Advances in Sentiment Analysis- Leveraging Deep Learning

Sentiment analysis, or opinion mining refers to the analysis of opinion, subjective sentiment, attitude and emotion from written textual content as naturally produced by people (Liu, 2012). There is increasing interest in field of sentiment analysis due to digitization, social media marketing and Ominchannel experience. Sentiment analysis provides tools for extracting trends in the mindscapes of users, and applications of these mindscapes are multifold- stock market predictions, political election prediction, advertising, hate speech detection.

Popular research method was CAVE- content analysis of verbatim explanation, that was used by both social psychologists and positive psychologists to study sports team, living legends with outsized performance and also the dead- e.g. presidential speeches given by former POTUS. Sentiment in the content was mined and The preferred method was NLP or Natural Language processing, which would be trained to classify sentiment/opinion . The job involved training the data set and NLP was largely rule based system with statistical approaches like Bayesian classifiers.

A sentiment analysis task was binary or ternary classification problem, where sentiment was classified positive or negative . It was binary classification. (Jurafsky and Martin , 2009).

With time, this evolved to ternary classification where sentiment got classified as Positive, Neutral or Negative. This further evolved to more nuanced classification and sentiment intensity was introduced -strongly positive or strongly negative.

However, both Binary and Ternary classification can be considered primitive, because they are one dimensional approaches . This is because Sentiments are classified on one general scale of polarity (Ohman et el, 2016).

Today’s political and economic arena is complex and intensely competitive , hence one dimensional approach to sentiment analysis does not yield any competitive advantage/moat. This realization has lead to more fine tuned classification of sentiment/emotion. The latest trend is fine grained classification of emotion/sentiment to put label such as Anger, Sadness, Joy etc. This is called multi-class emotion classification.

SuccessNeurons pioneers multi-class classification of sentiment/opinion by using advanced deep learning algorithms.

Muti-class emotion classification gives advantage, e.g politician may want to know voters turnout on election day. This can be determined by multi class sentiment analysis and going beyond binary/ternary classification (Positive, Neutral, Negative) to nuanced multi class (Angry, Sad, Happy, Excited etc.) Those with intense emotions-e.g. Excited, Elated, Angry , Rage are more likely to come out for voting and influence outcomes.

Voter having positive emotion , but lacking intensity is unlikely to turn up for voting.

Multi class emotion classification is big leap forward , because it empowers candidates with feedback on impact their messages are having on target voters.

At SuccessNeurons, we use the most advanced algorithms for multi-class classification – leverage latest research in deep learning, WE extensively used - multilevel perceptron , recurrent neural networks, convolutional neural networks that have been trained and delivering high accuracy.