Relevance Feedback and Information Retrieval

Mall, Priyanshu Raj (2023) Relevance Feedback and Information Retrieval. Masters thesis, Department of Physical Sciences.

Text (MS dissertation of Priyanshu Raj Mall (18MS118),) Thesis_18MS118.pdf - Submitted Version Restricted to Repository staff only Download (5MB)

Abstract

In recent years, there has been a significant increase in the production of information due to the easy availability of computing resources, and this trend shows no signs of slowing down. This information is usually stored for future reference, either locally or publicly. The establishment of standards and the interconnection of these systems has led to the creation of the World Wide Web, which contains mainly text documents, but there has been an increase in multimedia content over time. As a result of the massive amount of data, automated systems are necessary to help users locate specific information. Information Retrieval (IR) is the field concerned with organizing, storing, and retrieving information, with a particular emphasis on textual data, as evidenced by the numerous search engines available today. This thesis aims to explore various information retrieval models and their effectiveness in retrieving relevant information from large collections of data. Specifically, we begin by examining popular ranking models such as the LMDir and BM25. We then focus on implementing the RM3 [Abd+04], a widely used language model for query expansion. We seek to further investigate the performance of RM3 and the Rocchio algorithm in Pseudo Relevance Feedback (PRF) and True Relevance Feedback (TRF) scenarios using the TREC Disk 4, 5 collection, with the objective of comparing their performance. To improve the performance of the relevance-based language model, we also investigate different smoothing methods. Finally, our ultimate goal is to integrate Negative Feedback in RM3, which involves identifying irrelevant documents and updating the query model to improve the retrieval of relevant information. By conducting these experiments and analysis, we aim to contribute to the ongoing efforts in improving the effectiveness of information retrieval models.

Actions (login required)

View Item