Prepoznavanje imenovanih entiteta adresa

master's thesis

• Title (english): Named Entity Recognition for Addresses
• Author: Helena Čeović
• Mentor: Goran Delač (mentor)
• Committee member: Goran Delač (predsjednik povjerenstva)
• Committee member: Marin Šilić (član povjerenstva)
• Committee member: Klemo Vladimir (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog diplomskog rada je razvoj modela visoke točnosti za prepoznavanje imenovanih entiteta adresa. Izazovi uključuju raznolikost, dvosmislenost i kompleksnost entiteta adrese. Različite arhitekture modela korištene za treniranje klasifikatora su jednostavniji modeli logističke regresije i algoritam slučajnih šuma te složeniji model temeljen na dvosmjernoj povratnoj neuronskoj mreži s dodatnim slojem koji sadrži uvjetovana slučajna polja (BiLSTM-CRF) i implementiran je pomoću Flair knjižnice.
Sva testiranja provedena su na dva skupa podataka koji se razlikuju po načinu označavanja temeljenom na granularnosti imenavanog entiteta: adresa kao cjelina i adresa
sastavljena od podelemenata. Za oba skupa podataka najbolje rezultate postiže model
temeljen na BiLSTM-CRF arhitekturi s jednim RNN slojem te treniran ili samo na
BERT-u ili na kombinaciji BERT i GloVe embedding-a.

Abstract (english)

This thesis focuses on developing a high-performing named entity recognition
model for addresses. Challenges including diversity, ambiguity and complexity of
the address entity are introduced as well as the different model architectures used for
training the classifier. These include the simpler logistic regression and random forest
models as well as the more complex bidirectional LSTM network with a conditional
random field layer (BiLSTM-CRF) implemented using Flair framework. Experiments
are conducted using variously configured models on two sets of corpora tagged differently based on address entity granularity: entire address and address consisting of
subparts. For both corpora, the best results are achieved on a BiLSTM-CRF architecture model with a single RNN layer trained on either standalone BERT embeddings or
a stacked combination of BERT and GloVe.

• Keywords: prepoznavanje imenovanih entiteta; obrada prirodnog jezika; entitet adrese; IOB označavanje; ekstrakcija značajki; logistička regresija; algoritam slučajnih šuma; Flair; GloVe; BERT; povratne neuronske mreže; LSTM; uvjetovana slučajna polja; BiLSTM-CRF arhitektura
• Keywords (english): named entity recognition; natural language processing; address entity; IOB tagging; feature extraction; logistic regression; random forest; Flair; embeddings; GloVe; BERT; recurrent neural network; LSTM; conditional random fields; BiLSTM-CRF architecture
• Language: english
• URN:NBN: urn:nbn:hr:168:736413
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 12:01:06