of Github, please refer to the Google Drive link for model weights and .csv files saved in sub-folders.