Reorganization of historical data to support the analysis of organization's behavior

Abstract

If the historical data about the operations of an organization can be systematically and continually used, the analysis of organizational behavior for management control will be much easier and effective. Based upon Ashby's concept of black box and protocol, the paper proposes an approach to reorganizing the historical data of an organization. A data model containing two time variables to represent the cycle time of operations in the organization and the time length for behavior analysis has been developed. Using this model, a new data base containing a) the data files of all protocols describing the input state and output state of the activities concerned and b) the numerical data files of the statistics of interactions between and/or among entities concerned can be created from historical data removed from daily operational data base (DODB). The evidence from a case study in a general teaching hospital in Taiwan shows that the proposed approach can be implemented and the use of historical data to support the analysis of organization's behavior analysis for management control can be realized.

[Retracted] Multichannel CNN Model for Biomedical Entity Reorganization

“…As shown in Table 3 and Figure 6 , the evaluation results on the DDI dataset show that the method proposed in this paper is comparable to other methods. SCNN [ 16 ] introduced syntactic information into the model by analyzing sentence structure and training syntactic word vectors. This approach has achieved some results, but some errors are generated when parsing the sentence structure, which will lead to the accumulation of model errors.…”

“…The Transfer Model [ 11 ] adopts the transfer learning method to initialize the model through the network parameters of related tasks so that the F value reaches 61.5%. The Ensemble Model [ 16 ] includes three independent models of SVM, CNN, and Bi-LSTMs, adds additional features to construct manually, and then obtains the highest accuracy through voting. Still, the model is relatively complex and challenging to transfer to others.…”

“…First, by designing rich semantic features, determine whether there is a relationship between entities and then further classify the relationship, which is cumbersome. Recently, deep learning methods have developed rapidly, and the literature [ 16 ] proposed syntactic CNN, which uses syntactic word embedding to improve the model's performance. Reference [ 1 ] used the voting method to identify compound-protein relationships in the literature by combining models such as LSTM and CNN.…”

“…These deep learning models use static word vectors and cannot generate corresponding word vectors for new and combined words. In addition, since the imbalance of positive and negative samples in the corpus affects the model's performance, previous methods usually use a model combination or filter samples to ensure sample balance [ 16 , 23 ]. Still, filtering samples requires solid professional knowledge to design filtering rules, which the process is cumbersome and reduces the generalization of the model.…”