Improving Accuracy of Sentiment Analysis for Depression Recommendation using Multi-Domain Fuzzy Rules

Social media & e-commerce has opened up the doors for human behavioral analysis in ways which were not possible before. Companies have the ability to track user’s mood and suggest advertisements which can trigger buying decisions based on it. This is possible due to user’s real time social media updates. Users nowadays are willing to provide information like their location, their age, nearby friends information, their mood, their buying patterns, etc. Companies do not intentionally collect all this information, but it has become a matter of social pride to post it as social media status and updates. The information available can be put to use in multiple forms- predict election results, movie success, product liking/disliking, travel destination recommendation, health care, etc. In our work, we utilize this textual information posted by different users and analyze their depression level focusing on negative sentiments. In order to perform this task, we have considered user’s tweets, any links which they might have posted, the time of the tweet, their age group and any previous depression history of the user. All these parameters are given to a novel fuzzy decision tree that uses sentiment analysis and game theory-based scoring in order to evaluate the depression score for the user. We analyzed the system on different real-time users, and observed that the system predicts depression level with more than 90% accuracy. Our work can be used to generate a prototype to identify if a person is in a depressive state and figure out the intensity of his/her depression.