Text Analytics

GovTech's cool ever-changing logo

Check out the Pure CSS cat animation by Johan Mouchet (@johanmouchet) on CodePen.

GovText's Landing Page

I have written about GovText (Beta) briefly last year when I first explored it while still teaching in a school. Recently, I explored the text analytics platform for (Sg Gov) Public Service Officers (with a .gov.sg email) again especially amidst all the amazing public-facing work GovTech has been doing since the COVID-19 pandemic disrupted our lives (TraceTogether, MaskGoWhere and more). I decided to look into GovText a bit more this time so as to build on my schema on how it came about and what it could be used for.

I came across this article from Civil Service College entitled 'Data Science in Public Policy - The New Revolution'. Inside there is an insightful segment about Machine Learning where GovText was mentioned and how it was meant to improve ground-sensing methods for public officers.

Machine Learning
Machine learning, a branch of artificial intelligence (AI),4 is a statistical process that starts with a body of data and tries to derive a rule that explains the data or can predict future data. Unlike older AI systems where human experts determine the rules and criteria for the system to make analytical decisions, machine learning can be used even where it is difficult or not feasible to write down explicit rules to solve a problem.
Machine learning is already an essential feature of many commercial services such as trip planning, shopping recommendation system, and online ad targeting. It has also been applied in strategic games, language translation, self-driving vehicle, and even public services. In the public sector, machine learning software has helped the US Military to predict medical complications and improve treatment of severe combat wounds,5 and cities to schedule, track and provide just-in-time access to public transport.6 In Singapore, the Housing Development Board (HDB) in collaboration with GovTech used machine learning to identify customer concerns more accurately and adapt its policies to cater to citizens’ needs.
Building on the success of the HDB project, GovTech developed a text analysis platform, “GovText”,7 to enable public officers to apply unsupervised machine learning to discover topic clusters from textual data without any coding knowledge. GovText not only scales data science capabilities across all levels within the whole of government but also allows officers to improve their “ground-sensing” methods.

That was exactly what I attempted to do (ground-sensing) with textual data and qualitative comments I had access to during this Full Home-Based Learning (FHBL) period. It was quite a fun exploratory process for me except for the tedious task of collating and cleaning the dataset. The other interesting thing was the process of coming up with suitable topics based on the analysed and clustered/grouped texts. Then, the visual representations of the data makes more sense and tells a story about ground sentiments. If you are wondering how to do it. Go check out the User Manual and Sample Datasets on the site.

Further Analysis

1st level analysis

Barchart with Probabilities

Scatterplot with linkages

Coming up with topics based on 'word cloud'

From IBM's NLU site

 I also wrote about IBM Watson's text analytics demo last year. This time I just experimented and plugged in similar data in their demo with Natural Language Processing (NLP) capabilities on their Natural Language Understanding (NLU) site. It is able to go deeper and detect sentiments and even categorise based on its machine learning data. Do check the 2 sites out if you ever have to do some text analytics.