Skip to content

Iterative Relevance Feedback based Answer Passage Retrieval

Notifications You must be signed in to change notification settings

pranavdeepak13/IIR

Repository files navigation

Information Retrieval Coursework Project

Overview

This project for the Information Retrieval course delves into the development and comparison of various retrieval algorithms. Using a TREC dataset annotated for the task, we focused on understanding and implementing key algorithms in the field, demonstrating their efficiency and effectiveness in retrieving relevant information.

Key Features

Implemented Algorithms

  1. Boolean Retrieval:

    • Implemented the fundamental Boolean retrieval model, allowing for exact matching of queries with documents using logical operators.
  2. Similarity-Based Retrieval:

    • Developed algorithms that rank documents based on their similarity to a query, providing more nuanced and relevant results compared to Boolean retrieval.
  3. TF-IDF (Term Frequency-Inverse Document Frequency):

    • Utilized the TF-IDF model to evaluate how important a word is to a document in a collection, aiding in the identification of relevant documents.

Pseudo-Relevance Feedback

  • Incorporated pseudo-relevance feedback mechanisms to refine search results. This approach assumes that top-ranked search results in an initial query are relevant and uses this information to improve the search accuracy in subsequent iterations.

Efficient Information Retrieval Techniques

  • Explored and implemented various techniques to improve the efficiency and effectiveness of information retrieval, focusing on the optimization of search algorithms and query processing.

BM25 for Ranking

  • Used the BM25 algorithm, a more advanced approach compared to traditional TF-IDF, for ranking documents. BM25 provides better handling of term frequency saturation and document length normalization.

Dataset and Practical Application

  • The project utilized the PSGRobust dataset, enabling hands-on experience with real-world data.
  • Emphasis was placed on practical application, comparing different retrieval methods and understanding their strengths and limitations in various scenarios.

Educational Objective

  • The primary goal was to gain a comprehensive understanding of different information retrieval algorithms and their applications.
  • The coursework facilitated a deeper insight into the mechanics and theoretical underpinnings of IR systems.

Citation

This project was heavily influenced by the work of Keping Bi, Qingyao Ai, and W. Bruce Croft in their paper:

  • Keping Bi, Qingyao Ai, W. Bruce Croft. "Iterative Relevance Feedback for Answer Passage Retrieval with Passage-level Semantic Match." Proceedings of the European Conference on Information Retrieval (ECIR 19), Cologne, Germany, April 14-18, 2019, pp. 558-572.

This seminal paper provided crucial insights and methodologies that shaped our understanding and implementation of information retrieval algorithms.


For more information or inquiries about the project, please contact the project team and any queries feel free to create a PR

About

Iterative Relevance Feedback based Answer Passage Retrieval

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published