PROJECT TITLE :

Context-aware Service Recommendation based on Knowledge Graph Embedding

ABSTRACT:

Over the course of the past two decades, context awareness has been incorporated into recommender systems in order to provide consumers not only with highly rated items but also with items that are appropriate for the context in which they are using the system. The goal of context-aware service recommendation (CASR), which belongs to the category of context-aware systems, is to connect users with high-quality services while taking into account the requirements that the users' context imposes. These requirements can include the invocation time, location, social profiles, connectivity, and so on. The current CASR approaches, on the other hand, are not scalable due to the enormous amount of service data (QoS and context information, users reviews and feedbacks). In addition, because they utilize a straightforward matrix view, they do not adequately represent the relevant contextual information. In addition, the current CASR approaches take the traditional user-service relation as their starting point, and they do not permit multi-relational interactions to take place between users and services in a variety of different contexts. We provide a comprehensive and multi-relational representation of the CASR knowledge, which is based on the idea of a knowledge graph, in order to be able to provide a scalable and context-sensitive service recommendation that is also capable of great analysis and learning. After the context-aware service knowledge graph (C-SKG) has been constructed, it is then converted into a low-dimensional vector space so that its processing can proceed more quickly. In order to accomplish this, we make use of dilated recurrent neural networks in order to propose a context-aware knowledge graph embedding. This embedding is founded on the principles of first-order and subgraph-aware proximity. In the end, a recommendation algorithm is devised in order to provide the highest-rated services in accordance with the context of the user who is being targeted. Experiments have shown that our solution is accurate and scalable in comparison to other CASR methodologies that are considered state-of-the-art.