Welcome to the repository for the Information Retrieval course at the Indian Institute of Technology, Kharagpur, for the Autumn 2022 semester.
This repository contains the code for assignments related to Information Retrieval. The primary focus is on building effective retrieval systems for documents using different techniques.
In the initial assignment, we built a simple inverted index. The Boolean Retrieval system treats each query as an AND of individual words, utilizing this encoding to retrieve all relevant documents.
For the subsequent assignment, we advanced to a tf-idf based ranked retrieval system. The tf-idf weight, representing the term frequency of a term in a document, was used for each term. This enhanced system aims to answer free-text queries.
To assess the performance of our information retrieval system, we gathered the top 20 documents and compared them to the gold standard, provided in qrels.csv. Evaluation metrics such as Average Precision and Normalized Discounted Cumulative Gain were employed.
In the final phase, relevance feedback methods were applied to enhance system recall. For relevance feedback, a gold standard relevance judgment score of 2 was considered as relevant, and the rest as non-relevant. Pseudo Relevance Feedback was implemented by deeming the top 10 ranked documents as relevant, and then modifying the query vector using Rocchio's algorithm.
All the code and results related to the assignments have been uploaded in this repository. You can explore the respective folders for detailed implementations, code documentation, and evaluation results.
Feel free to clone or fork this repository to explore the code, replicate the experiments, or adapt it for your own learning purposes.
We appreciate your interest and hope you find the contents of this repository valuable for your understanding of Information Retrieval.
Best Regards, Kushagra Chitkara