Explainable AI for Sentiment Analysis For English Language

[akolonin]

This MSc student research task may or may not end-up with actual code submitted into this repository

Main Goal: Exploration of applicability of explainable artificial intelligence technique for sentiment analysis applied for English language
Russian: Исследование применимости технологии объяснимого искусственного интеллекта для анализа тональностей применительно к английскому языку

Subgoals:

1. Explore possibilities of explainability of sentiment analysis problem solutions for English language in two different aspects:
   1.a) Given a model trained on a corpus, extract the significant features or combinations of features indicative to particular sentiments
   1.b) Given a discernment of a sentiment for a text, explain which significant features or combinations of features in the exposed text were involved into making the discernment
2. The models used for the study may be identified during the study, better have few models to compare
3. The models in use should be trainable for different domains like, need to use at lease one domain in the scope of work, but trying more than one domain may be a plus
   3.a) positive/negative/neutral
   3.b) interrogative/declarative/directive/imperative
   3.c) like/love/wow/haha/sad/angry (Facebook style)
   3.d) certain/uncertain/neutral
4. The models in use should be to provide multiple sentiment discernments per exposed text like sentiment may new "neutral" or "negative" or "positive" or "positive and negative" (mixed feelings)

For the reference, the following information may be used (or not used):

1. Collection of resources on past works on Sentiment Mining: https://docs.google.com/document/d/1dG_wz8UJVi0CTy3FzWzSHRcrpYHG33utqoK27epbIQM/edit
2. MSc thesis on sentiment analysis by Juan Fernando Pinzon ad NSU, year 2019
3. Collection of posts from Steemit social network

[akolonin]

The first round of work would include the following:

1. Development on the written overview of the subject field, enumerating
   1.a) different models (available as open source) of sentiment analysis for English
   1.b) different models or approaches of explainable sentiment analysis for English
2. Exploring availability of the testing/training corpora ad selecting the one for further work
3. Trying to apply explainable sentiment analysis using Lime in based on one of the models found above
4. Continuous learning on explainable AI, involving:
   4.a) Book https://www.springer.com/gp/book/9783030289539
   4.b) Approaches highlighted in the following presentation:
   https://www.youtube.com/watch?v=ZusjVvxEinw
   http://aigents.com/papers/2019/AI-state-2019-en.pdf
5. Drafting the semester-by-semester plan for the work on thesis with milestones and deliverables identified