Recommendation system for mood stabilization using content recommendation

Abstract

In the era of accelerating technological development, society is confronted with the paradoxical situation of making technological advancements while experiencing a decline in mental health. The importance of mental health seems to be declining significantly. The impact of our daily content intake on emotional well-being is clearly visible. For instance, while a melancholic song can make a person feel sad, an inspirational movie can charge a person’s spirit to come up stronger. Hence we intend to employ this concept to propose a system designed to recommend “Feel Good” YouTube videos with the aim of stabilizing an individual’s mood when it wavers or becomes low. To do this efficiently, we worked on the SEED Dataset, which is composed of EEG signals and Eye Movement data. We implemented a multifaceted approach, including the extraction of Differential Entropy Features, Wavelet Transform, Shannon Entropy features and Eye movement features. These were further harnessed by Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks to ensure accurate emotion classification. A thorough evaluation of these two deep learning models in the context of emotion classification is presented by focusing on their relevant merits and demerits. Based on the comparisons it is found that CNN is the most suited for our study with an accuracy of 93.01%. Once a mood classification is achieved, our proposed system will curate and suggest trending “Feel Good” content. To tackle this, we implemented a recommendation system based on the fusion of two prevalent techniques. Initially, text classification was employed to extract the emotion associated with the video and later, Pearson Correlation was utilized to obtain accurate correlation between the contents of the videos based on their corresponding ratings from viewers. Furthermore, concepts of Analytic Hierarchy Process (AHP) have been implemented to come up with an efficient algorithm which works in stabilizing an individual’s mood gradually. In essence, our innovative system encompasses two primary objectives: the detection of an individual’s emotional state through EEG signal analysis and the subsequent stabilization of their mood through targeted content recommendations. By combining these components, we envision a tool that not only comprehends the user’s emotional well-being but actively contributes to its enhancement.