Matching Support Seekers with Support Providers and Responsible Decision Making with Knowledge-infused Bandits
Super Excited to share news on the acceptance of two research ideas:
- Matching Support Seekers with Support Providers
- An automated method to identify supportive users is challenging due to diverse roles of users, but it is more challenging to #associate them with #supportseekers for suitable help.
- Annotating such matches is exceptionally time consuming for experts and #moderators on online platforms. We demonstrate an expert validated methodology that uses #diverseknowledge to match support seekers and support providers.
- 7.2 out of 10 confidence ratings from subject matter expert and users of Reddit for support
- We tested our strategy on 25000 users from Coronavirus and covid19_support community pairs on Reddit who express anxiety and depression with certainty.
- Preprint
- Kudos to team: Kaushik Roy, Aditya Sharma (intern), Biplav Srivastava, Amit P. Sheth
- Knowledge Infused Policy Gradients with Upper Confidence Bound for Relational Bandits
- Knowledge Infusion in the Relational Bandits learning setting.
- Deriving a Upper Confidence Bound style strategy for effective explore-exploit trade-off.
- Performing regret analysis of our approach when we include human expertise and extensive experimentation on several benchmarking datasets.
- Preprint
- Kudos to Team: Kaushik Roy, Qi Zhang, Amit P. Sheth