Search

The prediction of chronic diseases and their comorbidities is an essential task in healthcare, aiming to predict patients' future disease risk based on their previous medical records. The accumulation of administrative data has laid a solid foundation for applying deep learning approaches in healthcare. Existing studies focused on the patients' characteristics such as gender, age and location to predict the risk of the different diseases. However, there are high dimensional, incomplete and noisy problems in the administrative data. In this research, using administrative health data, we implemented graph theory and content-based recommender system approaches to analyse and predict chronic diseases and their comorbidities. Firstly, we used bipartite graphs to represent the relationships between patients and diseases. Then, we projected this graph to a one-mode graph, namely 'disease network'. After that, six recommender system models with patient features and network features were trained. The outputs of these models are the severity levels of diseases and the predicted diseases with rank. Finally, we evaluated the performance of these models against the same models without network features. The results demonstrated that the models with network features have lower prediction error and better performances for predicting chronic diseases and their latent comorbidities on large administrative data. Among these models, the graph convolution matrix completion model reveals the least amount of error and the best performance for prediction. Further, using a case study of a specific patient, we demonstrated the application of these models in predictive disease risk analysis. Thus, this study showed the potential application of the recommender system approaches to the health sector utilising administrative claim data, which could significantly contribute to healthcare services and stakeholders. [ABSTRACT FROM AUTHOR]

In recent years, many researchers take advantage of graph convolution network in multi-behavior recommendation to further alleviate the data sparsity problem. However, most of current works directly use graph convolution network, which makes time complexity of the model relatively high. These works also ignore the different weights of neighbors and the different contributions of each behavior to user' s preference . Therefore, this paper proposed a multi-behavior recommendation model based on light graph convolution and enhanced attention ( MB-LGCA) . Firstly, the model constructed a user-item bipartite graph according to the multi-behavior data, and used a light graph convolution network to aggregate the features of neighbors to obtain high-order collaborative information. At the same time, it used attention mechanism to integrate the neighbors' weights to enhance embedding representations of nodes . It used the k-order user 's embedding propagation to obtain the different importance of each behavior to user ' s preference, so that the model had better interpretability. Finally, it combined embedding representations of different layers for prediction. The experimental results on two real datasets show that the model has better performance. [ABSTRACT FROM AUTHOR]

Neighborhood-based top-N recommendation techniques for implicit users ' inte raction data are ranking models, of which similarity functions are the important ingredients. Traditional similarity functions suffer from two major issues: sparse data and high dimensional data. Sparse data hinder the recommendation system from scoring on smooth neighborhood, high dimensional data causes the curse of dimensionality problem. This paper proposed a representation-based approach named multi object representation learning ( MO) for recommendation, MO was a bipartite node representation learning algorithm, which embedded the different level of network structures and item ordering information into nodes representations to help leverage the recommendation performance. Experimental results on three real data sets of different scales show that the algorithm has higher accuracy and recall than the commonly used recommendation model based on implicit feedback, especially for large-scale data sets, which can effectively alleviate the problems of matrix sparsity and dimensionality disaster, and improve the recommended performance. [ABSTRACT FROM AUTHOR]

Many bipartite networks describe systems where an edge represents a relation between a user and an item. Measuring the similarity between either users or items is the basis of memory-based collaborative filtering, a widely used method to build a recommender system with the purpose of proposing items to users. When the edges of the network are unweighted, the popular common neighbors-based approaches, allowing only positive similarity values, neglect the possibility and the effect of two users (or two items) being very dissimilar. Moreover, they underperform with respect to model-based (machine learning) approaches, although providing a higher interpretability. Inspired by the functioning of Decision Trees, we propose a method to compute similarity that allows also negative values, the Sapling Similarity. The key idea is to look at how the information that a user is connected to an item influences our prior estimation of the probability that another user is connected to the same item: if it is reduced, then the similarity between the two users will be negative, otherwise it will be positive. We show that, when used to build memory-based collaborative filtering, Sapling Similarity provides better recommendations than existing similarity metrics. Then we compare the Sapling Similarity Collaborative Filtering (SSCF, an hybrid of the item-based and the user-based) with state-of-the-art models using standard datasets. Even if SSCF depends on only one straightforward hyperparameter, it has comparable or higher recommending accuracy, and outperforms all other models on the Amazon-Book dataset, while retaining the high explainability of memory-based approaches. [ABSTRACT FROM AUTHOR]