Personalized IR on SE-PQA

Live Demo

Try the web application here: Personalized IR App

Downloading Files

Download all the files at the following link: pir-notebooks-data.zip

In this file, we have the following folders:

• cache: used for storing the output of a PyTerrier Transformer
• experiments: all the experiments that we executed
• index_sepqa: all the PyTerrier indexes that we have created
• models: Scikit-Learn trained models

Notebooks

• notebook1-data-analysis: Look at the data
• notebook2-baseline-retrieval: BM25 and TF-IDF
• notebook3-neural-reranking: Reranking with a Bi-Encoder
• notebook4-query-expansion: Expand the query with an LLM
• notebook5.1-personalize-ir: Personalize the query with the Tags Score
• notebook5.2-personalize-ir: Other scores for personalization
• notebook5.3-cold-start-problem: Formulation of the Tags Score dependent on the number of questions written by the user
• notebook6-ltr-personalized-ir: Learn to rank on top of the personalized information retrieval pipeline

Requirements

The file environment.yaml is the output of the command:

conda env export --no-builds > environment.yaml

Acknowledgment

This project uses the SE-PQA dataset from:

Kasela, P., Braga, M., Pasi, G., & Perego, R. (2023). SE-PQA: a Resource for Personalized Community Question Answering [Data set]. Zenodo. https://doi.org/10.5281/zenodo.10679181