Estimados Académicos y Académicas FEN,

Les extendemos la invitación al seminario académico organizado por el Departamento de Administración, en el que se presentará el trabajo titulado Predicting fan engagement and social media performance metrics from unstructured data content using deep learning techniques.

Exponen: David Díaz, académico FEN y Mohamed Zaki, Deputy Director of the Cambridge Service Alliance at the University of Cambridge

Abstract: This study presents a research approach for predicting social media performance metrics and fan reactions and emotions of posts published by Manchester United Football Club (MUFC) in their brand Instagram account. We propose an AI approach (deep learning architecture) that processes and extracts features from content published by the MUFC and use those features to predict the level of fan engagement as captured by Instagram performance metrics: number of likes, comments and shares. In addition, the given features were also utilized to predict the level of fans’ emotions, such as, surprise, joy and sadness, and positive, negative and neutral sentiments that each post will elicit. Our sample included all post generated by MUFC and samples of their fans replies between January 2014 and December 2020. Our AI approach, make use of state-of-the art Natural Language Processing (NLP) and Computer Vision (CV) algorithms to extract features from the multimedia content of each post, that is, the main elements that an Instagram carrousel typically contains: images, videos, captions and hashtags. In particular, we use transformers NLP algorithms to extract topic, sentiment and emotion recognition features from textual captions and comments, and transfer learning CV algorithms to extract celebrities, objects, faces, and emotions features from images and videos. Multi-input Deep Learning Algorithms were also used to generate automatic caption descriptions of images and videos, and to combine NLP and CV features to predict post success and sentiment and emotions elicited. Our models can achieve high levels of accuracy to predict which posts will get high/low engagement rates with fans, as well as, to predict the level and type of sentiments and emotions that will be mentioned by fans in the post replies.

Link de inscricpión al evento.

Saludos cordiales,

Dirección de Investigación