Primjena modela dubokog učenja za analizu sentimenta

undergraduate thesis

• Title (english): Application of Deep Learning for Sentiment Analysis
• Author: Bruno Gavranović
• Mentor: Jan Šnajder (mentor)
• Committee member: Jan Šnajder (predsjednik povjerenstva)
• Committee member: Domagoj Alagić (član povjerenstva)
• Committee member: Martin Tutek (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2016-07-11, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Uzevši u obzir eksponencijalni rast dostupnih podataka generiranih u cijelom svijetu, postoji rastući interes za stvaranje modela sposobnih za analizu podataka koji imaju semantički nepoznat kontekst. Ovaj rad se fokusira na istraživanje i analizu teorijske osnove koja stoji iza modela zvanih povratne neuronske mreže (engl. recurrent neural networks, RNN) kroz konkretnu implementaciju na problemu semantičke analize. Poseban slučaj povratnih neuronskih mreža,mreže s dugotrajnim kratkoročnim pamćenjem(engl. Long short-term memory networks, LSTM) su implementirane na problemu klasifikacije sentimenata na skupu podataka recenzija filmova. Testirane su i uspoređene mnogobrojne arhitekture i konfiguracije hiperparametara spomenutih neuronskih mreža. Značajano poboljšanje performansi je primjećeno s dubokim višeslojnim LSTM mrežama, u za razliku od rezultata dobivenih s plitkim mrežama.

Abstract (english)

Given the exponential growth of amount of available data generated worldwide, there has been a rising interest in the creation of models capable of analyzing the data without knowing the semantic context. This paper focuses on exploration and analysis of the theoretical foundation that stands behind models called recurrent neural networks through concrete implementation on a problem of semantic analysis. Special case of recurrent neural networks,long short-term memory networks (LSTM networks) were implemented on a task of semantic classification on the IMDb dataset. Various architectures and hyperparameter configurations were tested and compared on the said neural networks. Significant performance boost was detected with deep multi-layer LSTM networks, compared to performance with shallow ones.

• Keywords: obrada prirodnog jezika; strojno učenje; neuronske mreže; duboko učenje; analiza sentimenata; LSTM; povratne neuronske mreže
• Keywords (english): natural language processing; machine learning; neural networks; deep learning; sentiment analysis; LSTM; recurrent neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:737684
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 08:52:08