Utilize Sentiment Analysis for identifying and/or applying a desired "tone" for content generation

Open
#781 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
20/100
Issue type
Feature
Clarity
Needs clarification
Activity status
Stale
Tech stack
aws, azure, machine-learning, php
Domain
ai, content

Research direction

Start by reviewing the existing AI service integrations and the related issues #489, #81, and #117. Compare the available sentiment-analysis services and define a small proof of concept that measures existing content, stores adjustable tone values, and evaluates whether prompts produce visibly different tones before proposing formal integration.

Written by the indexing model from the issue text.

Description

help wanted needs:engineering
Is your enhancement related to a problem? Please describe.

As related to (perhaps a corollary of) #489 and as mentioned in #81 and #117, we could look to leverage Sentiment Analysis as an alternative to the approach referenced in 489 to identify the "tone" for a site and as input to a prompt for content generation to ensure future content generations align to that "tone".

Designs

Various sentiment analysis tools are available within AI service providers we already integrate with (e.g., Watson's NLU emotion and sentiment text analytics features, Azure's Language sentiment analysis feature), with other AI providers that we've considered (e.g. Amazon's Comprehend sentiment feature), and likely others that would arise with some searching on the topic.

The concepts that we could explore with a rough proof of concept / prototype to see if this could work as an alternative for #489 is as follows:

  • utilize Sentiment Analysis on the existing site content to identify the average plotting of said content across various sentiment scales (formal-to-informal, humorous-to-non-humorous, academic-to-non-academic, happy-to-sad, optimistic-to-pessimistic, etc.) and store that as the site "tone"
  • present those averages of existing content on those different sentiment options with the ability to adjust where the content ideally exists on respective scales to match the "desired tone" for a site
  • utilize the stored sentiment option scales in prompting for content generation to attempt to affect the resulting output from an AI Service provider to match the "tone" for site content
Describe alternatives you've considered

One approach for a proof of concept here would be to have stored values for where sentiment options are on respective scales and send the same prompt for content generation with wildly different scale values to see if the resulting output content varies and appears to match the "desired tone" for the sentiment option scale values. If that appears to work, then building out the proof of concept further for assessment before building more formally into ClassifAI would be helpful.

Code of Conduct
  • I agree to follow this project's Code of Conduct
Dominant language
PHP
Stars
714
Forks
68
Avg merge
4h 32m
Merged PRs (30d)
23

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from 10up/classifai

All issues in 10up/classifai

Similar issues

More PHP issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.