@MASTERSTHESIS{pgi2020011, author = "L. Stoyanova", supervisor = "W. Wallace", title = "Topic Modelling, Sentiment Analsys and Classification of Short-Form Text Customer Journey of Insurance Purchases", school = "Department of Computer and Information Sciences, University of Strathclyde", year = "2019", abstract = "Upon consultation with professionals in the field of social media textual data analytics and a systematic review of literature in the field of topic modelling and sentiment analysis of shortform text, a research gap was identified, for which a prototype system was developed. The business problem faced is the lack of an automated approach to topic-sentiment extraction and classification of user-generated text based on the stage that the user is situated at in their customer journey in association with the purchase of a product or service. The following research proposes a system of tools that can extract topics and associated sentiment polarity from social media data, and subsequently allocate user-generated text in pre-defined classes that correspond with stages of the customer journey. The research involved experimental procedures in the field of sentiment classification, topic modelling and text classification. To evaluate the models{\^a}€™ performance a survey was distributed, which engaged a total of 58 respondents to perform the same tasks that the algorithms were given. The technical and human-agent experiment results were compared with the aim of evaluating the ability of an automated approach to solve this business challenge in a timely and efficient manner, which would emphasise the organisational benefits of cost-cutting and intelligent decision-making, which could be achieved following the implementation of the system. Considering the scope of the research project, the data used was extracted from social media websites Facebook and Twitter, and thus lacked labels, hindering the application of supervised learning for the classification task. Nonetheless, unsupervised and semi-supervised approaches were implemented, with the script for supervised model being annexed to support the work of other researchers. The conceptualised system of algorithms has measurable benefits to organisations and has been approved for implementation as part of the initial stages of a strategic project in the University of Strathclyde. The research presents exciting opportunities for future research, as well as actionable recommendations and implications for both text analytics professionals, business owners and academics.", }