[WIR] Applying Science Models for Re-Ranking in IR

Introducing bibliometric-enhanced metadata to IR
This is the repository for the course 'Web Information Retrieval' for the study 'Digital Sciences' at the University of Applied Sciences Cologne.
It contains the project code for the project 'Applying Science Models for Re-Ranking in IR'.

An exploration of utilizing power laws in combination with graph-based metrics and bibliometrics such as coreness in an IR setting

• Four Graph Implementations: Co-citation Graph, Citation Graph, Lotka-Inspired Graph, Journal Graph
• Metrics: Degree Centrality, Closeness, Betweenness, Distance to most popular node
• Comparing BM25 Baseline vs. Re-ranking with Boosting Factor
• Dataset: TREC-COVID / Cord19 dataset
• Metadata Enrichment: SemanticScholar API

Structure of this Repository

• data\: Metadata used for the project
• doc\: Documentation and presentation of the project
• scripts\: Jupyter notebooks used for the project

Notebook Description

Filename	Description
Experiments.ipynb	Creating, experimenting & evaluating retrieval perfomance with the graphs
Scrape_metadata.ipynb	Scraping relevant metadata using SemanticScholarAPI

Sources

cord19
Semantic Scholar
pyterrier