True Insights for Jira

Inspiration

User-generated issues can hint at larger problems and uncover opportunities, but reading through every single issue is a painful experience, specially with popular products that have a constant flow of issues. There are already many tools that process quantitative data to gain insights, but we wanted to take it one step further using Natural Language Processing and turn qualitative data into actionable insights.

What it does

True Insights applies a custom language model to the issues in a Jira project and generates two main views:

Cluster: A graph where the nodes represent words and the edges represent how frequently they appear together.
Wordcloud: A sentiment wordcloud colored either black, green or red depending on whether the users's sentiment toward that keyword is, respectively, neutral, positive or negative.

You can use the cluster to discover problems that are common to different issues. For example, you could see a pair of keywords mentioned together in several separate issues, discover they are duplicates and link the issues accordingly.

Also, the cluster shows semantically related keywords. Without having to manually comb through hundreds of issues, you can see a common problem even when it's worded differently in each issue and assign these issues a specific label.

How we built it

React.js as the frontend framework for Forge's Custom UI
The Atlaskit library for native React components that fit into the Atlassian UI
AWS API Gateway and API Lambda for the backend infrastructure
Our own custom Natural Language Processing models to process text

Challenges we ran into

The main problem we had was not having a large project on which to test our Natural Language Processing models, so we used a slice of Jira's public issue tracker for testing.

Accomplishments that we're proud of

We believe that Natural Language Processing can greatly improve the user experience of an issue tracker and bridge the current gap between automation and qualitative data. Our app is a small demonstration of what a language model can uncover from user input like issues.