Data asset valuation in health informatics for Spatial Environment Design (SED) is critical for bridging health data insights with effective, human-centered spatial environment planning. However, existing data asset valuation methods for SED face some limitations. In this paper, we proposed a data asset valuation method combining sparse self-attention mechanism and reinforcement learning (RL) for spatial environment design in health informatics. By centering spatial relevance, dynamic needs, and stakeholder equity, our method ensures data assets reliably translate into health-promoting spatial environment designs.

In reality, sample imbalance in medical assisted diagnosis is a common phenomenon. However, the problem of imbalanced binary samples can have several impacts on medical assisted diagnosis. To overcome the limitations, we proposed a SparseAttentionRNN-based method for explainable learning from imbalanced datasets in medical assisted diagnosis. At the data level, we use an improved SMOTE to balance the samples of majority category and minority category. At the feature level, we introduce a dropout technique based on approximate Shapley values of attribute importance to enhance the efficiency of model training. At the algorithmic level, we construct the embedding vectors based on complex valued continuous functions to better represent contextual dependencies, multiply the scaling factor by the product of the query matrix dimension and the key matrix dimension to enhance the stability of model training, and concatenate the outputs between the recurrent neural network and self-attention mechanisms. Our proposed method improved the efficiency of learning from imbalanced datasets in medical assisted diagnosis and enhanced the interpretation of attribute importance in medical assisted diagnosis.

Because using CSV files for data storage and exchange in bioinformatics has some disadvantages, it is usually necessary to load the bioinformatics CSV files into a database. However, the plain text and flexible nature of the CSV format make CSV files with Chinese headers often difficult to parse and correctly load their content. In this paper, we propose a multi-byte hash based bulk load method to load the bioinformatics CSV files with multi-byte headers into a database table. As a result, our proposed method improved data consistency and availability.

Medical insurance claims usually require the review of various medical receipts. However, it’s challenging to recognize table contents from medical receipts because of various form for-mat and printing quality. In this paper, we propose a color-wise context-splitting Transformer-based table recognition method for recognizing table contents from medical insurance claim receipts containing some handwritten invoice amounts under red stamps.

Patient Clustering for elderly care in a continuing care retirement community (CCRC) is a very significant problem on management of diversified patient data. In recent years, deep clustering methods which learn a clustering-friendly representation result in a significant increase of clustering performance. In this paper, we present a Transformer-based patient clustering method based on tag graph embedding and binomial expectation-maximization (BEM) Transformer. First, we present a tag graph dimensionality reduction algorithm to represent a patient as an embedding vector. Then, we propose a patient clustering method based on BEM Transformer where we multiply the length of query vector and the length of key vector with the scaling factor of the original Transformer [1] to make the training process more stable. Our proposed Transformer-based patient clustering method is effective for clustering the patients in the CCRCs.

A continuing care retirement community (CCRC) establishs large-scale, full-function and high-quality chain communities of medical and senior caring which combined physical and psycological feathures of the older adults by meeting multiple types of care to echo different levels of care needs for the older adults. In this paper, we propose a Transformer-based method for recommending the CCRCs. Our proposed Transformer-based method involves a 3-stages embedding layer and a self-attention layer. The first embedding stage is based on the care preference similarities of the older adults in the CCRCs. The second embedding stage is based on the promotion interaction preference similarities of the older adults out of the CCRCs. The thrid embedding stage is based on the social relationship similarities of the older adults. In self-attention layer, we multiply the length of query vector and that of key vector with the scaling factor of original Transformer to make the training process more stable.

Generally, a continuing care retirement community (CCRC) can benefit older adults by allowing them to move among and through independent living, assisted living, skilled nursing care, and Alzheimer’s care, and so on. In this paper, we propose a visit prediction method to predict next visit of the patients in the CCRC from their visit sequences of a certain in-patient room in the CCRC. There are three contributions of our proposed method. Firstly, we use a complex-valued continuous function to represent the ordered relationship in visit sequences, which replaces the representations of absolute positions in the original Transformer. Secondly, we multiply the length of query vector and the length of key vector with the scaling factor of the original Transformer to make the training process more stable. Thirdly, we evaluate the period of visit session interval based on its mean value and variance by using the maximum likelihood estimation.

The periodic trajectories of people provide an insightful and concise explanation over a long moving history, which helps to predict their future movements. In this paper, we present a Transformer based trajectory prediction method to predict the trajectories of people. First, we combine the spatial contexts of a trajectory with the sequential, temporal, and social contexts of a trajectory. Then, we predict the trajectories of people by applying Transformer based model where we multiply the length of query vector and the length of key vector with the scaling factor of the Transformer in order to make the training process more stable. As a result, our proposed Transformer based trajectory prediction model was effective.

In this paper, we present a Transformer based risk prediction method to infer the risk level from risk factors affecting the residents’ decisions on check-out from a continuing care retirement community (CCRC). In this Transformer based model, we multiply the length of query vector and the length of key vector with the scaling factor of the original Transformer in order to make the training process of our model more stable.

In nowaday, there are more and more popular smartphone apps. It is important for us to know why a certain app is chosen and whether an alternative app will be selected if the reasons using this app are changing. Once we know the reasons of current app choice, it will be helpful for us to design better app or provide better services so that more users are attracted to use this app. In this paper, we propose a causal inference model to analyze the reasons why a certain app is used. In this model, we define a regret loss function to evaluate the causal influence of a cause on its effect after an intervention on this cause. And, we use maximum likelihood method to estimate the parameter of a tranfer distribution.