Združeni model za ekstrakciju relacija između imenovanih entiteta

master's thesis

• Title (english): A Joint Model for Named Entity Relation Extraction
• Author: Dino Radaković
• Mentor: Jan Šnajder (mentor)
• Committee member: Jan Šnajder (predsjednik povjerenstva)
• Committee member: Siniša Šegvić (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2016-07-04, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Pojavljivanja imenovanih entiteta (npr., ljudi, organizacija i lokacija) u tekstu često tvore odnose ili relacije, kao što su to prijateljstva, poznanstva, utjecaj, suradnja i mnoge druge. Ekstrakcija relacija među imenovanim entitetima važan je zadatak, posebice s obzirom na trenutno dostupne količine teksta pisanog prirodnim jezikom. Najčešće izvedbe temelje se na algoritmima nadziranog učenja, koji zasebno klasificiraju svaki od parova imenovanih entiteta s ciljem zaključivanja relacije... More među njima. Takvi pristupi ne uzimaju u obzir strukturiranu prirodu relacija u obliku međurelacijskih odnosa i ograničenja prilikom učenja modela, što ponekad djelomično nadoknađuju upotpunjavanjem predikcija globalnom optimizacijom. Aktualna istraživanja u području strukturne predikcije temeljene na pretraživanju pružaju nove mogućnosti za rješavanje problema ekstrakcije relacija. Ovaj rad usmjeren je na jedan algoritam strukturne predikcije temeljene na pretraživanju, SEARN, prilagođen za zadatak ekstrakcije relacija među imenovanim entitetima. Uvodi se novi skup podataka za ekstrakciju relacija na hrvatskom jeziku. Opisuje se model temeljen na pravilima, model temeljen na strojevima potpornih vektora te model temeljen na SEARN-u. Kvaliteta spomenutih modela ocijenjena je na uvedenom skupu podataka, korištenjem F1 mjere. Less

Abstract (english)

Occurrences of named entities (i. e. people, organizations, locations, etc.) often
appear within text in form of relations, such as acquaintance, influence, collaboration,
and various others. Named entity relation extraction is a task of great importance, espe-
cially considering the sheer scale of presently availabe text written in natural language.
Typical implementations are based on supervised learning algorithms, classifying each
pair of named entities individually, in order to... More determine the relation between the two.
However, such approaches do not take into account the structural nature of relations in
the form of intra-relational influences and constraints at training time, sometimes com-
pensating for the resulting drawbacks by post-processing outputs via global loss opti-
mization. Recent research in search-based structured prediction opens new venues for
attempting to solve relation extraction. This thesis focuses of a search-based structured
prediction algorithm, SEARN, adapted to the task of named entity relation extraction
in Croatian language. We introduce a novel dataset for relation extraction in Croatian.
We then proceed by describing and implementing a rule-based predictor, a configura-
tion of support vector machines, and a SEARN-based model for named entity relation
extraction. The models are then compared on the produced dataset, their performance
being evaluated in terms of F1 scores and related metrics Less

• Keywords: Združeno učenje; SEARN; ekstrakcija relacija; imenovani entitet; hrvatski jezik; obrada prirodnog jezika
• Keywords (english): Joint learning; SEARN; relation extraction; named entity; Croatian language; natural language processing
• Language: english
• URN:NBN: urn:nbn:hr:168:627481
• 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: 2018-02-06 11:29:19