top of page
  • mhoeschele

Interesting Comment Detection

I’ve interviewed thousands of people across the federal government/industry, and there is one thing that always surprises me as a data minded person.


If you talk to 100s of people, less than 5 people will stand out as providing the most interesting or critical insights.


Here at InsightAI, we love to hear what people have on their minds and make sense of it. Our meaning based topic analysis is key to doing this at scale, but we have found that identifying a handful of highly insightful responses is critical to increasing the actionability and impact of these insights.


  • Example

    • Topic: IT Outdated

    • Summary: The IT systems for the contracting team are outdated and slow, regularly causing deadlines to be missed.

    • Insightful Comment: “My computer regularly crashes 30min after opening the contract writing software, causing me to lose all my progress and miss deadlines.”


As you can see, with the addition of the insightful comment, much more detail about what is going on is identified, making it easy to identify how to get to work on addressing the issue and likely someone you would want to learn more from.


Our proprietary AI model that identifies these insightful comments is called Critical Comment detection. It scores how interesting each response is within a topic cluster and across all responses to a question. Of course this begs the question, how to define “interesting?”


To design this analytic I interviewed a number of people about what makes something interesting to them and found the below heuristics:


  1. New information on a topic I care about

    1. “Our fuel pumps keep falling apart.”

    2. “There is lots of wear on the pumps for fule.”

    3. Interesting: “The fuel pumps are degraded, this could cause an explosion.”

  2. Easy to understand and detailed descriptions of a topic

    1. “The software keeps crashing when rendering video.”

    2. “Video rendering tools crash all the time.”

    3. Interesting: “Cinema 4D crashes when I try to render a fluid simulation on my Alienware laptop

  3. Someone unexpected mentions a topic

    1. Engineer: “Modeling software is too slow for my bigger projects.”

    2. Interesting: “The modeling and simulation tools are inadequate to deliver projects on time.” ~ HR representative


Identifying highly accurate topics across 10,000s of people’s responses has been the foundation of the analysis we deliver from day 1. Now you don’t have to dig through those responses to find great quotes or detailed exemplars to highlight!


10 views0 comments
bottom of page