AI Mining provides high accuracy emotion analysis of text. We are the next generation of sentiment analysis.
We extract emotions from words in a new way. We take a sequence of words as input, and output percentages of different emotions. For example, our analysis shows that the word “lovely” contains 79% love and 9% surprise, while “banana” is 33% laughter and 29% surprise. “My beard is on fire” is 63% laughter. “My fish just died” is 60% sadness. We sell access to our analyser via an API.
Our analyser is a custom design multi-layered artificial neural network. Neural networks are software algorithms inspired by the way the neurons function in brains. They are one of the core technologies used in machine learning, which itself is part of the re-emerging field of artificial intelligence. Neural networks are also referred to as models. Our emotion analysis models the way humans encode their emotions into words. Essentially, it is an incredibly complex mathematical equation, which converts words like “banana” into numbers, performs millions of calculations and produces an emotional score.
Our emotion analysis has several use cases:
- Writers of news headlines, blog headlines and text ads
- Brand and social analytics companies measuring the public’s reaction to news, ads, Facebook posts and tweets
- Customer service departments who need to bulk analyse or prioritise support requests
- Developers who want to add text emotion detection to their own analytics packages
These audiences are currently using the more simple sentiment analysis, which only labels text as positive or negative. We believe that upgrading to emotion analysis would significantly improve their businesses.
In terms of size, the market for text analytics is expected to grow 27% annually between up to US$9.2 billion by 2020. Emotion detection for voice and face is projected even higher at US$22 billion by 2020. So there’s plenty of business out there for the right provider.
There are very few other companies providing emotion analysis of text. This is mainly because of the lack of large corpuses of labelled emotional data. The biggest companies use a technique called Linguistic Inquiry and Word Count. This essentially counts how many good and bad words there are in a piece of text, and performs various mathematical operations to produce a set of emotional scores. There are far more companies providing sentiment analysis but they also tend to use lexicon based analysis.
We, on the other hand, have based our product around machine learning, putting us at the cutting edge of today’s text analysis companies. Our data mostly comes from analysing social media networks looking for emotional tags.
We currently provide two products. Emowatch that extracts 5 emotions as above, and a traditional sentiment analysis. Please read more about our products and try our demonstrations and use cases.