Machine learning approach to auto-tagging online content for content marketing efficiency: A comparative analysis between methods and content type

Abstract

As complex data becomes the norm, greater understanding of machine learning (ML) applications is needed for content marketers. Unstructured data, scattered across platforms in multiple forms, impedes performance and user experience. Automated classification offers a solution to this. We compare three state-of-the-art ML techniques for multilabel classification-Random Forest, K-Nearest Neighbor, and Neural Network-to automatically tag and classify online news articles. Neural Network performs the best, yielding an F1 Score of 70% and provides satisfactory cross-platform applicability on the same organisation's YouTube content. The developed model can automatically label 99.6% of the unlabelled website and 96.1% of the unlabelled YouTube content. Thus, we contribute to marketing literature via comparative evaluation of ML models for multilabel content classification, and cross-channel validation for a different type of content. Results suggest that organisations may optimise ML to auto-tag content across various platforms, opening avenues for aggregated analyses of content performance.

A machine learning approach to the digitalization of bank customers: Evidence from random and causal forests

“…Instead of ex-ante selecting a default machine learning technique, we initially consider a number of machine learning techniques that have proved their value in this field (see among others [ 28 – 32 ]. The following machine learning approach is employed to examine bank customer digitalization:…”

“…Methodologically, instead of ex-ante selecting a machine learning technique, we consider a number of machine learning techniques that have proved their value in this field: random forest, extreme gradient boosting, k-nearest neighbor, support vector machine, Bayesian networks and extreme learning machine (see among others [ 28 – 32 ]). After selecting the machine learning with the best performance (in terms of predicted accuracy) we use this algorithm to identify the main features predicting bank customers’ digitalization process.…”

Topic-driven toxicity: Exploring the relationship between online toxicity and news topics

“…These add no information for the classifier algorithm and are thus removed. As machine learning models take numbers as input [48], we convert our articles into numbers using the Term Frequency-Inverted Document Frequency (TF-IDF) technique that counts the number of instances each unique word appears in each content piece. TF-IDF scores each word based on how common the word is in a given content piece, and how uncommon it is across all content pieces [49].…”

“…Here, we report the key evaluation methods and results of the topic classification. Note that a full evaluation study of the applied FFNN classifier is presented in Salminen et al [48].…”

Content marketing strategy and its impact on customers under the global market conditions

“…For instance, an algorithm applied to website content is not necessarily expected to be effective across channels that vary in content-type, such as when classifying online videos whose titles and descriptions tend to be considerably scarcer than website content such as news and blog articles. [11] The presented paper consists of four parts, the first part of the article focuses on the theoretical aspects of the issue of global content marketing. Second part of paper focuses on the methods of data collection that are needed to perform the analysis and also the background for the third part of paper.…”