FOOD TRANSITIONS

What online data say about eating habits

Understanding how individuals shift to diets with much smaller ecological footprints may help us in persuading more people to change their habits and transition to more sustainable food systems. Online interactions provide important answers.

Food’s central role for human lives goes beyond mere sustenance and includes health and well-being, social contact and control, and culture and rituals1. Understanding what humans eat, when and why, has been the research focus of several academic disciplines for decades. Recently, a new research domain has emerged which uses traces of human behaviour — logged via interactions with online-recipe portals, search engines and social media — to learn about food preferences and habits. So-called behavioural log data offer several advantages relative to traditional methods. Despite providing rich, detailed insight, conventional methods are often restricted to a small sample of people when using direct observation, or, when using surveys, they suffer from the value-action gap — differences between reported and actual preferences.

By analysing such log data, researchers have been able to identify patterns in human behaviour, such as the overall preference for fatty and calorie-rich foods2, seasonal trends3, geographical differences4 and divergent preferences across genders5. Using such insights, it is possible to algorithmically predict the recipes that will become popular or that individual users will interact with6. Moreover, online behavioural data have been correlated with the prevalence of diet-related illness in the population7.

These analyses usually aim to understand behaviour to inform systems that can assist people to nourish themselves more healthily. Accordingly, these sorts of findings have inspired recipe-suggestion algorithms and adapted user interfaces to influence food choices. Recipe-recommending systems have been shown to ‘nudge’ users towards healthier options2. They can also help generate dietary plans that conform to government nutritional guidelines8.

Writing in Nature Sustainability, Asano and Biermann9 analyse interaction data from the most popular recipe portal in Germany, Chefkoch.de, comprising 2.5 million user interactions with over 240,000 recipes. This study is exciting for two main reasons. First, the motivation for the work is centred on sustainability, rather than on human health and lifestyle-related disease, as most previous studies in this domain have been. The premise is the idea that eating less meat and animal produce is environmentally friendly.

Second, the analytical approach is novel. Most often, this kind of research is focused on how to change behaviour. However, actual changes in habits are rarely studied. To study such observed behaviour, the authors derive an innovative method that focuses on the tags describing the content and preparation of a recipe (Fig. 1). Then, they analyse the tags applied to recipes at different levels of granularity.

Fig. 1: Data science pipeline.
figure1

icons made by Freepik and adapted from www.flaticon.com.

Schema of a data science pipeline (from input to output) to reveal behavioural dietary change from online user–recipe interaction data.

At a coarse level, they study how users interact over time with recipes featuring specific tags. The authors find a large proportional increase in the sharing of recipes with the tags ‘vegetarian’ and ‘vegan’, whereas meat-related tags such as ‘beef’, ‘pork’ and ‘poultry’ were not associated with such an increase. They also detect a rapid increase in the proportion of recipes submitted to the platform that are vegetarian (30% of the submitted recipes) and vegan (14%). The authors cite survey research to show that these findings mirror current market dynamics in Germany.

In a next step, they investigate the probability of an individual changing their behaviour, such that they interact with recipes associated with tags that they had infrequently interacted with in the past. The authors measure this with a metric referred to as the ‘infection rate of a given tag’. Moreover, they derive a metric that measures the extent to which a user continues this pattern of interaction — the tag’s ‘retention rate’. They find complementary evidence for the increasing popularity of meat-free meals. The tags ‘vegan’ and ‘vegetarian’ are associated with a high infection rate whereas the meat-based tags have low and declining infection-rates. Using both infection and retention rates, the authors are able to estimate the frequency of vegetarian and vegan diets from the user population, obtaining results close to those reported by large surveys10.

At a fine level, they explore how individual users transition from meat to meat-free recipes, by analysing the sort of recipes these users interacted with most during transitions. The authors identify strong replacement patterns; when transitioning, the recipes users liked are more similar to the original meat-based recipes than to other recipes. For example, a courgette patty replacing a meat-based burger patty.

In a final step, Asano and Biermann complement the study with qualitative data, contacting and interviewing users identified as exhibiting extreme dietary change. The interviews confirm that users prefer to transition by eating meat-free equivalents of favourite meat dishes before they learn about more novel vegetarian dishes as they become more experienced.

In conclusion, this study has two major contributions: an innovative approach to trace dietary change on a large scale using online recipe interaction; and novel findings about how this change looks. This work opens up new interesting avenues for research. For example, examining temporal dynamics of dietary change can shed light on the duration and drivers of such transitions.

References

  1. 1.

    Anderson, E. N. Everyone Eats: Understanding Food and Culture 2nd edn (NYU Press, 2014).

  2. 2.

    Elsweiler, D., Trattner, C. & Harvey, M. In Proc. 40th International ACM SIGIR Conference on Research and Development in Information Retrieval 575–584 (ACM, 2017).

  3. 3.

    West, R., White, R. W. & Horvitz, E. In Proc. 22nd International Conference on World Wide Web 1399–1410 (ACM, 2013).

  4. 4.

    Wagner, C., Singer, P. & Strohmaier, M. EPJ Data Sci. 3, 38 (2014).

    Article  Google Scholar 

  5. 5.

    Wagner, C., & Aiello, L. M. In Proc. ACM Web Science Conference 63 (ACM, 2015).

  6. 6.

    Trattner, C., Moesslang, D. & Elsweiler, D. EPJ Data Sci. 7, 20 (2018).

    Article  Google Scholar 

  7. 7.

    Abbar, S., Mejova, Y. & Weber, I. In Proc. 33rd Annual ACM Conference on Human Factors in Computing Systems 3197–3206 (ACM, 2015).

  8. 8.

    Trattner, C. & Elsweiler, D. In Proc. 26th International Conference on World Wide Web 489–498 (International World Wide Web Conferences Steering Committee, 2017).

  9. 9.

    Asano, Y. M. & Biermann, G. Nat. Sustain. https://doi.org/10.1038/s41893-019-0316-0 (2019).

  10. 10.

    Statista Vegetarismus und Veganismus (2018); https://go.nature.com/2KdrLov

Download references

Author information

Affiliations

Authors

Corresponding author

Correspondence to Christoph Trattner.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Trattner, C., Elsweiler, D. What online data say about eating habits. Nat Sustain 2, 545–546 (2019). https://doi.org/10.1038/s41893-019-0329-8

Download citation

Further reading

Search

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing