Building a Simple Information Retrieval System using BM25 and GPT-3 and evaluated in the CISI collection
Enrollment exercise fot the course "Deep Learning aplicado a sistemas de buscas", FEEC-Unicamp, first semester of 2023.
A brief report describing the implementation details, results, how to test the IR system and how chatGPT helped with the project. The markdown version can be found here.
Jupyter notebook with all the associated functions and libraries, with the code for the IR system.
We know that one of the measures of a good search engine is its UI and how the user interact with it alongside with the query language expressiveness and the latency of the response.
Thus, we build a poc of a search engine interface for our models using streamlit. This poc can be accessed on streamlit cloud or hugginface spaces.