Supporting digital content marketing and messaging through topic modelling and decision trees

Andreas Gregoriades, Maria Pampaka, Herodotos Herodotou, Evripides Christodoulou

Research output: Contribution to journalArticlepeer-review

Abstract

This paper presents a machine learning approach involving tourists’ electronic word of mouth (eWOM) to support destination marketing campaigns. This approach enhances optimisation of a critical aspect of marketing campaigns, that is, the communication of the right content to the right consumers. The proposed method further considers aggregate cultural and economic-related information of the tourists’ country of origin with topic modelling and Decision Tree (DT) models. Each DT addresses different dimensions of culture and purchasing power and the way these dimensions are associated with the topics discussed in eWOM, thus revealing patterns relating tourists’ experiences with potential explanations for their dissatisfaction/satisfaction. The method is implemented in a case study in the context of tourism in Cyprus focusing on two hotel groups (2/3 and 4/5 stars) to account for their differences. Patterns emerged from the extraction of rules from DTs illuminate combinations of variables associated with tourist experience (negative or positive) for each of the two hotel categories and verify the asymmetric relationship between service performance and satisfaction. The approach can be used by management during marketing campaigns to design messages to better address the desires and needs of tourists from different cultural and economic backgrounds, as these emerge from the data analysis.

Original languageEnglish
Article number115546
JournalExpert Systems with Applications
Volume184
Early online date6 Jul 2021
DOIs
Publication statusPublished - 1 Dec 2021

Keywords

  • Cultural and economic distance
  • Decision trees
  • Shapley additive explanation
  • Topic modelling
  • Tourists’ reviews

Fingerprint

Dive into the research topics of 'Supporting digital content marketing and messaging through topic modelling and decision trees'. Together they form a unique fingerprint.

Cite this