Investigating banks’ social media content and consumer reactions with machine learning

Abstract

Purpose

The research examines the Twitter (X) accounts of 20 retail banks in the USA, selected from the Top 100 banks on Twitter, over the period of 2012–2022, focusing on consumer reactions to posts. By analysing 136,150 tweets, the study aims to explore how banks utilize Twitter, and examine the effects of the visual and textual elements used in their posts on consumers.

Design/methodology/approach

– Text and image analysis of the Twitter accounts available to the public was performed using different natural language processing methods (topic analysis, sentiment analysis), computer vision techniques (object detection, optical character recognition, image captioning), and statistical and machine learning models (linear and logistic regression, Catboost). The research investigates how certain brand identity elements (e.g. words, colours, visual elements, symbols) in retail banks' Twitter posts trigger consumer reactions (likes).

Findings

– We found that although some factors statistically significantly increase the popularity of a post, there is no well-identified, general brand identity element that has a commercially relevant impact on consumer responses, i.e., the popularity of a post.

Originality/value

The study sheds new light on the effectiveness of social media tactical tools and strategies in the banking sector, which are common knowledge in the general marketing practice, focusing on popular posting activities.