Contributor :
Chaithanya Kalyan
Mentor :
Denny George
When dealing with online harms on social media in India you come across challenges that are unique to India. Firstly a majority of the data is in the form of images, videos and audios and secondly its in various Indian languages. These challenges make it hard to study the extent of harms on a large scale in an automated manner. Some of the foundational tools and datasets that we are building aims to close this game. Feluda is one such tool that makes analysing multimodal and multilingual content easier. As part of this project we reviewed state of the art ML models, evaluated their capabilities and limitations in working with the kind of data we find on Indian social media and built the capability to analyze a collection of audios and run automations to group similar audios together and assign tags to them. This will feed into downstream annotation and review workflows that will help surface thematic trends in audios.
Contributor :
Chaithanya Kalyan
Mentor :
Denny George
When dealing with online harms on social media in India you come across challenges that are unique to India. Firstly a majority of the data is in the form of images, videos and audios and secondly its in various Indian languages. These challenges make it hard to study the extent of harms on a large scale in an automated manner. Some of the foundational tools and datasets that we are building aims to close this game. Feluda is one such tool that makes analysing multimodal and multilingual content easier. As part of this project we reviewed state of the art ML models, evaluated their capabilities and limitations in working with the kind of data we find on Indian social media and built the capability to analyze a collection of audios and run automations to group similar audios together and assign tags to them. This will feed into downstream annotation and review workflows that will help surface thematic trends in audios.
About Contributor
Chaithanya is an undergraduate student pursuing Computer Science and Engineering with specialisation in Artificial Intelligence at Amrita Vishwa Vidyapeetham. He likes solving real world problems with the help of AI and very interested in developing models at scale. His other interests includes listening music and dancing.
Key Impact Takeaways:
- This project streamlines the workflow for clustering audio and video files, reducing manual effort and increasing scalability.
- The tools and operators developed will influence the analysis of large-scale multimedia datasets, with potential applications in AI research, content recommendation systems, and media archives.