Text analytics have taken major steps forward with the advent of Chat GPT and OpenAI.  Collecting and analyzing text from survey comments is an important part of survey analysis but it can be challenging to deal with so much unstructured data and apply statistics to the data.  In early 2022 we took thousands of workplace survey comments collected between 2008 and 2023 to create a Natural Language Processing (NLP) topic model for a global design firm, Gensler, that would result in a data-driven survey comment analysis process for it’s team of strategists.  

Before this, analysts were using crude methods to extract insight from the written comments that were limited to key word searches and assumptions about sentiment and themes based on a handful of comments.    In the presentation we will share findings from two years of research into how our team at Gensler improved our existing workplace survey tool by creating a data-driven process for analyzing open-ended survey comments using machine learning.  

In this session the audience will learn about the use case for applying Natural Language Processing (NLP) to our workplace survey comments, the size and type of data in our set and how we went about creating topic models, labeling functions, and supplementing data with other sources.  We will share information about the size of our data set, the survey questions we used to build our model using NLP methods, and interesting challenges on our way to building the topic model.

What will I learn from this session?  
How to think about unstructured data your organization already collects
How to build a topic model tuned to your specific needs
Methods to improve model accuracy
Value of in-process reviews to help the team pivot
Difference between silver and gold data in text classification
Ways to collect gold labeled text data
How to refine your methods as more text tools become available