On Mid-Summer’s Eve (June 21st 2017), Dominic Cassidy of Bricolage and Dr Sophie Carr of Bays Consulting came together to give a short, case study based presentation to the London Text Analytics Meetup group. Bringing together practitioners and researchers in the field of text analytics and natural language processing, this incredibly welcoming group provides a vibrant, open and supportive atmosphere in which to discuss approaches and findings, and learn from peers and colleagues working in the same area.
So, it was on one of the hottest days of the year that Dominic and Sophie presented their thoughts on how structured ontologies and taxonomies could help improve the computer algorithms used for natural language processing. Using a series of case studies in different fields, Dominic explored how it was possible to constrain knowledge, semantically represented in unstructured natural language text, into a structured format which could be readily mined for useful knowledge. Whilst a practical reality, Dominic argued that in doing so, the analyst runs the very real risk of gaining quantitative oversight at the expense of qualitative insight. In the end, is forcing a structure on unstructured data really delivering the benefit that the analyst is looking for? How can you start to measure and ascertain the qualitative and holistic understanding of the text (literally reading between the lines) that is lost in order to achieve an understanding of the text in terms of numbers? By reducing text to numbers and structure, do the natural language processes really answer the questions being considered? By reviewing how ontologies and taxonomies have been widely applied in the field of text analytics and natural language processing, Dominic presented a method for creating and developing bespoke ontologies based on a qualitative social research approach, which compliments more quantitative approaches with additional qualitative context.
Extending the thoughts, research and case studies presented by Dominic, the presentation turned towards how Bayesian approaches could provide an opportunity to develop dynamically and automatically updating ontologies. Starting with a brief introduction to Bayes Theorem, Sophie explored the possibility of developing approaches to determine conditional probabilities for phrases in order to understand and predict how key words and phrases of interest change over time within written reports.
Both parts of the presentation are areas of interest for Bricolage and Bays Consulting, we’ll keep you updated on our work as it continues. For the latest updates, make sure you follow us @bricolage_RandA and @baystatsic on Twitter, LinkedIn, Facebook and Instagram.