|
Title:
|
A PRELIMINARY EXPLORATION OF DEEP LEARNING TO IDENTIFY UNQUALIFIED USERS AND LOW QUALITY NUTRITION ADVICE ON TWITTER |
|
Author(s):
|
Guto Leoni, Patricia Endo, Pierangelo Rosati and Theo Lynn |
|
ISBN:
|
978-989-8704-26-9 |
|
Editors:
|
Piet Kommers and Pedro Isaías |
|
Year:
|
2021 |
|
Edition:
|
Single |
|
Keywords:
|
Deep Learning, Twitter, Nutrition, Diet, eHealth, Public Health Communication |
|
Type:
|
Full |
|
First Page:
|
45 |
|
Last Page:
|
52 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
The promotion of healthy nutrition and diet advice is a key tenet of public health strategies to combat obesity worldwide.
Increasingly, social media is a major source of health information however is not subject to the same filtering and quality
control as required by public health bodies or commercial sources. Consequently, unqualified users, poor quality nutrition
and diet information, and social bots are an increasing feature of the nutrition and diet discourse on social media. Such
activity may result in adverse impacts to the public and interfere with official public health education and promotion
campaigns. In this paper, we propose a deep learning model based on a Long Short-Term Memory (LSTM) architecture to
detect (i) unqualified Twitter accounts, and (ii) tweets featuring low quality nutritional advice. We test the proposed model
on dataset comprised of tweet data and profiles associated with over 1.2m tweets on nutrition and diet over a 16-month
period from January 2018 to April 2019. Preliminary results suggest that the LSTM performs significantly better for profiles
than tweets, averaging between 87% and 94% for accuracy, sensitivity, precision and F1-Score, and achieving 47% for
specificity. These values are much lower for the tweets classifier. |
|
|
|
|
|
|
