Künstliche Intelligenz: OTTO revolutioniert Produktbewertungen

Do the sneakers look bigger? What does the material feel like? How comfortable are they? Many customers may well have already asked these questions themselves – and publicly answered them in their review on otto.de after trying the products out. To help other customers keep track of hundreds of product reviews and benefit from the experience of others, OTTO now shows a review of specific aspects that customers can then use to filter out the information they need. Such ‘aggregated reviews’ also appear in all other assortments and therefore provided for over 2.1 million articles on otto.de – with washing machines, for example, where factors that require explanations such as ‘operation’ or ‘washing result’ are the most important.

The as-yet unique, self-developed product-assessment algorithm from OTTO filters out the most frequently named aspects from the customer reviews. Customers can then sort the comments that other users have made most often in terms of these aspects. Beyond that it also shows which percentage of users have expressed themselves positively, negatively or neutrally about the product.

“No-one wants to buy a washing machine enough to want to read over 1,700 evaluations. Yet user feedback is an important source of information from other users, and a crucial factor in the buying process. So it’s all the more important that customers get easy access to the information they require”, says Marc Opelt, Member of the OTTO Management Board, Marketing and Spokesman OTTO.

Powered by artificial intelligence: each assessment analysed by algorithm

The self-developed solution from OTTO is based on artificial intelligence. Or to put it more precisely, it applies an algorithm based on machine-learning principles and analyses each review text. The algorithm is ‘fed’ over a million product reviews every night; it then automatically analyses aspects from these text entries and identifies them as positive, negative or neutral, and finally groups these aspects in semantically similar sets. ‘Deep Learning’ is deployed for this process – learning based on artificial neural networks.

“The subject of artificial intelligence will continue to shape the online retail of the future. Our objective must always be to give our customers genuine added value with the aid of innovative technologies. The new filter function for product evaluations on otto.de is therefore a good example of the smart deployment of machine-learning methods that will make the buying experience on otto.de even better”, says Marc Opelt.

Information on customer evaluations on otto.de

In an internal study in 2016, the online retailer OTTO determined that two-thirds of people surveyed consulted the customer assessments before buying in the last twelve months. Customer assessments on otto.de vary from one assortment to another. So as a rule, technical articles are assessed much more frequently and in greater detail than fashion products. The reason for this is in the simpler technical characteristics that can be more objectively reviewed than with articles of clothing.

Explanations of AI, Machine Learning and Deep Learning

The broad term ‘AI’ (Artificial Intelligence) is used from the moment when technology can perform intelligent, human-like tasks. ‘Machine Learning’ is the term used for self-learning systems that recognise patterns from examples and can apply this knowledge to unknown data. With ‘Deep Learning’, artificial neural networks are used to ensure that learning success is achieved more quickly. With the help of Deep Learning, technology is able to identify structures within large quantities of data, for example, and to constantly improve itself in terms of more precise recognition.