Seasonal Trends in Global Dieting Online: A Big Data Survey

Background We aimed to explore whether the massive amounts of data generated during online search interest in dieting and weight loss could be harnessed, using big data analysis, with a view to its potential incorporation in global health obesity prevention efforts. Methods We applied big data analysis to the major global health practice of dieting for weight management. Data was collected from Google and Naver search engines from January 2004 to January 2018 using the search term ‘diet’, in: A) selected six Northern and Southern Hemisphere countries, B) ve primarily Arab and Muslim countries grouped as (i) conservative, (ii) semi-conservative, and (iii) liberal, and C) South Korea.


Big-Data in Public Health
Big-Data is de ned as "data sets that are so voluminous and complex" that they overwhelm traditional data analytic methods 1 . The "three Vs"-volume, velocity, and variety-is a popular concept used to describe big-data. This re ects not just the huge volumes of data, but also the speed at which such data is generated, and the wide range of data involved. Big-data analytic methods are better suited for analyzing massive datasets in a myriad of rapidly evolving scenarios 2,3 .
One of the major advantages of big-data is that it can analyze global data cost-effectively, reliably, and accurately, The Pillbox project of the United States National Library of Medicine (NLM Pillbox) is an often cited example of how healthcare can be improved using big-data 4,5 . This powerful public service tool is a massive database that provides information on a wide range of both prescription and over-the-counter (OTC) drugs. It was designed to help users to rapidly identify such drugs based on their ingredients and appearance. Pillbox simultaneously provides and collects information based on user queries, which can enhance convenience, save costs, and improve consumer safety, among other bene ts. Big-data analysis by supercomputers such as IBM Watson that utilize machine learning and arti cial intelligence algorithms, can improve diagnosis, minimizing errors, and improving care 6 .
Another example of Big-data application was Google's in uenza forecast (Google Flu Trends). Massive amounts of data obtained from global online search patterns from dozens of countries, generating realtime insights and "nowcasts" about suspected in uenza activity worldwide. In 2009, Google predicted the spread of the u 7-10 days earlier than the U.S. Centers for Disease Control and Prevention, based on such online search data about the u 7 . Several countries, including South Korea, India, and China have compared the predictive value of such data from online searches, compared to actual numbers using traditional public health approaches [8][9][10][11] . Predictive analyses using social big-data can also reliably predict other global seasonal trends, for example, mental health issues like depression 12,13 . Big-data analyses have limitations, but upgrades and improvements are constantly provided, with the aim of improving accuracy and precision 14,15 .

Dieting
Dieting and weight loss efforts are global pursuits, considering the known health risks of obesity. Dieting mostly refers to a change in eating habits, but is also juxtaposed to increasing physical activity as part of a weight management regimen 16,17 . Obesity, especially central or visceral obesity, is an established risk factor for several diseases, especially cardiovascular disease, type II diabetes, musculoskeletal diseases, and cancer [18][19][20] . Obesity is also a factor in mental health disorders and depression [21][22][23] , with adverse effects on interpersonal relationships 24 .
Global rates of obesity is on the rise, with an obesity epidemic observed in the Arab/Muslim world 25 , with seasonal trends observed in weight loss efforts in Western societies [26][27][28] . Therefore, global health efforts aimed toward the prevention of obesity are warranted 29,30 .
With the advent on the Internet and social media, there is a heavy pro t-driven fad industry online 31 .
Much of this is wholly cosmetic, driven by a heavy emphasis on body image [32][33][34] . Such fads and food trends based on pseudo-science and quackery not only fail to deliver the results promised, but they also pose a real risk to health and well-being 32,35,36 .

Aims and goals
We aimed to explore whether the massive amounts of data generated during online search interest in dieting and weight loss could be harnessed, using Big-data analysis, with a view to its potential incorporation in global health obesity prevention efforts. We aimed to explore whether there were season trends, and perhaps an optimal time to potentially target people with online search interests in dieting? In this pursuit, we hypothesized that 'Interest or attempt to explore dieting online would be tend to be seasonal'. Our study therefore (i) examined the time series and seasonality of dieting globally, using social big-data collected from online portals, and (ii) aimed to suggest timely healthy intervention strategy based on such ndings. This search terms 'diet', 'dieting', and 'weight loss' were used, and the monthly correlation coe cient between the three words was 0.946 ~ 0.980 (Table 1). Of these, the search volume for 'diet' was overwhelmingly high. Since the terms weight loss and weight control could convey other meanings, the search term 'diet', was nally selected. K.]) and three from the Southern Hemisphere (South Africa, Australia, New Zealand) which had the highest search volumes for the search term "diet". In other words, the reference is based on the search volumes, which means that the interest in the six selected countries is the highest. As diet and weight loss are also in uenced by socioeconomic, cultural, and religious factors 39,40 , we also selected ve predominantly Arab and Muslim countries, excluding Iraq and Turkey, categorized as (i) conservative,

Searching tools and keyword
(ii) semi-conservative, and (iii) liberal 41 . South Korea was also studied using the Naver search engine, due to its local dominance as aforementioned.

Theoretical model
Cosinor analysis is a method used to evaluate the periodic ow of time series data as a cosine function.
This analysis has been frequently used to analyze body cycles (24 hours), such as circadian rhythm 42,43 , It is also used to demonstrate the seasonality of differences in blood pressure, stroke incidence, and vitamin D concentration in the 12-month follow-up period [44][45][46]

Statistical analyses
The data was analyzed by standardizing the number of periods with longest search period as 100 and the lowest as 0. In this study, we used web-search data from January 2004 to July 2018, and the time unit was one month. On the other hand, South Korea used data starting from October 2007 to March 2019, since no data from Naver was available prior to October 2007.

Descriptive statistics
The combined results from the two Hemispheres was the highest in January  Table 2).

Cosinor analysis
As a result of cosinor analysis of the Northern and Southern Hemispheres, there was seasonality (amplitude = 6.94, C.I = 5.33 ~ 8.56, P > 0.0000); searches were the highest in April and the lowest in October. As a result of analyzing only the Northern Hemisphere, seasonality (amplitude = 6.68, C.I = 5.13 8.22, P > 0.0000) was highest in April and lowest was in early October. In the Southern Hemisphere, seasonal variation of the curve was not statistically signi cant (amplitude = 1.21, C.I= -0.26 ~ 2.67, P > 0.1058) (Fig. 1).

Discussion
Obesity is an important global public health challenge, as it is a major risk factor for cardiovascular and chronic diseases, with major impact on morbidity, mortality, and health care costs. Effective management of obesity includes prevention of premature death and disability, reducing the economic burden of disease, and the promotion of healthy diets and lifestyles 29,48 .
With respect to the promotion of healthy diet and lifestyle, our aim was to analyze global dieting and weight loss trends, being cognizant that sociocultural, societal, and traditional practices could potentially play a role. Since dieting and weight loss pursuits are a global enterprise, with the Internet a major portal for disseminating information and advertisements, we considered Big data analysis ideal for studying this vast amount of data on global dieting practices and trends [49][50][51] . No attempt was made to exclude fad diets and weight loss programs, even though these also have potential health risks, including increased risks of eating disorders, mental health problems, including stress, anxiety, and depression. 52-54 32,55−60 .
In this study, we found that the search volume for the Northern and Southern Hemisphere diets was the highest in January, which coincides with the New Year, where people traditionally make New Year's resolutions following the Christmas holidays and festivities. On the other hand, for the predominantly Arab and Muslim countries, the highest search volumes were in April. For South Korea, the highest search volume was in February.
On cosinor analysis, which analyzes periodic trends, online search interest in dieting in the Northern Hemisphere was statistically signi cantly seasonal, but for the Southern Hemisphere, it was not.
Studies using cosinor analysis tend to show opposite dieting trends of Southern compared to Northern Hemisphere countries 46,61 , probably re ecting the divergent seasons. The data on global seasonal trends in dieting is apparently limited, but studies on weight changes in three major countries, including Japan, the United States, and Germany, showed a sharp increase from December, with the greatest increase in weight in early January, just after the Christmas holiday festivities 62,63 .
In this study, the search volumes of both Northern and Southern Hemispheres were the highest in January. Overall, search interest reached its peak before summer (April) in the Northern Hemisphere, and November in the Southern Hemisphere [64][65][66] . For predominantly Arab and Muslim countries, seasonality was not striking, and the magnitude was smaller than that observed for the Northern Hemisphere, with April being the highest point in the periodic rate. Seasonality tended to be a bit more pronounced in the liberal Arab and Muslim countries, compared to their more conservative counterparts.
Finally, in South Korea, data from Naver showed seasonality, with April being the peak month for online diet searches, with the trend of rhythm being similar to that of the Northern Hemisphere. However, there was no statistically signi cant seasonality in the data from Google, which may be a re ection of the lower percentage of Google searches on this topic in South Korea. Because of this skew, such Google searchers on the subject was most likely unrepresentative.
Although our study is exploratory, our Big-data analyses could suggest the potential for seasonal emphasis on weight control programs. More-cost effective health awareness and prevention weight loss strategies could harness the power of online Big-data analyses and real-time "nowcasting", for optimal timing of public health interventions for obesity. In the same way that marketing strategists use such Big data to target consumers, so too could public health authorities utilize Big data for optimal timing of public education and intervention programs.

Strengths and limitations
The authors are unaware of any previous studies to analyze the global seasonality of diets using social big-data. Big-data analysis of seasonal dieting trends is rather easy to access and analyze, and therefore potentially more cost-effective. This approach can also hold relevance to other areas of public health.
Our study has clear limitations. Firstly, we conducted the keywords search terms in English only, which is less representative for countries that do not use English as their primary search language, such as in South Korea and in the Arab and Muslim -majority countries we studied. It may therefore be more accurate to include searches using the preferred language of such countries. Secondly, it is di cult or near impossible to examine the individual characteristics of each person who performed each search, without breaching social media con dentiality or other agreements. Thirdly, we could not accurately predict the actual gures by analyzing the search volume using web-based methods only.
Big-data analysis can reliably analyze the huge metadata trends such as global seasonal patterns such as search interests in dieting and weight loss. In our study, some degree of seasonal patterns emerged, e.g. the highest search volume during the summer months in the Northern Hemisphere countries. Weight management and weight loss strategies could take such trends into account for optimal timing of their health promotion and intervention strategies. Big-data analytics, including arti cial intelligence algorithms, can be harnessed to provide cost-effective insights and optimal approaches for global health promotion and intervention programs.

Declarations
Authors' contributions M-B Park designed the study model and analyzed data. J Wang reviewed related literature and suggested the scope. BE Bulwer involved designing the study and interpreting the data. C. Ranabhat checked statistical method and framework.
Ethical Approval and Consent to participate Not applicable.

Competing interests
The authors declare that they have no competing interests.

Funding
None.