A Sentimental Analysis System Using Zero-Shot Machine Learning Technique

Abstract

The internet has turned our lives upside down and has become a global means of communication. As the world is rapidly advancing, many new and challenging calls for humankind are associated. One of those challenges is analyzing the sentiments, i.e. opinions or feelings of the person or any user such as customers, while choosing and buying any product. For cases and situations like this, analysis of sentiments or opinion mining plays a significant role. Sentiment Analysis is vital because the customers can get an overview and understanding of reviews of the customers who have already purchased that particular product. Also, it helps them make decisions about their purchase and hence proceed forward accordingly. In comparison to the existing work, the proposed work considers all the sentiments throughout any conversation or review, whether they are good or bad, and hence classifies them further as positive and negative with their extent i.e. percentages of positivity and negativity in the statement. It also finds out the label of the review or any other conversation so that the users can get an idea about the domain of the conversation. Even though related research work has already been done, there is still a need to improve the accuracy and understandability of sentiment analysis. This work is mainly done by using the zero-shot learning technique. After classifying the reviews and predicting labels, the spaCy model is used with it to get essential keywords and phrases for the conversation. In the proposed work, this is done by discarding greetings with a score greater than 80%.