Twitter Twitter on the wall, which university's the fairest of them all? Exploring brands' social perception on social media using Big Data.

Background. Every University is becoming more and more like a brand. In fact, universities compete against each other to attract scholars, public or private funds, and students. For this reason, it is of growing importance to understand how universities are perceived by various stakeholders, put differently, their brand image. So far, brand image has typically been investigated through interviews, questionnaires and focus groups. Today, thanks to their increasing diffusion, the internet and social media are becoming useful information sources to understand how universities are perceived. Objectives. The aim of this work is to test a new method to understand universities’ social perception and to investigate important facets of such perception using users’ digital footprint in social networks. Such method provides quantitative metrics that allow for a rigorous investigation without losing brand image’s content and the shared representation used by users to explore the social environment. Method/Approach. One basic step for every work that aims to understand a given phenomenon is to observe it and to place it in context. To correctly frame every phenomenon a timely assessment phase is necessary. Such need brought us to focus on the comparison between our target (the universities) and other accounts of the social environment. Our goal is to measure some dimensions, in order to map the positioning of our target along them. Such dimensions are defined a priori, based on the nature of the different universities. Data have been obtained using a validated methodology (Culotta & Cutler, 2016; Permalink: http://dx.doi.org/10.1287/mksc.2015.0968). We used a modified version of the technique and we applied it to the universities under investigation. A difference with Culotta and Cutler’s work is the nature of our targets. Indeed, Culotta and Cutler investigated how various consumer brands (cars, apparels, foods and beverages, personal care), while universities are non-profit organizations. Universities are perceived along different attribute dimensions (e.g. research, employability, teaching, ecc…) in comparison with profit organizations. Therefore we are extending Culotta & Cutler’s work to the investigation of social perceptions of a very different type of targets. Following Culotta and Cutler, we employed online behaviors to track social perception dimensions. We focalized on a large group of prototypical accounts, used as benchmarks of the attribute dimension, and compared them to targets accounts. The online behavior that we considered was the act of following a prototypical (i.e., attribute related) or target account (i.e., Universities). We relayed on available online data in contrast with data generated by means of direct question such as those collected using classical brand image methods. The main advantage of this kind of data is its high ecological validity. In fact, they are free behaviors and not answers to direct questions. The spontaneous nature of online behaviors makes them very valuable and interesting. Moreover, online behaviors allow the researcher to understand how a given target is represented in the subjects’ mind because the behavior is performed without external influence such as in the case of responding a questionnaire. The proposed method is based on three stage. In the first stage, attribute dimensions are defined according to the target under investigation. In the second stage, target and prototypical accounts are selected. Target accounts are the ones of interest while prototypical accounts are accounts that show characteristics is desirable to have in order to have a high score on the dimension. The selection of prototypical accounts will determine a dimension’s benchmark, therefore, it allows for the construction of nuanced or ad-hoc dimensions. In our work, the selection of prototypical accounts was negotiated among the authors. In the third stage, the similarity among target and prototypical accounts of each dimension is computed. During this phase, a score for each target along each dimension is produced. Such score is a purely quantitative variable drawn upon the overlap between the target and the prototypes’ followers. The degree of overlap is based on the computation of a Jaccard Similarity Coefficient that summarizes the ratio of the intersection over the union of the followers’ sets. The Jaccard Similarity coefficients between the target and each prototype are then weighted to produce a unique social perception score along the given dimension. The method is based on two assumptions. First, to follow prototypical accounts of a given dimension implies that the follower is interested in such dimension. It means that accounts followed by the same user share some feature of the dimension. Second, if an account is followed by many users interested in a specific dimension the account followed is perceived as related to the dimension. Results/Findings. The intersection between followers of a target account and sets of followers of prototypical accounts is meant to capture and quantify the interest in the dimension of the follower of the target. Such intersections can be aggregated in order to obtain a score for each target along each dimension, therefore, the score can represent how targets are perceived along the dimensions of interest. Conclusion and Implications. Once a profile for each target on different dimension had been obtained, it is possible to map the environment where targets are perceived. Moreover, our method allows for an estimation of the relative importance of each follower. Such estimation provides useful insights for understanding which users drive a target’s social perception and which are to acquire in order to improve the target’s perception on a specific dimension.