Living in a Branded Look-A-Like World: Measuring and predicting brand preferences in social networks
(A2019-10040)

AUTHORS

Willem Smit, Asia School of Business in collaboration with MIT Sloan; Daphne Leent, Deloitte

KEYWORDS

social networks; prediction algorithms; brand preferences

ABSTRACT

As a consequence of the increase in available data, marketers move their focus to more data-based strategies. When still data about individual consumers is missing, marketers resort to look-a-like audiences. In this paper, we determine how analysis on social networks could benefit identifying brand like-minded audiences. Using surveys we create two sets containing data regarding both social relations and brand preferences. The first survey of 401 respondents gauges the ego-centered network and applies tests to show whether related people are more likely to appreciate the same brand than unrelated people. The second dataset with 254 observations measures people’s social network to construct several models predicting whether a person would choose a brand. The considered prediction methods are (1) nearest neighbour, (2) Markov random fields and (3) kernel based regression. The preciseness of the predictions highly depend on the sample, but roughly 60% of the predictions are correct. In this paper we will focus on the nearest neighbour method and its outcomes.