A. Figueira1, L. Oliveira2

1University of Porto (PORTUGAL)
2CICE / ISCAP Polytechnic of Porto & CRACS / INESC TEC (PORTUGAL)
The adoption of social media channels by Higher Education Institutions (HEI) has not been accompanied by a carefully planned and strategically design process. The adoption of social media channels has mainly been sustained by HEI joining the trend to adopt social media environments, aiming at mediatization, but lacking alignment with organizational goals and the adoption of efficient monitoring and benchmarking methods that can address the need to measure the efficiency of communication and conversation on social media.

Bearing in mind that a social media strategy needs to be aligned with and framed in the overall organizational strategic management goals, the structure of its communication strategy should provide a clear indication of the HEI’s positioning. However the right question relies on: how can one attempt to measure the efficiency of social media approach that has not been strategically designed and is the result of a set of unarticulated and situational messages? Thus, on a first stage it is not relevant, neither possible, to determine the social media strategies’ level of efficiency, which organizations constantly seek. Consequently, determining the organizational positioning of HEI current strategies will allow to combine monitoring and benchmarking methods to foster the identification opportunities and threats, which can serve as inputs for the internal evaluation of social media strategies’, for the necessary strategic readjustments and a subsequent efficiency measurement.

The model:
In order to establish a comparison of social media strategies between HEI’s, we developed a seven category model, encompassing the fundamental communication areas of focus for higher education service providers: Education; Research; Society; Identity; Administration; Relationship and Information.

The experiment:
We focused on the full set Portuguese HEI using their Facebook pages. From an initial list of the agent Page Id’s, our systems accessed and retrieved the posts of an entire school year. In the end we got 15.409 posts. Then, we performed s classification of these posts according to our model. For this step used six of the most promising, and prominent, classifiers: Support Vector Machines, Random Forests, LogiBoost, K-Nearest Neighbours, MultiLayer Perceptrons and Deep Neural Networks. First, we classified a set of 350 posts, manually. Then we run the full set of posts on our 6 trained classifiers to obtain a predicted category for each post, tagging the post with the prevalent category.
With all posts classified, each one belonging to one of the 7 categories of our model, we grouped all posts from each HEI in order to find the centroid of each group, which is a vector of the form: Centroid(HEI) = (#C1, #C2, #C3, #C4, #C5, #C6, #C7).
This vector gives as a notation that represents a signature of the communication strategy of a particular HEI. Using standard clustering algorithms, we group all the centroids into a set of k clusters, where k was given by the Silhouette coefficient.

By clustering the overall social media strategies and corresponding response rate we expect to unveil the sector’s monitoring HEI and, through a benchmarking process, retrieve useful inputs for the design of social media strategies for HEI in general. In the paper we describe in detail the classification and clustering process, as well as the conclusions we can take from these results.