CrystalFeel - Multidimensional Emotional Intensity Analysis Engine
01 Jan 2021
CrystalFeel is a collection of machine learning based emotion analysis algorithms for analysing the emotional intensity properties in natural language. It is developed by researchers studying Affective Computing from A*STAR's Institute
of High Performance Computing.
To date, major language analytic services and theoretical models pertaining to emotion understanding are limited in that they predominantly consider discrete, categorical sense of emotions (happy vs. not happy, sad vs. not sad, angry vs. not angry)
and of sentiments (positive, negative, neutral). Although such discrete classification outputs of emotions and sentiments can be also represented as real-valued scores, these scores indicate the confidence or probability of the classifier but
not the intensity of the emotional experience. Scoring emotional intensity along a continuous scale is a relatively less explored feature.
Features
- Analyse a wide range of written/spoken natural language text input such as tweets, Facebook posts, comments, news headlines and articles, or speech transcripts
- Produce highly accurate and rich analysis outputs in terms of quantitative emotion intensity scores, across five analytic dimensions: fear, sadness, anger, joy and overall valence, as well as additional qualitative sentiment category and emotion
category labels
- Ground with explainable and theoretically sound affective AI research behind the engine, with industry-strength software robustness
CrystalFeel is developed with a solid ground of theoretical concepts and empirical knowledge from emotion science, in conjunction with innovative computational algorithms development incorporating in-house curated emotion intensity lexicon, latest natural
language processing and affective computing techniques. The underlying predictive algorithms underneath CrystalFeel include novel sociolinguistics inspired, psychologically meaningful and explainable features and system components, and have been validated
high predictive accuracy with average Pearson r value of 0.798 on predicting anger intensity, fear intensity, sadness intensity, joy intensity and valence intensity tested with human annotated ground truth data.
Industry Applications
CrystalFeel’s accurate, rich and predictive emotion analysis insights from natural language empowers a wide range of consumer, social and public applications in but not limited to the following industry-leading innovations and use cases:
- Market research: Understand consumer experiences and feedback in naturalistic context and inform product and marketing innovation
- Media and publishing: Understand, predict and recommend popular and viral news in digital platforms
- Ground sensing: Understand commuter experiences in real time and inform policy evaluation and calibration
- Disease surveillance and management: Timely and wide range surveillance of social media indicators of infectious disease development
- Mental wellbeing: Early community sensing of threats and signs of wellbeing issues & enable more effective healthcare resources allocation
Test out our interactive demo here: Crystalfeel
For more info or collaboration opportunities, please write to enquiry@ihpc.a-star.edu.sg.