Prepoznavanje imenskih entitet na domenskih besedilih iz farmacije

KOVAČ KEBER, BENJAMIN (author), Žitnik, Slavko (mentor)

Prepoznavanje imenskih entitet je ena od nalog problema procesiranja naravnega jezika. Gre za označevanje besed in besednih zvez z oznakami v naprej določenih tipov imenskih entitet. Primeri uporabe prepoznavanja imenskih entitet so klasifikacija vsebine za ponudnike novic, učinkoviti iskalni algoritmi, priporočanje vsebine, organizacija člankov in podpora strankam. Preučili smo problem prepoznavanja imenskih entitet na domenskih besedilih iz farmacije. V ta namen smo uporabili štiri različne metode in za učenje modelov uporabili dva korpusa (CHEMDNER in n2c2), ki imata ročno označene imenske entitete iz področja farmacije (in kemije). Modele smo evalvirali tudi na besedilih, ki smo jih sami ročno označili. Najbolje se je odrezal model BERT. Za praktično uporabo pa bo verjetno potrebno v modele vložiti še nekaj truda za izboljšave.

• Language: Slovenian
• Keywords: procesiranje naravnega jezika, prepoznavanje imenskih entitet, farmacija
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2023
• PID: 20.500.12556/RUL-144589
• COBISS.SI-ID: 144127491
• Publication date in RUL: 02.03.2023

Secondary language

• Language: English
• Title: Named entity recognition in pharmaceutical domain texts

Abstract

Named entity recognition is one of the tasks of the natural language processing problem. It is about tagging words and phrases with labels of predefined types of named entities. Examples of named entity recognition use cases are content classification for news providers, efficient search algorithms, content recommendation, organization of research papers and customer support. We have studied the problem of named entity recognition on domain texts from pharmacy. For this purpose, we used four different named entity recognition methods using two corpora (CHEMDNER and n2c2) that contain manually annotated named entities from the pharmacy domain. We also evaluated the models on texts, which we manually annotated. The BERT model performed best. For practical use, it is probably necessary to put some more effort in the model in order to improve it.

• Keywords: natural language processing, named entity recognition, pharmacy