Identifying Purchase Intent of Tweets using Natural Language Processing

Abstract

Social Networks such as Twitter have grown immensely in the recent past with rapid adoption by users where many thoughts/ideas/opinions are publicly tweeted in massive scales. This includes a person’s intention to purchase a product or service in the foreseeable future. Or in other words ‘Purchase Intent’. The ability to effectively and efficiently identify purchase intent of tweets will revolutionize the search engine keywords and social media advertising dominated landscape of online marketing and create a paradigm shift towards creating unique, direct, and quality conversations between organizations ready to serve a need/want of a customer in a timely and appropriate manner. This research focuses on identifying purchase intent of twitter using natural language processing techniques. The scope of the research has been limited to the domain of iPhones. Using a fusion of techniques, the domain specific methodologies introduced in this research have successfully delivered the best results in comparison to contemporary related research. This research considers bag of words based features, tone based features, lexicon based features and part of speech tags based features to identify purchase intent of tweets using natural language processing. Furthermore a novel technique termed as derived nGrams based feature extraction methodology has been successfully introduced in this research which has shown significantly improved results over all other feature groups. Emoji has shown growing popularity on social media such as twitter where Emoji are used to convey contextual information in non-text format. As Emoji are mostly disregarded in many research in this domain, this research has considered an approach where emoji are converted to textual information and incremental results have proved its pertinence in the process. Overall, the domain specific methodologies used in this research has shown a 98% accuracy of identifying purchase intent of tweets in the iPhone domain using a supervised learning approach.