Sentiment Analysis on User-generated Video, Audio and Text

Social Media prevails in today's world for voicing an opinion. That makes it crucial for businesses, artists, content creators, and pretty much anyone else on the internet to analyse what people are saying about them and what they offer. It provides vital information on what they can change and improve upon. Hence, Sentiment Analysis is going to be a significant aspect to help organizations and individuals grow. Contrary to previous work, a broader viewpoint is taken into consideration and any kind of discrepancies is eliminated since the sentiment classified by any one modality is confirmed by the analysis of the other two modalities as well. In all the previous work done on multimodal sentiment analysis, the weightage given to all the modules was the same for fusion but to get perfect results, a trial-and-error logging method is used with many different weightages for fusion. This helped us in knowing which module is best suited for sentimental analysis and how each of them can alter the results obtained. With this type of the development in the sentiment analysis domain, a system which can extract the information in terms of the emotions of the people regarding a specific product from any of the three mediums (text, audio or video) can lead to the better technological aspects in the market. The objective was to build a system which can identify the sentiment categorized into six types: anger, joy, disgust, sadness, fear, and surprise of a video when the data is fed into it. The system developed depicts how much of each of these sentiments are present in a particular input.