Detecting trends in product consumer preference using sentiment visualization


Tommy Chua

Date of Publication


Document Type

Master's Thesis

Degree Name

Master of Science in Computer Science


College of Computer Studies


Computer Science

Thesis Adviser

Charibeth Cheng

Defense Panel Chair

Ralph Vincent Regalado

Defense Panel Member

Leif Romeritch Syliongka
Charibeth Cheng


As more and more consumers use the web space to voice out and share their opinions, the Internet as a whole continues to grow as a valuable asset especially for companies to support business solutions. This is because consumer opinions are very indicative of consumer preference which aid marketing companies in becoming more consumer-centric. Sentiment Visualization offers its application when dealing with user-generated data to collectively product a general view of consumer preference among the abundant consumer opinions currently available in the web space. Several efforts have delved into graphically representing sentiments across a wide number of texts for users to effectively draw pertinent information on. This research designed a framework that detects trends in product consumer preference through sentiment visualization. To identify consumer preference, consumer reviews were processed to extract product attributes along with their corresponding descriptions through term frequency inverse class frequency and dependency parsing. These information were then aggregated to sentiment visualizations which present and highlight the trend in product consumer preference.

The main output of the research was evaluated by comparing it to the actual trends that are seen in the mobile market. Two different evaluations were conducted to better assess the produced framework. The first evaluation method focused on looking at the frameworks capability to detect top attribute descriptions for each year. The research output was able to correlate with the the market movement when only one match was surveyed. This was able to garner an average of 89%. The second evaluation method looked at comparing the performance or progression of each attribute description through peak years. The result for this evaluation was satisfactory as a 70% correlation was attained. Although market research was only assumed, the framework produced by the research was able to present information that mostly correlated with the actual production trends in the market.

Abstract Format




Accession Number


Shelf Location

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

Physical Description

computer disc ; 4 3/4 in.

This document is currently not available here.