Start Date
16-8-2018 12:00 AM
Description
Overweight is epidemic in the United States and elsewhere in the world, causing major health concerns. Based on self-disclosure theory, i.e., people have the tendency to disclose information concerning their feelings, intentions, and acts (e.g., food consumption) online, we aim to leverage social media platforms to develop an unobtrusive approach to predicting overweight. However, traditional statistical and machine learning-based approaches either deliver unsatisfactory performance or demand a large number of features. In this paper, we present a novel social media-based overweight prediction approach based on deep learning as applied in the context of Natural Language Processing (NLP). The input to this approach is food-related Twitter posts. Our computational results show the effectiveness of our method, with remarkable improvement in terms of accuracy over a set of benchmark methods.
Recommended Citation
Huangfu, Luwen and Zeng, Daniel, "Social Media-based Overweight Prediction Using Deep Learning" (2018). AMCIS 2018 Proceedings. 16.
https://aisel.aisnet.org/amcis2018/DataScience/Presentations/16
Social Media-based Overweight Prediction Using Deep Learning
Overweight is epidemic in the United States and elsewhere in the world, causing major health concerns. Based on self-disclosure theory, i.e., people have the tendency to disclose information concerning their feelings, intentions, and acts (e.g., food consumption) online, we aim to leverage social media platforms to develop an unobtrusive approach to predicting overweight. However, traditional statistical and machine learning-based approaches either deliver unsatisfactory performance or demand a large number of features. In this paper, we present a novel social media-based overweight prediction approach based on deep learning as applied in the context of Natural Language Processing (NLP). The input to this approach is food-related Twitter posts. Our computational results show the effectiveness of our method, with remarkable improvement in terms of accuracy over a set of benchmark methods.