Evaluation of a Supervised Approach to Information Retrieval

Bc. Matej Meško

Bachelor's thesis

Abstract:

Táto bakalárska práca sa zaoberá konceptami z Information Retrieval. Zameriava sa na problematiku využitia modelov založených na supervised prístupe v IR. Porovnáva viacero verejne dostupných datasetov a vyberá reprezentatívnu vzorku s rôznorodými vlastnosťami, na ktorých následne testuje niekoľko supervised a unsupervised IR modelov. Na základe týchto výsledkov vyhodnocuje efektivitu týchto dvoch prístupov k IR, vhodnosť vybraných datasetov pre supervised vyhľadávanie, a ich ďalšie následky.

Abstract:

This bachelor’s thesis deals with the concepts of Information Retrieval. It focuses on the apparent difficulties in employing supervised-based models in IR. It compares several publicly available datasets and selects a representative sample with diverse properties, on which it subsequently tests several supervised and unsupervised IR models. Based on these results, it evaluates the effectiveness of the two considered approaches to IR, the suitability of selected datasets for supervised retrieval, and further implications.

Keywords

information retrieval, dataset, supervised, unsupervised, TF-IDF, BM25, reranking, logistic regression, cross encoding

Language used: English

Date on which the thesis was submitted / produced: 19. 5. 2022

Thesis defence

• Date of defence: 28. 6. 2022
• Supervisor: Mgr. Michal Štefánik
• Reader: Mgr. Martin Geletka

Masaryk University

Faculty of Informatics

Bachelor programme / field:
Informatics / Informatics