Emotive ontology: extracting fine-grained emotions from terse, informal messages

With the uptake of social media, such as Facebook and Twitter, there is now a vast amount of new user generated content on a daily basis, much of it in the form of short, informal free-form text. Businesses, institutions, governments and law enforcement organisations are now actively seeking ways to monitor and more generally analyse public response to various events, products and services. Our primary aim in this project was the development of an approach for capturing a wide and comprehensive range of emotions from sparse, text based messages in social-media, such as Twitter, to help monitor emotional responses to events. Prior work has focused mostly on negative / positive sentiment classification tasks, and although numerous approaches employ highly elaborate and effective techniques with some success, the sentiment or emotion granularity is generally limiting and arguably not always most appropriate for real-world problems. In this paper we employ an ontology engineering approach to the problem of fine-grained emotion detection in sparse messages. Messages are also processed using a custom NLP pipeline, which is appropriate for the sparse and informal nature of text encountered on micro-blogs. Our approach detects a range of eight high-level emotions; anger, confusion, disgust, fear, happiness, sadness, shame and surprise. We report f-measures (recall and precision) and compare our approach to two related approaches from recent literature. © 2013 IADIS.