Green Product Recommendations via AI

shutterstock © Konstantin Yuganov

03.08.2022
Green Product Recommendations via AI

Brain City Berlin

Keeping an eye on sustainability when shopping online is not easy. The AI-based Green Consumption Assistant (GCA) is designed to help consumers make sustainable purchasing decisions. The cooperation project between the Technische Universität Berlin, the Berliner Hochschule für Technik and the green search engine Ecosia is located in Brain City Berlin at the Einstein Center Digital Future. It is funded by the German Federal Ministry for the Environment as an AI lighthouse project.

“Sustainability” is written on a small, green label next to the product photo. And a leaf symbol next to the retailer's listing in the search engine “Ecosia” also indicates that this is an item produced in an environmentally friendly manner. The label, as well as information on sustainable product alternatives such as repair, rental or sharing options, are fed into the green search engine via an AI-driven browser extension: The Green Consumption Assistant (GCA) generates its recommendations using an AI that obtains its data from the “Green Database”. And this in turn is fed with ecological and social information.

Felix Biessmann, Professor of Machine Learning at the BHT, tells us more about the cooperation project which is funded by the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection (BMUV) as an AI lighthouse project between the Technische Universität Berlin (TU Berlin), the the Berliner Hochschule für Technik (BHT) and Ecosia. In the Green Consumption Assistant project, he is the sub-project manager responsible for the database and the Machine Learning components.

Prof. Biessmann, what goal are you pursuing as an interdisciplinary team with the Green Consumption Assistant project?

Above all, we want to help consumers make more sustainable consumer decisions online. We create the basis for the necessary functionalities with three sub-goals: First, we are building an AI-based product database – as a central point of competence for sustainable consumption. To do so, we extract information on the sustainability of products and services, have them checked by experts and then convert the data into a uniform format. These data allow us to analyse comprehensively the range of sustainable consumption options in online shopping. We also hope that making sustainability information available in GreenDB will have a catalytic effect on new innovative AI products – often new AI applications are only made possible by the provision of new datasets.

And the other two sub-goals?

We want to develop a recommendation assistant that makes sustainable product options comparable and available in real time. Sustainability information of different products and product categories is aggregated and displayed during online shopping. Resource-conscious consumption options, such as sharing or buying used products, are also becoming more visible. In addition, other features such as company ratings can be developed to support sustainable behaviour. Our third sub-goal is to adapt the application design to the behaviour of the users. Experienced UX/UI designers from Ecosia and behavioural researchers from TU Berlin are working together to develop methods and representations for this.

Why is it so difficult for consumers to find sustainable products online?

This is mainly due to the fact that information about product sustainability is often not displayed. And when such information is available, it is often unclear where these data come from. There are so many sustainability labels that refer to different aspects and require different sustainability claims – or have different certification requirements.

What is the scientific knowledge interest of the GCA team?

As part of the project, we would like to make scientific contributions in several disciplines: On the one hand, we use empirical studies such as experiments and surveys to research which presentation and which type of sustainability information most effectively encourages sustainable consumption. On the other hand, we are creating a database that is open for research purposes and is constantly growing: More than 220,000 products and their associated sustainability information have now been collected in the GreenDB. These data allow scientists to carry out new analyses of sustainable consumption options. In addition, researchers in the field of Machine Learning can use them to train new AI systems that automatically recognise and extract information relevant to sustainability.

Where does the project get its data from – and how does it learn?

The GreenDB primarily collects product data that are publicly available on the Internet. In addition, data from the evaluation of sustainability information on the Internet are used. These are partly based on publicly available product data from online shops. They are also generated from Machine Learning-based finding and extraction of product attributes, i.e. from unstructured and semi-structured data. Soon, we also want to include data from the AI-based search of sustainable product alternatives.

Can it already been seen from the Ecosia data how the service of the Green Consumption Assistant is being received?

The evaluation of quantitative and qualitative usage data from the first prototype, a browser extension, shows that while satisfaction with the first version of the GCA is high, it was not able to attract a relevant number of users. Despite the low click-through rates, even this first version of the GCA fulfilled the conditions for effective information interventions. The most important components of the browser extension have been integrated directly into the search engine in recent months. Since then, green product alternatives from the areas of clothing and electronic devices have been displayed in the shopping area of Ecosia. These are classified and selected as credible in various sustainability dimensions on the basis of assessments by sustainability experts at the TU Berlin in cooperation with the State information portal “Siegelklarheit”. These products are marked with the easy-to-understand “Sustainability” banner and a leaf icon on the product images.

What could a possible further development of the Green Consumption Assistant be like?

It is planned that in relation with the green product recommendations, in addition to the current product-based recommendations, brands and companies will also be evaluated in the future, and that recommendations will appear directly on the search page. In addition, the “sustainability” banner on the product images should be visualised in a more differentiated and multi-dimensional manner. Planned features include: a company rating based on climate promises, sustainability-related recommendations for search terms, and recommendations for resource-conserving behaviour. User interviews and A/B tests to evaluate different system variants will accompany the development of future features.

How are the TU Berlin, BHT and Ecosia working together on the lighthouse project?

The technical part of the research, especially that of Machine Learning and the provision of product data, is taken over by BHT. The sustainability experts at the TU Berlin evaluate the sustainability information and the social science research on user behaviour and acceptance of the GCA. And everything concerning the productive software, i.e. what Ecosia users see as well as the backend, is developed by the Ecosia team in close coordination with BHT.

Is the GCA team also in contact with other scientific institutions?

The project emerged from the Einstein Center Digital Future (ECDF), which was created specifically to promote transdisciplinary scientific exchange. The GCA is a good example of this. At the ECDF, and also independently of it, our team is in constant contact with researchers from all Berlin universities and other research institutions, as well as with related projects of the funding guideline.

Why is Berlin a good location for application-related interdisciplinary research in the field of AI and sustainability?

With the ECDF, Berlin has created excellent conditions for interdisciplinary exchange between Berlin’s universities. There is also very good research in Berlin in the field of Machine Learning and sustainability. And many people in Berlin who are not actively researching it are also interested in these topics. This can be seen not least from the fact that there are many young and successful companies in the areas of IT and sustainability in the city. (vdo)