JMIR Publications

ABSTRACT

Background:

Dietary management is vital for maintaining human health. Social media platforms are widely used for daily recording and communication for individual’s diets, activities, and emotions. These social media texts and images are valuable resources for dietary management.

Objective:

We developed, iFood, an applet providing personal dietary management based on social media data. The seamless connection between iFood and social media enabled the personal diet record and manage efficiently.

Methods:

We developed iFood with 7 key components: the social media data collector, the diet information collector, the diet image identifier, the text extractor, the visual-textual data integrator, the nutrient calculator, and the dietary reports generator. The social media data collector and the diet information collector were used to collect diet texts, images, user information, and diet-related information to construct User Visual-Textual Dataset, Dietary Information Expansion Dataset, and Diet Recipe Dataset. To extract diet entity from diet text, multiple named entity identification methods were applied as the text extractor. Images of diets were recognized using the diet image identifier. The diet identification findings under the unimodality data were fused using the visual-textual data integrator. Nutrient calculator was used to calculate the nutrients intake of user using data from the Diet Recipe Dataset and the Chinese Food Composition (including energy, carbohydrate, protein, and fat). Additionally, we developed a WeChat applet as the dietary reports generator to analyze and display outcomes of diet management across several dimensions, such as food records, diet preferences, diet behaviors, nutrients content, etc.

Results:

We collected social media data from the Weibo platform, which included 105,227 dietary textual records and 407,133 diet images from 96 Weibo users. 1,286 diets images and information were obtained. Applying machine learning algorithms, we integrated the identification results of diet texts and images, the accuracy of the visual-textual integrator after being trained on the aforementioned dataset is 51.38%. Furthermore, we displayed the nutrient intake, distribution of diet categories, and daily, weekly, monthly, and yearly diet records via iFood applet.

Conclusions:

We developed a social media based dietary management applet, iFood. It was demonstrated to be useful for personal dietary record and management with efficient connection with individual social media data and accurate analysis of individual’s dietary behavior. This study may prompt utilization of the social media data for achieving a healthy lifestyle.