Refine
Document Type
- Part of a Book (3)
- Working Paper (1)
Reviewed
Keywords
- Collaborative Filtering (1)
- Empfehlungssystem (1)
- Item-based (1)
- Kollaborative Filterung (1)
- Recommenderalgorithmus (1)
- Recommendersystem (1)
- User-based (1)
Institute
Is part of the Bibliography
- yes (4)
Mit dem rapiden Wachstum der Informationstechnologie zum Anfang des 21. Jahrhunderts und der damit verbundenen starken Zunahme der Datenmenge, ist Software zum Filtern und Auffinden von relevanten Informationen unverzichtbar geworden. Gerade für Unternehmen, deren Geschäftsmodell auf digitalisierten Unternehmensprozessen beruht, ist es essenziell dem jeweiligen Kunden automatisiert passende Produkte anzubieten. Recommenderalgorithmen befassen sich genau mit dieser Problemstellung und versuchen Produktempfehlungen für Kunden möglichst gut zu personalisieren.
Die bislang vorliegenden Veröffentlichungen beschreiben Recommenderalgorithmen lediglich in groben Zügen. Der vorliegende Artikel befasst sich daher mit der didaktischen Aufarbeitung zweier grundlegender Algorithmen und bietet einen transparenten und allgemein verständlichen Zugang zu diesem Thema. Die beiden Algorithmen werden anhand eines durchgängigen Beispiels von der Berechnung der Ähnlichkeitsmaße bis zur Erstellung der Empfehlungsrangliste illustriert. Darüber hinaus wird eine Methodik vorgestellt, um die Empfehlungsqualität im Rahmen von Simulationsstudien mit ROC-Kurven zu evaluieren.
During the last decades, the internet has become an increasingly important channel for businesses to sell products and communicate with customers. Web analytics helps companies to understand customer behavior and optimizes processes to satisfy the customer needs but there is still room for improvement in real-time visualization in the context of business content. In this paper, we describe a graph-based visualization showing the entirety of the website activities at a glance. To increase the tangibility of customer behavior, the graph adapts to the website interactions in real time using smooth transitions from one state to another. Furthermore, we incorporate machine learning in our data integration process to deal with the dynamics of change of website content over time. Finally, we conduct an evaluation in the form of expert interviews revealing that our approach is suitable to optimize digitalized business processes, initiate marketing campaigns, increase the tangibility to the customer, and put a stronger focus on customer needs.
Like in many other branches of modern economies, the internet changed the behavior of suppliers and customers in real estate markets. Recommender systems became more and more important in providing customers with a fast and easy way to find suitable real estate items.
In this paper, we show different possibilities for embedding the recommendation engine into the user journey of a real estate portal. Moreover, we present how additional information regarding real estate items can be incorporated into the recommendation process. Finally, we compare the recommendation quality of the state-of-the-art approaches deep learning and factorization machines with collaborative filtering (the currently used recommender algorithm) based on a data set extracted from the productive system of the Immowelt Group.
World-wide research on recommender systems has resulted in great, highly effective algorithms based on a large variety of different concepts. Two of these promising recommender approaches are the Markov Random Walk and (higher-order) Factorization Machines. Unfortunately, due to the substantial effort for optimizing hyperparameters, most articles that describe new recommender approaches do not compare the obtained results with other state-of-the-art approaches in the recommender domain.
This paper demonstrates how different state-of-the-art recommender algorithms can be compared in a consistent manner. Furthermore, we investigate under which circumstances Factorization Machines should be preferred and in which situations Markov Random Walk is the most striking algorithm. In addition, we include the restart concept into a Markov Random Walk with an optimized walk length and show how the number of factors of each order in a higher-order Factorization Machine can be optimized.