search

MyMood-Alexa / MyMood

0 stars 1 forks source link

[SA-3.1] Research API's for sentiment analysis #26

Closed danqo closed 6 years ago

danqo commented 6 years ago

-Compile a list of ready-to-use sentiment API's we can consider using -Pros and cons of API's found

child of #19 connects #19

danqo commented 6 years ago

All of the following have the following cons: -May have to see if this violates terms of service, since we're technically sending data to a third party. -Another dependency variable that we have to track and have limited control over.

  1. Stanford Sentiment140 API Since we are using the Stanford data set, they also have an API trained off that same data set. API Documentation: http://help.sentiment140.com/api Architecture used: https://cs.stanford.edu/people/alecmgo/papers/TwitterDistantSupervision09.pdf Pros: -Research and development is backed. Cons: -Closely related, but not customized for our use case.

  2. IBM Watson API: https://www.ibm.com/watson/developercloud/tone-analyzer/api/v3/curl.html?curl Demo: https://tone-analyzer-demo.ng.bluemix.net/ "I aced my exam" => Fear Cons: -Does not seem accurate for our use case (Uses multi-label classification instead of multi-class classification)

  3. Text-Processing.com API: http://text-processing.com/docs/sentiment.html Demo: http://text-processing.com/demo/sentiment/ "I aced my exam" => 0.6 negative, 0.4 positive "I pooped in my pants" => 0.6 negative, 0.4 positive Pros: -Simple and easy to use Cons: -Does not seem to be well developed

**Also has tutorials for using NLTK. E.g. Stemming http://text-processing.com/demo/stem/

  1. Microsoft Cognitive Services API: https://westus.dev.cognitive.microsoft.com/docs/services/TextAnalytics.V2.0/operations/56f30ceeeda5650db055a3c9 Price: 1,000 transactions free per month Pros: -Supports multiple languages

  2. TheySay PreCeive API: https://theysay.docs.apiary.io/#reference/0/analysis-recipes Demo: https://apidemo.theysay.io/ "I dropped my ice cream." => 0.149 neutral, 0.851 negative "I aced my exam." => 0.8 positive, 0.2 neutral "I pooped in my pants." => 0.149 neutral, 0.851 negative Price: Free => Maximum 500 requests per day, reset at midnight UTC. Pros: -Python support, easy to use