René GötzIdentifikation von nutzerorientierten Produktähnlichkeiten mithilfe künstlicher Intelligenz für Empfehlungsagenten in der Kleidungsindustrie
Studien zur Wirtschaftsinformatik, volume 108
Hamburg 2021, 234 pages
ISBN 978-3-339-12500-2 (print)
ISBN 978-3-339-12501-9 (eBook)
About this book deutschenglish
The high product variety in online retailing often leads to consumers being overwhelmed and ultimately abandoning their purchases.
To counteract this flood of information, product recommendation methods are used that can provide consumers with individualized and personalized products. Product suggestions are often based on historical click and purchase behavior with the goal of reflecting the preferences and wishes of consumers. However, click behavior alone is often not sufficient to fully understand consumers and interpret their behavior.
The author of this paper first considers various data sources that can reflect preferences regarding specific products from a consumer perspective. Explicit customer opinions in the form of product reviews as well as click behavior, product images and product attributes provide relevant perspectives for this purpose. Unstructured text data first needs to be preprocessed using Natural Language Processing methods and is then getting converted into a uniform data model. The set of product attributes primarily contains information regarding various color metrics (saturation, brightness, etc.) and the presence of individual construction and design elements of a product. Based on consumer click behavior, products can be identified which are frequently viewed together within a session. The various data sources are used to identify product similarities. Machine Learning (Word2Vec) and Deep Learning (Variational Autoencoder) methods are used for this task.
The result of the algorithms is the representation of products as vectors in a multidimensional vector space, which can be compared with each other based on distances. The different perspectives on product similarities are investigated and compared based on the use case of product recommendation. In addition, a web-based recommendation agent is developed to serve as a decision support tool for product search in online retail. Users can interactively narrow down the product selection based on their individual preferences and have suitable alternatives displayed.
Keywords
AIAutoencoderBetriebswirtschaftDeep LearningDesign Science ResearchEmpfehlungsagentInformatikIntelligenzKIKleidungsindustrieMachine LearningNatural Language ProcessingProduktempfehlungenTechnology Acceptance ModelWirtschaftsinformatikWord2VecIhr Werk im Verlag Dr. Kovač
Möchten Sie Ihre wissenschaftliche Arbeit publizieren? Erfahren Sie mehr über unsere günstigen Konditionen und unseren Service für Autorinnen und Autoren.