Here you can find publications, press releases, and other downloads about the GCA.

You would like to get in touch with us? Write to us at info[at]

Research Articles

Article: Driving Sustainable Behavior with Persuasive Technology

In the article published by Maike Gossen and Patricia Jankowski, crucial design elements are identified. It is about refining the system and evaluating its effectiveness in changing users’ behaviour. The evaluation is operationalised through the Persuasive System Design (PSD) model.


Gossen, Maike, Jankowski, Patricia (2022): Driving Sustainable Behavior with Persuasive Technology: The Green Consumption Assistant. Ökologisches Wirtschaften, 2.2022 (37), p.12-13.

Nudging Sustainable Consumption: A Large-Scale Data Analysis of Sustainability Labels for Fashion in German Online Retail

Labelling The publication addresses the question of how online retailers use sustainability pledges to inform consumers about the sustainability of products. In a study, an large data set with sustainability information on almost 17,000 fashion products of the leading online retailers in Germany were examined. The results show that many fashion products are labelled as sustainable, with two-thirds of the products using their label and one-third using a label verified by a third party. Only 14 percent of labelled products have trustworthy, third-party verified sustainability labels. This low percentage makes it difficult for consumers to understand the scope of a product’s sustainability. Moreover, the heterogeneity of labels can confuse consumers and make them feel unsure.

The practical recommendations are given in the publication support policy initiatives that address the risk of greenwashing through uncertified and insufficient sustainability information.

GreenDB: Toward a Product-by-Product Sustainability Database

One aim of our project is to build an AI-based product database that enables the development of a recommendation assistant. This paper explains how scraping technologies can be used to request publicly available data about products and information about their sustainability, and how this information can be integrated into a central database. Furthermore, the first version of the GreenDB is presented and published.

Paper in "Frontiers in Big Data"

Paper: A Benchmark for Data Imputation Methods

Improving the data quality of applications that use machine learning (ML) helps to increase their performance and enables the use of more efficient models. One of the most common problems of data quality is missing values. In this peer-reviewed article, Sebastian Jäger, Arndt Allhorn, and Felix Bießmann evaluated different methods based on different data sets that can be used to predict these missing values, which we encounter again and again with regard to sustainable product recommendations. Finally, recommendations are made for a wide variety of situations based on the results of our experiments.


With the increasing importance and complexity of data pipelines, data quality became one of the key challenges in modern software applications. The importance of data quality has been recognized beyond the field of data engineering and database management systems (DBMSs). Also, for machine learning (ML) applications, high data quality standards are crucial to ensure robust predictive performance and responsible usage of automated decision making. One of the most frequent data quality problems is missing values. Incomplete datasets can break data pipelines and can have a devastating impact on downstream ML applications when not detected. While statisticians and, more recently, ML researchers have introduced a variety of approaches to impute missing values, comprehensive benchmarks comparing classical and modern imputation approaches under fair and realistic conditions are underrepresented. Here, we aim to fill this gap. We conduct a comprehensive suite of experiments on a large number of datasets with heterogeneous data and realistic missingness conditions, comparing both novel deep learning approaches and classical ML imputation methods when either only test or train and test data are affected by missing data. Each imputation method is evaluated regarding the imputation quality and the impact imputation has on a downstream ML task. Our results provide valuable insights into the performance of a variety of imputation methods under realistic conditions. We hope that our results help researchers and engineers to guide their data preprocessing method selection for automated data quality improvement.

Working Paper

GCA Working Paper II – Driving Forces of Green Shopping Behavior

In our second publication, Robin Jadkowski looks at psychological determinants that lead to more eco-friendly purchasing decisions. This involves developing a method that matches user tracking data from our GCA with survey data and investigating the relationship of different variables to each other that could intensify eco-friendly behaviour.


This study evaluates psychological and socio-demographic driving forces of pro-environmental behavior (PEB) in an online shopping environment. Previous empirical studies substantiate the role and strengths of numerous psychological driving forces that influence individual PEB. However, the type of PEB operationalization is heterogenous, as well as the type and operationalization of the driving forces. Steg & Vlek (2009) stress the point that effective interventions have to be aimed at these driving forces and that the strengths and relationship differs for specific types of PEB. Up to date, studies that use actual behavior as operationalizations of PEB in online shopping environments are rare, therefore this study pursued two main goals: (1) Test a method to link user tracking data of a digital shopping assistant (as indicators for pro-environmental shopping decisions) with self-reported survey items; (2) Exploratively assess the relationship of four socio-demographic and 14 psychological variables that may act as driving forces for PEB. The click behavior of N = 35 beta users of an online shopping assistant was tracked over a period of five months and successfully linked to previously obtained survey data. It was not possible to reliably detect effects of the assessed driving forces – PEB relationship with the achieved sample size. In conclusion, this studies procedure revealed great potential for future research to evaluate the effects of psychological variables on PEB in a real shopping environment.

GCA Working Paper I – Scaling Sustainability Advice

We are proud to publish the first working paper for the Green Consumption Assistant project. In this paper, our colleague Cathérine Lehmann summarises our decision-making processes for sustainable product recommendations, as well as possible approaches for scaling up.


Data availability on the sustainability of products is low which poses challenges for actors from all sectors dealing with promoting sustainable consumption. We describe how we currently provide users of a Chrome browser extension with general sustainability advice and with recommendations of best-in-class products in terms of sustainability. Then we outline a possible concept towards more automatisation and thus scalability of the current approach. For the latter, we discuss six different schemes for generating large-scale green recommendations on a product level, finding that currently product sustainability can be best evaluated in terms of data availability when resorting to lists of labelled products. In the future, Product Environmental Footprints and similar data should be more easily available in order to have quantifiable data for research and for showing more information to users. Overall, an integrated approach, including e.g. aspects of  organizational sustainability, might help to fill data voids and/or to provide a more complete picture of a product’s sustainability level.

Blog Posts


03.06.2021 Article in KI-Berlin Blog

In the article titled „Using Artificial Intelligence to make more sustainable behaviour easy for people” #KI-Berlin introduces the GCA to the broader public and describes how our first iteration works.