Deep Learning Methods for Emotion Detection from Text – Dr. Liron Allerhand
Sentiment analysis is an active research field where researchers aim to automatically determine the polarity of text [1], either as a binary problem or as a multi-class problem where multiple levels of positiveness and negativeness are reported.
Recently, there is an increasing interest in going beyond sentiment, and analyzing emotions such as happiness, fear, anger, surprise, sadness and others. Emotion detection has many use cases for both enterprises and consumers. The best-known examples are customer service performance monitoring [2], and social media analysis [3].
In this talk, we present a new algorithm based on deep learning, which not only outperforms state-of-the-art method [4] in emotion detection from text, but also automatically decides on length of emotionally-intensive text blocks in a document.
Our talk presents the problem by examples, with business motivations related to the Microsoft Cognitive Services suite. We present a technique to capture both semantic and syntactic relationships in sentences using word embeddings and Long Short-Term Memory (LSTM) based modeling. Our algorithm exploits lexical information of emotions to enrich the data representation. We present empirical results based on ISAER and SemEval-2007 datasets [5,6]. We then motivate the problem of detecting emotionally-intensive text blocks of various sizes, along with an entropy-based technique to solve it by determining the granularity on which the emotions model is applied.
We conclude with a live demonstration of the algorithm on diverse types of data: interviews, customer service, and social media.
Sentiment analysis is an active research field where researchers aim to automatically determine the polarity of text [1], either as a binary problem or as a multi-class problem where multiple levels of positiveness and negativeness are reported.
Recently, there is an increasing interest in going beyond sentiment, and analyzing emotions such as happiness, fear, anger, surprise, sadness and others. Emotion detection has many use cases for both enterprises and consumers. The best-known examples are customer service performance monitoring [2], and social media analysis [3].
In this talk, we present a new algorithm based on deep learning, which not only outperforms state-of-the-art method [4] in emotion detection from text, but also automatically decides on length of emotionally-intensive text blocks in a document.
Our talk presents the problem by examples, with business motivations related to the Microsoft Cognitive Services suite. We present a technique to capture both semantic and syntactic relationships in sentences using word embeddings and Long Short-Term Memory (LSTM) based modeling. Our algorithm exploits lexical information of emotions to enrich the data representation. We present empirical results based on ISAER and SemEval-2007 datasets [5,6]. We then motivate the problem of detecting emotionally-intensive text blocks of various sizes, along with an entropy-based technique to solve it by determining the granularity on which the emotions model is applied.
We conclude with a live demonstration of the algorithm on diverse types of data: interviews, customer service, and social media.
