Background

EMOTIVE is a specialist analytics service for extracting and detecting fine-grained emotions in social-media and Twitter. Our state-of-the-art semantically driven engine is unique amongst existing approaches. It can analyse thousands of tweets a second to extract from each tweet a direct expression of one of eight basic emotions: anger, disgust, fear, happiness, sadness, surprise, shame and confusion, and much more.

Input

Campaign hashtags, keywords and/or phrases to extract real-time tweets about your product, service and/or brand.

Output

Emotions, strength of emotions, expressions, emotional extremes. Detailed reports, over location, time, terms, relative analysis against other campaigns and events.


Step 1

A clean state-of-the-art social-media specific Natural Languag Processing (NLP) pipeline developed at Loughborough University parses every tweet in real-time.

Step 2

A unique semantic network (an Ontology) of emotions is efficiently matched from memory to extract emotional content in Twitter and social-stream messages. A large variety of emotional expression phrases, terms, slang, substrings are analysed.


Step 3

Depending on the specific needs of the customer, the social-stream with emotion enriched data is automatically filtered and further analysed using Machine Learning (ML) techniques. Post processing and dataset aggregation takes place so that various facets of the emotional reactions in the dataset can be analysed and reported on.

Reports

An extensive set of reports with insight on the campaign will be made available to individual clients.


Processing speed

Our program and the entire natural language parsing engine were optimised for big data, so that potentially all UK based Tweets comming in each second can be processed. The analysis can be further scaled on a per client basis to cover the enitre world, which represents around 10,000 tweets/sec.

Performance

EMOTIVE has been extensively tested and evaluated, achieving an F-measure - a measure commonly used to evaluate the performance of these types of systems - of 0.96, the best that has been reported for such a task.


Academic Research