Network representation learning has attracted much attention as a new learning paradigm to embed network vertices into a low-dimensional vector space, by preserving network information. In this paper, in the light of user co-tag network and social network, we introduced network representation learning techniques into the learning of user preference, to encode user social relations into a continuous vector space. Firstly, we proposed a hybrid network representation learning approach to effectively utilize users’ tagging and social relationships, and then we took it for service recommendation. The experiment results show that the improvements of performance, compared with three baselines on two public data sets, the improvement ratio over the baselines is up to 50% in terms of Recall@10 and Precision@10, and the improvement is even more than 90% in terms of NDGG@10 and MRR@10.
Medical concept embedding which aims at learning interpretable low-dimensional representations of medical codes has become one of the key technologies to enable the machine (deep) learning models to imitate the doctor’s cognitive reasoning process in a variety of clinical tasks. Most existing works focus on leveraging the medical ontology to get the representations but remains ineffective in dealing with 1) the inconsistency between the knowledge of the medical ontology and the observations in health records, and 2) the deficiency of discovering the relations among multi-types of medical concepts. To address these challenges, this paper proposes MrER(Multi-relational EHR representation learning method). It’s a heterogeneous graph convolutional network with a self-adaptive adjacency matrix, to infer the multi-relations among different types of medical concepts and align them in the same subspace for the complex knowledge inference. Moreover, an temporal convolutional network is introduced to capture the dependency patterns in the sequence of medical records. The entire framework is trained in an end-to-end fashion. The experimental results show that MrER achieves competitive performance advantages in sequential diagnosis prediction task in comparison with state-of-the-art methods and the learned embeddings have good interpretability regarding the relationship between medical codes.
POI recommendation aims to predict the locations that users may be interested in at next time based on the user's historical check-in sequence information. It is a key task to improve customer experience and business operations, which has aroused widespread interest in academia and industry. But it is still challenging due to the diversity of human activities and the sparseness of the available check-in records. In order to cope with these challenges, The paper proposes a recommendation method based on graph representation learning: Temporal-Regional based Graph Convolutional Network (TRGCN) to further improve the accuracy of prediction. The model first builds a multi-graph representation based on the user's check-in record, and at the same time integrates contextual information such as time period and region into the graph. After that, the model learns the representation of each node at a specific time through the graph neural network. In addition, we apply different score functions to evaluate users' preferences for POIs and regions. The experiment performance of TRGCN proves the effectiveness of constructing a multi-graph structure of user check-in records based on spatio-temporal context to learn the representation of graph nodes. In addition, there is a strong correlation between sequence data. Experiments have also proved the effectiveness of recurrent neural network (or its variants) in processing sequence data. At the same time, the attention mechanism will learn more important information from the sequence to further improve model performance