Personalized Emotion Detection from Text using Machine Learning

Abstract

Social media platforms have evolved into essential tools for sharing emotions with the rest of the globe. People utilize text, photos, music, and video to express their emotions or their points of view. On the Internet, these social media platforms generate a vast volume of unstructured data. All the existing emotion detection models based on the supervised learning methodology require to collect labelled data which is time- consuming task. Next, for languages that are underutilized in terms of NLP, such as Polish vocabulary. To tackle these issues, this paper presents a model that functions as an emotional awareness system. It is based on open forums with automated audio information and text analysis for user-based emotion detection. The model is a key component of many natural computer programmers for processing natural language. The proposed model is implemented using PocketSphinx as an ASR. In addition to this, the model also uses the Word2Vec model, K-means clustering, and the Term frequency-inverse document Vectorizer for text analysis and segmentation. Further, the data from the International Survey on Emotion Antecedents and Reactions (ISEAR) was used to test and train the model. Finally, the concept evolves into a user-based system that will act as a personal assistant for segregating people's emotions as they experiment with new apps. The ISEAR database will be used for performance testing. When utilized in the database, the suggested model greatly outperforms previous models, with an accuracy of 80.25 percent and an F1 score accuracy of 88.86 percent. As the future work, this algorithm can be integrated into an application where emotion is used for recommendation.