Discipline: Mathematics and Statistics
Subcategory: Mathematics and Statistics
Amos Anzele - North Carolina A&T State University
Boasting over 100 million daily active users, and more than 500 million daily tweets sent per day, Twitter is one of the top 5 most used social media platforms. With such a large amount of data, it is a potent source for marketing analysis with text mining. This study focused on gauging an effective marketing strategy for a company’s product line based on Twitter users’ perception. The objective of this study was to characterize customers’ perception of Tesla, and identify competing brands using twitter messages. We utilized Twitter’s Application Programming Interface (API) along with the R ‘twitteR’ package to extract twitter message data in June 2017, before the launch of Model 3. After extensive text mining-related data cleanup, we created 22,500 corpora upon which we performed sentiment analysis. Our analysis relied on the sentiment lexicon defined by Hu and Liu (2004). Our newly proposed sentiment score based on adjusted weights on positive and negative words showed that tweets containing Tesla and related words are neutral (77%), negative (12%) and positive (11%). Our visualization tools, the word cloud and network analysis graph, identified two potential competing brands of Tesla, BMW and Apple. These results suggest that Tesla could be viewed as both an automotive company and a technology company by Twitter users, and a potential marketing strategy may target Apple and BMW customers rather than other electric car customers. Our future goal is to investigate the effect of the target marketing of a brand.
Funder Acknowledgement(s): This research was funded by the National Science Foundation (NSF) under Grant HRD#1623358.
Faculty Advisor: Dr. Seongtae Kim, firstname.lastname@example.org
Role: I created a Twitter developer account, which allowed me to have access to public tweets via Twitter's API. Utilizing R's TwitteR package, I was able to download the data in the form of an excel file, and load in R to perform the text mining. I performed the data cleaning that allowed for a 22,500 corpora upon which the text mining was executed. I chose the visualization that best translate the data to the audience. The word cloud and the network analysis were my two choices. Also, I decided to introduce an alternative sentiment analysis score based on an adjusted weight of positive words. By doubling the weigh of the positive words found in the sentiment lexicon, I felt it would give a more equitable representation of Twitter users' true sentiment.