Sentiment Analysis on Video Using 3D-CNN and Bidirectional LSTM

Sujay Angadi and R. Venkata Siva Reddy

Rapid development of technology and use of internet enabled users to easily connect, communicate and share their ideas with millions of other people. Billions of PC, smartphones with built-in camera is used to post videos on social websites like Facebook and YouTube which continues to increase. Opinions expressed across the social media platforms have great impact in making data driven decisions in businesses and organisations. It has become a challenging task to analyse such high volume of data which is posted every day on social media platforms. In proposed method, sentiment analysis on opinion videos is performed using hybrid deep learning network, comprising of 3D convolution neural networks followed by bidirectional long short-term memory. Proposed method effectively collects spatial and temporal information within video frames. Proposed method significantly outperforms the state of art sentiment analysis in terms of polarity detection on videos using publicly available datasets such as YouTube and MOUD.

Volume 11 | Issue 5

Pages: 227-240