Stereotype process

The challenge

Neither content-based nor collaborative recommendation attempts to include domain-specific knowledge in the recommendation. This is a missed opportunity in terms of grouping user profiles or items in accordance with a dynamically generated cluster. Such clusters provide additional knowledge about the users or the connections of items among each other. The formation of statistical clusters is a technology known from data mining that should be made usable for the dynamic generation of personalized recommendations in order to increase the quality of recommendations.

The YOOCHOOSE solution

Stereotypes are clusters of similar content. To be able to generate such clusters, processes from either the area of content-based or collaborative recommendation are used. This creates clusters based on item metadata or the user history. Furthermore, depending on the application in question, ontology or taxonomies can also enter into the generation of stereotypes. The advantage of the patented stereotype process is that the data which are not used for generation, i.e. either the item metadata or the user history, are optimized in a second step. This advantage can be explained by means of an application example of the stereotype process. If the stereotype model was calculated based on the historic usage data of the readers of an online newspaper (i.e. based on existing user profiles), the generated clusters can be used to produce a meta-metadata model for the included items. Based on this information, the cold start problem for new items can be solved (which represents a special challenge in the news environment) without having to forgo the benefit of collaborative recommendations for news. Over the three-year development of the stereotype process, YOOCHOOSE has developed additional application scenarios in which the stereotype process can significantly increase recommendation quality based on its characteristic, thereby opening up fields of application in which no satisfactory recommendations could be generated in the past.