Abstract:
Modern world has provided us the ability to communicate with each other all over the world instantaneously. Total sent emails in 2021 reached an estimated staggering number of 319.6 billion and is expected to cross 376.4 billion in year 2025. Email is used as a primary form of communication between people in business setting as well as formal. In a day a person can receive more than a hundred emails per day so it becomes quite tedious to answer every email so there is a need for email response to be automated to take burden of the employees. Although automated email response systems have existed for quite some time, but they don’t consider the emotions based on the sentiment which is in turn based on the context of the emails thus rendering the response less natural. A lot of work has been done in the field of context, sentiment analysis and emotion detection, but rarely all these fields have been combined. This work is a continuation of “Sentiment Analysis for Automated Email Response System” done by Muhammad Babar Abbas and Dr Muhammad Mukarram Khan, they have implemented the context-based sentiment analysis for email without taking into account the emotion as well. This research is to rectify this aspect and create a system that can generate a response while considering contexts and emotions and sentiments related to that context thus giving a more natural and accurate response. For the purpose of this research a new custom dataset was also created which comprised of 1000 tweets taken from the CrowdFlower dataset with their contexts. We looked at the various algorithms that were being utilized by the researchers in the separate fields of context analysis, sentiment analysis and emotion recognition and compared them to find out the best possible algorithm to be used for automated email response system. Finally the accuracy of the models are tested on real world datasets and are compared to gauge out the best suitable algorithm.
