(TFG) Automated detection of events through NLP-based analysis of mobile app reviews

Description: Mobile app stores have become a source for data-driven analysis combining developer documentation (i.e., app descriptions, changelogs, technical documents) and user generated data (i.e., app reviews, ratings). Concerning the latter, user feedback analysis is essential to build deep, up-to-date perspectives on a given domain. Specifically, user reviews provide immediate information about the internal and external events affecting the lifecycle of an app (e.g., feedback on a given feature, reactions to a new app release, social events affecting the vision of users for a given app). In this thesis, the student will design and develop a web-based service for the automatic detection and classification of app-related events extracted from app user reviews. These events will be extracted using generative large language models (e.g., T5, GPT-4) on an unsupervised approach using prompt-engineering techniques from a sub set of reviews for a domain-specific catalogue of applications.

Degree: GEI

Research Areas: Natural Language Processing (NLP), Large Language Models (LLM), Mobile App Stores

Technologies: Python, HuggingFace, PyTorch, Flask

Contact: Quim Motger

e-mail: jmotger (at) essi.upc.edu