Recognizing product emotions using deep learning for subtle expression recognition

Date of Publication

2017

Document Type

Master's Thesis

Degree Name

Master of Science in Computer Science

College

College of Computer Studies

Department/Unit

Computer Science

Thesis Adviser

Merlin Teodosia C. Suarez

Defense Panel Chair

Macario O. Cordel, II

Defense Panel Member

Joel P. Ilao
Rafael A. Cabredo
Merlin Teodosia C. Suarez

Abstract/Summary

A variety of products are designed and manufactured depending on the demands of the consumers. To further develop better products to the increasing wants and needs of the society, insights regarding these products were collected and analyzed. One aspect of these consumer insights is commonly known as product emotion. It is the study of customer's behavior and appreciation when interacting with a product.

One notable system had explored product emotion recognition via facial expression recognition. However, a traditional machine learning approach was used which utilized distances of facial points as features. Throughout his research, it has been observed that product emotions are difficult to recognize. These expressions show low intensity of emotions called, subtle expressions.

Deep learning, on the other hand, had proven to be a powerful approach in image classification because it learns multiple representations of patterns which make it different from using a specific hand-crafted feature: distances of facial points. For this reason, a best performing convolutional neural network was developed which is a product of extensive experiments to recognize subtle expressions from product emotions. Best results came from a 10 layer deep network. The final results are 70.69% top-3 accuracy, 60.08% top-1 accuracy, and 82.46% top-1 accuracy using 7, 3, and 2 emotion labels respectively. Moreover, this research compared the performance of the model with humans. The model performed better than humans which achieved 75% against 62.5%.

Abstract Format

html

Language

English

Format

Electronic

Accession Number

CDTG007054

Shelf Location

Archives, The Learning Commons, 12F Henry Sy Sr. Hall

Physical Description

1 computer disc ; 4 3/4 in.

Keywords

Consumers' preferences; Consumer behavior

This document is currently not available here.

Share

COinS