A Scalable Solution for Extreme Multi-class Product Classification: An E-commerce Case Study

Simple item page
dc.access.optionRestricted Campus Access Only
dc.contributor.advisorGudivada, Venkat N
dc.contributor.authorFathi, Ehsan
dc.contributor.departmentComputer Science
dc.date.accessioned2018-05-25T18:03:04Z
dc.date.available2020-05-01T08:01:54Z
dc.date.created2018-05
dc.date.issued2018-04-27
dc.date.submittedMay 2018
dc.date.updated2018-05-23T21:15:09Z
dc.degree.departmentComputer Science
dc.degree.disciplineMS-Computer Science
dc.degree.grantorEast Carolina University
dc.degree.levelMasters
dc.degree.nameM.S.
dc.description.abstractImage classification is the main task in image processing. Although, there were a lot of advances in recent years, it is still quite a challenge. On the other hand, due to the progress in technology, e-commerce has emerged as the fastest-growing sector of the U.S. marketplace. Product classification is an extremely important issue in e-commerce. In this work, we propose a scalable, flexible, practical, modular and efficient architecture to use image classification techniques for product classification just using product images. Considering the diversity of products offering in retail online retail stores it is not surprising that we confront an excessive number of classes. Case study is Cdiscount which is the biggest non-food e-commerce company in France which has made about 3 billion euros. As the trend of growing rate of this e-commerce shows they will have about 30 million products up for sale while they just had 10 million products until 2 years ago. As the next step to toward business expansion, they decided to employ image processing techniques. The structure of the dataset, diversity of the products and volume of it makes it unique between all the available public data sets. We focused on developing a CNN architecture to tackle this challenge and provide a more general, flexible, scalable and efficient solution for Cdiscount image classification business problem. Results of applying the proposed architecture shows a reasonable accuracy which shows the efficiency of the architecture. A comparison between proposed model and previous models is also provided.
dc.embargo.lift2020-05-01
dc.format.mimetypeapplication/pdf
dc.identifier.urihttp://hdl.handle.net/10342/6769
dc.language.isoen
dc.publisherEast Carolina University
dc.subjectproduct classification
dc.subjectConstitutional Neural Network
dc.subject.lcshElectronic commerce--Case studies
dc.subject.lcshImage compression
dc.subject.lcshInternet industry
dc.titleA Scalable Solution for Extreme Multi-class Product Classification: An E-commerce Case Study
dc.typeMaster's Thesis
dc.type.materialtext

Files

Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
FATHI-MASTERSTHESIS-2018.pdf
Size:
5.9 MB
Format:
Adobe Portable Document Format
Download

Collections

Master's Theses
Computer Science