Hierarchical Attention Neural Networks for the Detection of Advertising and Promotion in Texts

Overview

Novi High School Novi, MI As news consumption shifts to be based more and more online, there is also a growing trend of deceptive advertising practices in this area. Native advertising, as an example, consists of adverts written as an editorial or news article, and is, as a result, difficult to differentiate from organic content. In addition, more malicious forms of advertising have been growing in prevalence recently, with spam emails reaching an all time high in 2019. It is for this reason that a method to accurately detect and filter promotional texts is necessary. This paper proposes a novel, largescale natural language dataset built for the purpose of advertising detection, consisting of human annotated multi-class labels for 5 types of promotional language and a negative class for unbiased texts. In addition, this paper proposes a method to detect promotional texts through the use of a deep learning architecture that effectively models large texts by capturing a hierarchical feature structure, looking at both the word-level representation of a document and the overall sentencelevel representation. Evaluated using accuracy, precision, and bias as metrics, this method was found to be effective for the detection of promotion within texts.

Competition history

• JSHS 2020 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.