Open USN service is USN service to enable enhanced and flexible service interoperability and provisioning based on the use of standards interfaces. In this paper is to define an open USN service framework, and provide reference architecture of open USN service framework. The use of standard interfaces of open USN service framework will ensure USN service reusability, portability across several USN services, as well as accessibility and interoperability by USN application providers and/or developers. This paper will contribute to the development and activation of new variety USN service by deploying sensor node constructed on variety field or sensor network to share and utilize at different service field.
A Chinese Diseases Question Answering System(Hestia QA) is being developed by ISTIC. As a part of Hestia QA, a XML-based document retrieval and similarity calculation model is established here. The texts which describe diseases in Chinese are indexed and wrapped in XML tags. The query is compared with related tags in XML document and the similarity is calculated with a deformed cosine similarity algorithm. The Chinese terms semantic similarity calculation algorithm is used to get the similarity of two terms in the system. The result shows that our model works well. The Chinese disease XML datasets will be analyzed in different granularity levels or dimensions. The corpus of diseases in Chinese will be established after the automatic XML annotation software is completed in the next step.
The performance of ensemble, including AdaBoost, is determined by accuracy and generalization ability. However, the currently available AdaBoost’s weak learners mostly show high accuracy but rather low generalization ability. In this paper, we introduce three requirements that weak learners must satisfy in order to improve generalization ability of AdaBoost. Then, we propose w-delta learning rule based neural network(NN) as a weak learner that satisfies those requirements. Through experiments, we show that the proposed method improves generalization ability while maintaining the high accuracy.
In this paper we have proposed a fuzzy c-mean based clustering algorithm for categorization of different types of objects present in a heterogeneous information network. We have addressed a particular scenario in this paper when exact structure of objects and their relationships with other objects is either hidden or not known. We have performed the experiments on an agriculture information network and our results depicts that combining automatic extraction of structure of an information network with information objects can improve the quality of clustering.
The data set of forest cover types based on cartographic data consists of very large data set of 581,102 instances. So, decision tree-based data mining methods that need relatively less computing resources could be used for better classification models. Random forests consisting of multitude of special decision trees are known to be a good data mining tool, and a technique based on grid search of random forests was investigated to find very accurate classifier. Experiments showed that a classifier of high accuracy could be found for the data set of forest cover types.
