A Combinatorial Tweet Clustering Methodology Utilizing Inter and Intra Cosine Similarity

Data mining techniques are well known and are often used to analyze and explore datasets for meaningful information. Social media, such as Twitter, has emerged as a source of data where millions of tweets are generated everyday. They include tweets from individuals who share thoughts, commentary and their feelings about a wide variety of subjects. Social media also attracts marketers and businesses for the purpose of advertising, brand imaging and getting feedback from users. Twitter’s significant popularity and mass usage has resulted in a very large dataset where virtually any subject that is queried from the Twitter API may return a vast number of tweets. As a result, these tweets can be related to several distinctly different categories. Data mining a large amount of tweets to classify them into meaningful categories is a challenging task because of the often informal language used, the inclusion of URL links, spam and other irrelevant information. This thesis presents a combinatorial hierarchical clustering methodology that categorizes tweets into meaningful clusters by utilizing inter and intra cluster cosine similarity. Cosine similarity is the degree of relativity between two vectors. This thesis proposes a “Combinatorial Hierarchical Clustering Methodology” as a combination of both agglomerative (Bottom-Up) and divisive (Top-Down) hierarchical clustering approaches that attempts to maximize the clustering effectiveness and quality. The proposed methodology sub-categorizes, divides and combines clusters through an iterative process to help make sorted categories more meaningful. In addition, this approach does not require a-priori information about the numbers of clusters to be formed but rather forms clusters dynamically based on their determined similarity.

Name thesis:degree_name

Master of Applied Science (MASc)

Level thesis:degree_level

Master's

Discipline thesis:degree_discipline

Engineering - Software Systems

Grantor dc:publisher

Faculty of Graduate Studies and Research, University of Regina

Year dc:date.issued

2015

Author dc:creator

• Kaur, Navneet

Advisor dc:contributor.advisor

• Gelowitz, Craig

Committee members dc:contributor.committeemember

• Benedicenti, Luigi
• El-Darieby, Mohamed

Language dc:language.iso

en

DOI

https://doi.org/10.82465/4493

OAI identifier oai:identifier

oai:uregina.scholaris.ca:10294/6549