DualTaxoVec: Web user embedding and taxonomy generation

Learning web user embedding based on interaction data in the context of taxonomy is a way of studying the correlation between two web users. Such user embedding is important for further user analysis. Interaction data is made up of users and the items they interact within a domain, which is a group of entities with a basic common property. Usually a taxonomy of these items that users interact with is a hierarchical category structure for a domain. However, the taxonomy is not totally suitable for a particular task. To solve this problem, we propose a dual-way method DualTaxoVec, which learns the user embedding based on the taxonomy of the user interaction items. Meanwhile, it automatically constructs the taxonomy for the items that adapts the domain of users. It is composed of user–item and item–user tracks to construct the taxonomy and embed users in a dual-way. According to the experimental results, the validity and effectiveness of the DualTaxoVec has been demonstrated. •The DualTaxoVec takes use of the product taxonomy for generating web user embeddings.•It can also update the product taxonomy according to the generated embeddings.•The method is validated and superior to the popular and outdated methods.