SloBench : Slovenian Natural Language Processing Benchmark

DRAGAR, FRENK (author), Žitnik, Slavko (mentor)

Evaluation of natural language processing (NLP) tasks is an essential part of research and progress in the field. It provides an objective standard for comparison and performance on a specific task. We give an overview of recent public benchmarks and evaluation trends, with focus on the automatic evaluation of systems. We then propose, implement and document a general and extendable model-agnostic evaluation framework, along with the first online platform for the automatic evaluation of Slovene language NLP tasks with public leaderboards, showing the performance of submitted systems.

• Language: English
• Keywords: natural language processing, benchmarking, leaderboard, machine learning, web platform
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science ; FMF - Faculty of Mathematics and Physics
• Year: 2022
• PID: 20.500.12556/RUL-136261
• COBISS.SI-ID: 105616643
• Publication date in RUL: 21.04.2022

Secondary language

• Language: Slovenian
• Title: SloBench : Slovenski vrednotnik metod za obdelavo naravnega jezika

Abstract

Evalvacija nalog procesiranja naravnega jezika (NLP) je bistven del raziskav in napredka na tem področju. Zagotavlja objektiven standard za uspešnost in primerjavo sistemov pri določeni nalogi. Podamo pregled nedavnih javnih lestvic za najboljše sisteme in trendov njihovega ocenjevanja s poudarkom na avtomatskem vrednotenju sistemov. Nato predlagamo, implementiramo in dokumentiramo splošno, razširljivo in od sistemske arhitekture neodvisno ogrodje za evalvacijo sistemov, skupaj s prvo spletno platformo za avtomatsko vrednotenje NLP nalog v slovenščini z javnimi lestvicami, ki prikazujejo rezultate objavljenih sistemov.

• Keywords: procesiranje naravnega jezika, vrednotenje, lestvica najboljših, strojno učenje, spletna platforma