hurshd0 / trec-covid-ir

TREC-COVID 2020 Task - Building simple yet effective pandemic retrieval engine

---

TREC-COVID Challenge is an Information Retrieval (IR) challenge sponsored by NIST in colaboration with Allen Institute for AI and other industry leading research groups. I used Round 3 dataset which contains 40 topics, where each topic represent user's infomation need. The goal of challenge is to retrieve relevant documents for each topic, and evalute them against human judgements.

Method

I tried TF-IDF based approach raning from builting one from scratch using Python, and comparing it with two famous open-source commercial search engine - Solr and Elasticsearch. To improve rankins I also tried Automated Query Expansion using Word Embeddings and Deep learning based Re-ranking technique.

In this repo

• scripts - contains bash scripts to download dataset, and install trec eval tool
• notebooks - contains documentation, trec-eval results, and notes on results