Dobby

automated labeling
automated translation

Inspiration

We picked two major challenges developers face daily with focus on open-source developers.

Problem 1

The sheer volume of issues reported on GitHub's open-source projects can easily overwhelm users. It might be difficult for project maintainers to adequately manage and prioritize these issues. Using labels to group situations according to their seriousness, importance, and other pertinent factors is one technique to deal with this issue. But for larger projects with a high volume of issues, manually adding labels to every issue can be time-consuming and error-prone. To simplify the issue management process and guarantee that issues are resolved quickly and effectively, GitHub maintainers must be able to have issues automatically labelled based on keywords or other criteria.

Problem 2

Teams working on software development are frequently made up of people with different language origins in today's globally connected world. One of the most widely used sites for collaborative software development is GitHub, which is utilized often by engineers all around the world. But when concerns are posted on GitHub, they frequently have comments that not all team members can comprehend. Due to the communication barrier this generates, teamwork may suffer and problems may take longer to resolve. Because of this, developers must be able to translate comments in GitHub issues in order to effectively communicate with other team members who speak different languages.

What it does

Dobby is a GitHub bot that offers automated labeling and translation functionalities for your repository.

How we built it

Automated labelling

Dobby can automatically label issues based on the content of the issue (the issue title and description). Essentially what auto-label does is when a new issue is created, the bot samples the already labeled issues; It predicts the mostly likely label(s) for the new issue which is chosen if the confidence criterion is met. This functionality uses the Cohere's embed model embed-multilingual-v2.0.

Translation

Dobby can also translate the issue into the specified language and post a comment with the translated text. To translate an issue, mention Dobby in a comment with the query @dobby-gh-bot translate to , where is the target language to be translated to.

This functionality uses the Cohere's generator model command-xlarge-nightly.

Challenges we ran into

Confidence of the labels were quite low

Accomplishments that we're proud of

The translation worked pretty well for our use-case
Although the confidence were low the label worked fine for case where the confidence were high

What we learned

The strength and weaknesses of Transformer models and its potential use-case

What's next for Dobby

As we continue to improve our issue labeling bot, there are a few areas we want to focus on:

Subsampling of examples: To optimize our labeling process, we want to be able to subsample already labeled issues.
Support for multiple-labeling: Being able to apply multiple labels to an issue will help us get a more nuanced understanding of each issue.
Platform-agnostic integration: We want to be able to run our bot on any platform, such as Slack or Discord, to make it as widely accessible as possible.