Mining Celebrity Endorsement Perceptions Using Varieties of Twitter - PowerPoint PPT Presentation

Mining Celebrity Endorsement Perceptions Using Varieties of Twitter Account Automated Data Maria Oikonomidou GSC 2019 Computer Science Department - University of Crete

Celebrity Endorsement : Marketing strategy whose purpose is to use one or multiple celebrities to advertise a specific product or service. Computer Science Marketing Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Coffee Company Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Keywords Big Data Graph Analysis Social Media Content Analysis Similarity Analysis Marketing Strategy Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

The idea We measure directly the strength of association between a brand and a celebrity. Our approach represents a low-cost, real-time alternative to traditional survey-bassed elicitation methods. We measure consumers view of fit between pairs of celebrities and brands and validate through survey. Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Dataset twAler * Crawled data between 2016 - 2018 700 million tweets 52 million user accounts Four parts – different data selection Follow metric: Graph - 1 million Content metric: 11G tweets directed follow relations List metric: Weighted Graph - 89 million Favorite metric: Weighted Graph - relations 206 thousand relations Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Similarity Metrics

Follow Similarity Metric Structural similarity, shape of the Twitter Social graph A set of users follows a brand b, another set of users follows a celebrity c High intersection = high similarity Jaccard index Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

List Similarity Metric Twitter users curate Similarity between lists of other users​ users according to number of lists they are placed together High number of common lists = high similarity Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Content Similarity Metric Most active type of participation Content Comparison between followers of the targeted users Vectorize via TF - IDF Cosine Similarity Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Favorite Similarity Metric "Like" Smallest possible effort Big set of common users favored a pair of targeted users = high similarity Weighted Jaccard Index Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Evaluation

Targeted users Survey Design Evaluation Pearson 100 celebrities correlation 100 brands coefficients (r) 8 sectors Survey Similarity Association scale results metric Automobiles,Technology, 1 to 5 Retail etc results by consumers Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Findings Strong onger Correla lations ons Follow metric List Metric Industrial good Technology Favourite metric Content Metric Automobiles & parts Financial Services Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Questions? Maria Oikonomidou GSA 2019 mareco@ics.forth.gr

Thank you! Maria Oikonomidou GSA 2019 mareco@ics.forth.gr