A New Polar Transform‐Based Network for Multi‐Lingual Handwritten Text Sentiment Analysis

Sentiment analysis has attracted particular attention from researchers because it is valuable for various real‐world applications, such as opinion mining and market analysis in business. Unlike existing methods that use a natural language approach for sentiment analysis, this study uses a computer vision approach. In addition, the existing work ignores applications to multi‐lingual text (English and Kannada, which is the language of Karnataka state, India). To the best of our knowledge, this is the first work that uses images of handwritten text. Our observations indicate that positive handwriting typically features thick strokes, good quality, and well‐formed, readable characters. Negative handwriting tends to be sloppy and overwritten, making it difficult to read. Neutral handwriting exhibits neither distinctly positive nor negative traits. To extract these features, we propose a new model based on a polar transform to identify characteristics that distinguish three classes. Changes in theta and distance within the polar transform produce unique cluster points related to each class. The extracted features are then encoded using a convolutional neural network (CNN) in a novel manner. We test our method on our own dataset and compare the results with state‐of‐the‐art methods, demonstrating the superiority of our approach. The code and dataset are available at https://github.com/prajwalkumarprof/A‐new‐Polar‐Transform‐Based‐Text‐Sentiment‐Analysis.git .