Analiza sentimenta u recenzijama proizvoda

undergraduate thesis

• Title (english): Product Review Sentiment Analysis
• Author: Niko Palić
• Mentor: Jan Šnajder (mentor)
• Committee member: Jan Šnajder (predsjednik povjerenstva)
• Committee member: Domagoj Alagić (član povjerenstva)
• Committee member: Zoran Medić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog rada bio je izgraditi model strojnog učenja koji može što preciznije moguće
predvidjeti broj zvjezdica (rejting) neke recenzije proizvoda, koriste ́ci tekst recenzije. Recenzije su dane u Amazonovom skupu podataka. Svaka recenzija ima tijelo, naslov i broj
zvjezdica. Pripremanje skupa podataka uključivalo je tokenizaciju, micanje beskorisnih riječi i reduciranje riječi u korijen. Korišteni modeli su Gaussov naivni Bayesov klasifikator i sekvencijalna neuronska mreža. Trenirani su na različitim kombinacijama značajki (Word2Vec, mjera polarnosti, mjera subjektivnosti, brojač posebnih znakova i brojač riječi napisanih velikim slovima). Iz rezultata testiranja modela na prije neviđenim podacima, može se zaključiti da neuronske mreže djeluju bolje na ovakvom problemu. Kombiniranje svih značajki dalo je otprilike jednaku preciznost kao i korištenje samog Word2Vec-a. Najbolja mjera preciznosti i R2 mjera bile su umjerene, ali daleko veće od mjera osnovnog modela.

Abstract (english)

The task for this thesis was to build machine learning models to, as accurately as possible, predict star ratings of product reviews, by using only text of those reviews. The reviews are supplied in Amazon's product review dataset. Each review consisted of a title, body and accompanying star rating.

Preprocessing of the dataset included tokenization, removing of stopwords and stemming. The models used were the gaussian naive Bayes classifier and the sequential neural network. They were trained of different combination of features (Word2Vec, polarity score, subjectivity score, special sign counter, and uppercased words counter). From the results of testing the models on unseen data, it can be concluded that neural networks work better for this task. Combining all the features gave approximately the same accuracy as only using Word2Vec. The best classification accuracy and R2 score achieved were modest, but greater than those of the baseline model.

• Keywords: analiza sentimenta; strojno učenje; recenzije proizvoda; neuronske mreže; Amazon
• Keywords (english): sentiment analysis; machine learning; product review; naive Bayes; neural networks; Amazon
• Language: english
• URN:NBN: urn:nbn:hr:168:281357
• 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: 2020-01-07 20:54:36