Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Improving product categorization by combining image and title
KTH, School of Computer Science and Communication (CSC).
2017 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Förbättrad produktigenkänning genom combination av bild och titel (Swedish)
Abstract [en]

Companies strive to automate more of the work that earlier was done manually. Among applications that could be automated is product categorization. In this study it was investigated if a combination of image and text, could perform better than individual image or text classifier for product categorization. This was examined with focus on the popular method, deep learning.

To test which classifier that had the best performance, a case study was conducted. This case study used two data sets, one with 5 classes that had an even distribution of data between classes (balanced) and one with 33 classes that had an uneven distribution of data between classes (unbalanced). The results showed that the accuracy of the combined classifier were a few percent better than the accuracy of the best performing classifier that only used one information source (image or text) for both data sets. The accuracy of the combined classifier, for the balanced data set and the unbalanced data set, were respectively 2.6% and 4.2% higher than the accuracy of the best performing single source classifier.

Abstract [sv]

Företag strävar efter att automatisera arbete som tidigare gjorts manuellt. Bland tillämpningar som kan automatiseras är produktkategorisering. I den här studien undersöktes ifall en kombination av bild och text kan prestera bättre än en individuell bild- eller textklassificerare. Detta undersöktes med fokus på den populära metoden djupinlärning.

För att testa vilken klassificerare som kunde prestera bäst genomfördes en fallstudie. Fallstudien använde två data mängder, en bestående av fen klasser med jämn fördelning av data mellan klasserna (balanserad) och en bestående av 33 klasser med ojämn fördelning av data mellan klasserna (obalanserad). Resultatet visade att den kombinerade klassificeraren har några procent högre andel korrekt klassificerade produkter än den klassificeraren som presterade bäst av de som bara använde en källa av information (bild eller text). Andelen korrekt klassificerade produkter av den kombinerade klassificeraren, för den balanserade datamängden och för den obalanserade datamängden, var 2.6% respektive 4.2% högre än andelen korrekt klassificerade produkter av klassificeraren som presterade bäst av de som bara använder en källa av information.

Place, publisher, year, edition, pages
2017.
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-240071OAI: oai:DiVA.org:kth-240071DiVA, id: diva2:1269414
Supervisors
Examiners
Available from: 2018-12-11 Created: 2018-12-10 Last updated: 2018-12-11Bibliographically approved

Open Access in DiVA

No full text in DiVA

By organisation
School of Computer Science and Communication (CSC)
Computer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric score

urn-nbn
Total: 34 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf