Jumping Finite Automata for Tweet Comprehension

Every day, over one billion social media text messages are generated worldwide, which provides abundant information that can lead to improvements in lives of people through evidence-based decision making. Twitter is rich in such data but there are a number of technical challenges in comprehending tweets including ambiguity of the language used in tweets which is exacerbated in under resourced languages. This paper presents an approach based on Jumping Finite Automata for automatic comprehension of tweets. We construct a WordNet for the language of Kenya (WoLK) based on analysis of tweet structure, formalize the space of tweet variation and abstract the space on a Finite Automata. In addition, we present a software tool called Automata-Aided Tweet Comprehension (ATC) tool that takes raw tweets as input, preprocesses, recognise the syntax and extracts semantic information to 86% success rate.