This uses Microsoft's Text Analytics v2.1 API to analyze the sentiment of comments in a GitHub pull request.
The scores are the APIs best guess at a sentiment reading. Scores closer to 1 == "friendly".
senti.rb does its best to remove GitHub's code suggestion blocks from comments (```suggestion```), as it tends to confuse the sentiment analysis.
- First, sign up for the trial of Microsoft's Cognitive Services
- Get your
TEXT_ANALYTICS_SUBSCRIPTION_KEYand yourTEXT_ANALYTICS_ENDPOINT. - Generate a GitHub access token with repo permissions: https://github.com/settings/tokens
senti.rb is written in Ruby; you'll need to have a Ruby and bundler installed on your system. Macs have Ruby installed by default.
bundle install
You'll need to set these environment variables in your shell, with an export or at runtime.
To analyze PR #1 in this repo:
export TEXT_ANALYTICS_SUBSCRIPTION_KEY=your_key TEXT_ANALYTICS_ENDPOINT=https://westcentralus.api.cognitive.microsoft.com GITHUB_ACCESS_TOKEN=your_github_access_token GITHUB_REPO="seanknox/pr_sentiment_analysis" PR_NUMBER=1
bundle exec ruby -W0 senti.rb
===== SENTIMENT ANALYSIS =====
Comment: Looks great! I love it. You are the best. : Sentiment Score: 0.9979691505432129
Comment: I don't like this! And I don't like YOU!!!: Sentiment Score: 0.002584695816040039
Comment: This is an ambiguous comment? Maybe?: Sentiment Score: 0.7764753699302673
Comment: I don't like you but I put a smiley so thanks to machine learning not knowing passive aggressiveness it'll think this is nice 😄 : Sentiment Score: 0.9964126944541931
Overall sentiment average: 0.6933604776859283 / median 0.8864440321922302
GITHUB_REPO="seanknox/another_repo" \
PR_NUMBER=93 \
bundle exec ruby -W0 senti.rb