Crowdsourcing Personalized Weight Loss Diets
Total Page:16
File Type:pdf, Size:1020Kb
RESEARCH FEATURE Crowdsourcing Personalized Weight Loss Diets Simo Hosio, University of Oulu Niels van Berkel, University College London Jonas Oppenlaender, University of Oulu Jorge Goncalves, University of Melbourne The Diet Explorer is a lightweight system that relies on aggregated human insights for assessing and recommending suitable besity is a growing health problem around the world and weight loss diets. We compared has been described as a global its performance against Google epidemic by the World Health OOrganization. Not only is being overweight and suggest that the system, a psychologically sensitive issue, but it has been shown to negatively affect health bootstrapped using a public in a number of ways, including accelerat- ing aging or increasing the risk of diabe- crowdsourcing platform, tes and heart conditions. Examining the provides comparable results in issue strictly through the academic lens, two aspects matter in losing weight: diet terms of overall satisfaction, and exercise. However, previous work has shown that the former is substantially relevance, and trustworthiness. more important.1 Yet there is no consen- sus even among scientists on what is the optimal diet for losing weight (consider, for example, the constant battle between low-carb and low- platform designed for academic studies, we assessed all fat diets). As a result, one identified problem in the diet- 21 major weight loss diets listed in Wikipedia at the time ing world is information overload.2 There are simply too of this study across six distinct evaluation criteria. Then, many diets to choose from. using the same source to hire participants, we conducted In this article, we present The Diet Explorer (TDE), a a user study to compare TDE against Google in discover- crowd-powered tool that exploits crowdsourcing and wis- ing personalized weight loss diets. We consider Google as a dom of the crowd3 in first assessing and then recommend- fair yardstick in this case since people increasingly turn to ing personalized weight loss diets. Using paid labor from online resources to find health-related information. Prolific Academic (https://prolific.co/), a crowdsourcing Our results validate that TDE can be used to quickly offer personalized weight loss diets that meet the user’s needs much in the same way as Google but faster and Digital Object Identifier 10.1109/MC.2019.2902542 Date of current version: 15 January 2020 without its clutter, advertisements, and other identified COMPUTER 0018-9162/20©2020IEEE PUBLISHED BY THE IEEE COMPUTER SOCIETY JANUARY 2020 63 RESEARCH FEATURE pitfalls. Finally, we make a technical for time-consuming help from strang- pairs, the resulting knowledge is based contribution by providing a consid- ers than friends. Finally, Hosio et al.4 on the wisdom of the crowd, where the erably improved, reimplemented ver- discussed a lightweight decision sup- crowd is the people who assessed the sion of TDE in the plug-in repository of port tool that helps collect solutions to pairs. Specifically, in the health infor- the world’s most popular online con- any problem and rates those solutions mation field, a similar approach has tent management system, WordPress. in terms of different criteria. been successfully used in the past to Installing the plug-in version of TDE is In this article, we leverage crowd- recommend and capture data on low- thus a one-click process, and webmas- sourcing to assess weight loss diets er-back pain treatments.5 ters can use it for helping their visitors across a variety of criteria, using a data An important consideration with to discover diets and donate data structure similar to the one described any embeddable tool such as ours is the for science. in Hosio et al.4 In other words, we surrounding context, in other words, modeled human-contributed infor- the website. The context always plays TDE mation on diets across a set of criteria a role in user perceptions. To keep our by crowdsourcing multiple ratings per study design tidy, we deployed TDE Background and related work each diet-criterion pair. This crowd- as a solo installment on a blank page Losing weight is challenging. Both a sourced information repository—a online. There was no context to skew proper and balanced diet and exercise snapshot of the cumulative knowl- the users’ opinions about the tool and help, but of those two, diet has been edge of the respondents—was then its functionality. Instead of a website shown to matter more.1 Yet, as evi- used to bootstrap a crowdsourced that would normally introduce the tool, dent from the endless amount of diet- system that allowed users to discover we deployed an additional short splash ing fads and advice, choosing a diet to diets that best match their personal screen before loading TDE for the users begin with is extremely challenging preferences, described by a set of opti- and set out to examine how TDE suc- and confusing. To this end, crowd- mal/desired criteria values. ceeds in recommending diets. From the sourcing has emerged as an excellent However, it is important to note end users’ perspectives, TDE consists method to aggregate knowledge that that from a nutrition science perspec- of two main conceptual stages: one for can then be used in recommending tive, we do not consider whether these assessing different weight loss diets and suitable and trustworthy options: to diets work optimally and for whom. one for discovering personalized diets basically offer decision support.4,5 Our interest lies in matchmaking among all of the previously assessed Crowdsourcing has several advan- requesters with suitable diets using a diets. These are separate interfaces, tages in decision support, such as col- crowdsourced approach and investi- however, and in the study presented in lecting large numbers of potential gating how the approach system com- this article, each participant only used solutions and evaluating their quality pares to the contemporary de-facto one or the other, not both. to suggest the best ones.6 Related to way of discovering diets, Google. this, TaskGenies7 uses online crowds Assessing diets to create action plans that help peo- Implementation With TDE, every diet is assessed ple to be more productive in everyday TDE is a lightweight web-based tool against a set of different criteria, chores. In general, the more specific that can be embedded on any website using a slider input element that maps plans the participants were given, using a standard HTML iFrame tag to a numerical scale from zero to 100 the more productive they became. In (we describe the Wordpress plug-in [see Figure 1(a)]. This corresponds to a similar vein, PlanSourcing demon- later). TDE was implemented using how well the diet in question intui- strated how friends and strangers HTML, JavaScript, PHP, and MySQL. tively performs in terms of the crite- alike may be leveraged to create plans In essence, it is a crowd-powered deci- rion being assessed. The numerical that lead to behavioral changes in the sion support system6 that collects data value of the scale is displayed as the form of better personal decisions.8 on a question and provides answers by user moves the handle and is supple- This study highlights an interesting querying the data. As we collect sev- mented with a verbal scale to help the characteristic, or benefit, of crowd- eral subjective, independent ratings user understand the value. TDE can sourcing: sometimes it is easier to ask for each of the available diet-criterion host an arbitrary number of diets and 64 COMPUTER WWW.COMPUTER.ORG/COMPUTER any arbitrary criteria. A diet in TDE consists of a short title (for example, “The Paleo Diet”), a longer descrip- tion of the diet (“The Paleolithic diet is predominantly focused on consum- ing only foods presumed to have been the only foods available or consumed by humans during the paleolithic era […]”), and a hyperlink to an external information source about the diet (we used Wikipedia links). In a simi- lar vein, a criterion consists of a short title (such as “Rapid weight loss poten- tial”) and a longer description. The data model for storing struc- tured subjective knowledge on arbitrary questions has been pioneered earlier in crowdsourcing settings.9 The individ- (a) ual ratings are independent and from different people and therefore usable in estimating the relations on each diet-criterion pair, based on the theo- ries behind wisdom of the crowd.3 The results of the tool are as accurate (or inaccurate) as the people bootstrapping it with their knowledge. (b) Discovering weight loss diets Once data on every diet have been col- lected, TDE is ready to be used for dis- covering diets that best match the user’s preferences. In the discovery interface, the user indicates personal importance values for the same crite- (c) ria that were used to assess the diets. All the criteria are again represented with sliders, and the same verbal cues were provided to help users with their thought processes. In Figure 1(d), the user seeks to discover diets that, accord- ing to the crowd, have very high poten- tial for rapid weight loss and provide (d) (e) all of the nutrients needed for general FIGURE 1. The screenshots of the TDE platform: (a) the interface to assess diets across well-being. The interface also allows different criteria, (b) the splash screen used in the experiment, (c) the personal informa- the user to reset the sliders and start tion collection screen, (d) the ideal criteria configuration screen, and (e) the results screen over. The wordings of the criteria were depicting the top-five matches (diets) and links to read more about them. also slightly altered since assessing JANUARY 2020 65 RESEARCH FEATURE diets and indicating the importance of their weight loss goal, general activity and answer sites, weight loss and fitness the assessment criterion are two differ- level, and biggest problems in losing forums, and social (Web 2.0) platforms.