Clubhouse (https://www.clubhouse.com/) and Twitter Spaces (https://help.twitter.com/en/using-x/spaces) are prominent examples of audio-based social networks (relatively new types of social networks that are gaining a lot of popularity). Users spend time on these application speaking and listening with others in scheduled rooms. What the users speak about and listen to are indicators of their preference and can be used to profile users.

This project will study the design and development of an app that uses Large Language Models to analyse spoken data (see for example, LeMUR from AssemblyAI). Given an audio data file, you can build a system that transcribes and summarise the data. Results of this summarisation can be used in building user profiles.

Evaluation of the system with data collected from Twitter Spaces will be used to demonstrate how user profiles and recommendations is done on such platforms, with and without the summarised content.

Keywords: LLMs, audiobased social networks, text summarization, user profile creation and recommender systems