We've written our fair share of documentation, internal, external and everywhere between those. We've experienced first hand that when you write technical notes or documentation at the end of a long day, no matter how optimistic you are, things might get a little cynical and possibly negative. Obviously, this is not on purpose, but it will be perceived as negative by the reader. In our current team, a quick message or comment pointing out negative sentiment is usually enough, but ideally we would be able to immediately identify any sort of negative sentiment. That's why we built Tone Analyser for Confluence, to help us write positive, constructive content in Confluence.

What it does

Tone Analyser uses Natural Language Processing (NLP) and Machine Learning (ML) to analyse the sentiment of Confluence Pages. After publishing a page, Tone Analyser will start analysing your document and in mere seconds you will be informed about the sentiment via the Confluence byline section, at the top of the page using easy to understand and clear communication.

How we built it

After researching sentiment analysis and NLP, we came up with several visual prototypes to find the best way to visualize sentiment. Using Atlassian Connect, a sentiment API and some of our own frameworks we prototyped a total of 2 concepts, of which we implemented one in the final version.

Challenges we ran into

Sentiment can be calculated in multiple ways, each of which comes with a different sort of outcome in terms of value output. We've prototyped several options before we found the best way to visualize sentiment.

Accomplishments that we're proud of

In terms of team accomplishment, we're most proud of the fact we were able to release this app along with other apps, regardless of the time restriction and the need to make major changes to the way the app works. Most of all, the app is very much in line with our idea that plugins should be simple: easy to use, little explanation needed and non-obtrusive. Ironically enough, the simpler the app for users, the more difficult it becomes to build it.

What we learned

The importance of templates, frameworks and a clear goal has proven to be even more important when time is of the essence. In terms of machine learning and NLP, we've learned loads about types of sentiment, ways to analyse text and how people interpret sentiment..

What's next for Tone Analyser for Confluence Cloud

Although figuring out the sentiment of Confluence pages is a great help for people who write, we have about a dozen ideas on how we can improve content writing on Confluence using machine learning and NLP that we are excited to add to the application.

Share this project: