Meet Lucy: A Nutrition e-Assistant that Uses Crowdsourcing to Help You Lose Weight and Stay Motivated
By Lori Cameron
 

robot assistant

Staying motivated can be a pain for people struggling to lose weight. Wouldn’t it be great to be able to determine the nutritional value of your food at a glance or get encouragement when you need it?

Well, now you can.

Meet “Lucy,” a digital assistant that helps patients undergoing weight-loss treatment. Researchers from Ensenada Center for Scientific Research and Higher Education (CICESE) and ITSON teamed up to design Lucy whose database of nutritional information is created through crowdsourcing. They tested six crowdsourcing methods on 51 weight-loss participants and assessed them for how long it took to evaluate the food, how difficult the task was, and how accurate they were.

“We also describe the integration of one of these approaches into a conversational coaching agent to assist individuals who want to change their eating behaviors,” say Mario O. Parra, Jesus Favela, Luis A. Castro, and Arturo Morales, authors of “Monitoring Eating Behaviors for a Nutritionist E-Assistant Using Crowdsourcing” (login maybe be required for full text) in the March 2018 issue of Computer, our special issue on e-coaching.

Researchers designed their e-coaching model in three stages.

Informing the Design of a Nutritionist E-Assistant

Over three months, the researchers interviewed a nutritionist and 95 patients to discover what features Lucy should have. Here’s what they came up with:

  1. Allow the patient to keep a log of meals.
  2. System must oversee dietary regime compliance, without demanding additional effort from patients or the nutritionist.
  3. Provide alternatives for ingredient and dish substitutions.
  4. Estimate nutritional content, without giving the nutritionist additional burden.
  5. Provide nutritional counseling for diet compliance.
  6. Oversee progress in weight loss.
  7. Deliver motivational messages based on the patient’s goals.
  8. Answers to patients should be reliable.
  9. Answers should be provided quickly.
  10. System should run on hardware available to the patient.
  11. Patients should be able to access the system anywhere.
  12. The agent should be engaging either by voice or text.

“Informed by these requirements, we propose the design of a conversational agent that provides personalized support to patients during the intervention. The application is aimed at reducing some of the burden of counseling from the nutritionist,” the authors say.

Crowdsourced Assessment of Food Intake

Next, they compiled a list of six food intake features that they wanted crowdsource participants to evaluate based on photographs of the food:

  1. Number of calories
  2. Food groups
  3. Healthfulness scale
  4. Caloric range
  5. Ingredients
  6. Similar images

The participants were evaluated on how long it took them to assess all six features. In general, it took less than 30 seconds to evaluate all six, taking less than 7 seconds to determine healthfulness and calorie range and taking the longest to determine ingredients.

Comparison of six crowdsourced, photo-based approaches to assess the nutritional content of meals (see main text). (a) Latency per approach. (b) Cognitive load per approach across the four utilized subscales of the NASA-Task Load Index (NASA-TLX). Circles are suspected outliers; stars are outliers (with their corresponding observation number). Red lines show the overall mean across all approaches.

From this data, the researchers created a table measuring latency (how long to evaluate), cognitive load (mental effort), and accuracy.

Comparison of latency (LAT), cognitive load (CL), and accuracy (ACC) for crowdsourced approaches to assessing the nutritional content of meals.

Lucy—A Nutritionist Assistant Conversational Agent

As Lucy talks to you, the app displays a transcript of your conversation. You can ask Lucy a number of questions such as what you are having for dinner, what items can be substituted for others (tortillas instead of bread), how much you’ve eaten so far today, and how much weight you’ve lost. You can also tell Lucy the reasons you want to lose weight, and she will remind you when you need an emotional boost.

Now, it’s time to meet Lucy!

The Lucy conversational agent: (a) sample interaction with a patient, and (b) log of photographs of meals consumed in a given day.

Lucy was tested on 59 participants, and the results were primarily positive.

“The average scores indicate that the participants found Lucy to be useful and easy to use and seemed interested in adopting the system. The item with the lowest score was ‘It will be easy for me to become an expert in the use of the system.’ In contrast, the item ‘Using Lucy would help me with tasks related to my diet’ obtained the highest score. One of the participants gave a “neutral” answer to all questions, and two others had mostly negative ratings. All others seemed rather positive toward Lucy,” said the researchers.

Related research on health, diet, and fitness tech in the Computer Society Digital Library

Login may be required for full text.

 


 

About Lori Cameron

Lori Cameron is a Senior Writer for the IEEE Computer Society and currently writes regular features for Computer magazine, Computing Edge, and the Computing Now and Magazine Roundup websites. Contact her at l.cameron@computer.org. Follow her on LinkedIn.