Incremental Hierarchical Clustering driven Automatic Annotations for Unifying IoT Streaming Data

Abstract

In the Internet of Things (IoT), Cyber-Physical Systems (CPS), and sensor technologies huge and variety of streaming sensor data is generated. The unification of streaming sensor data is a challenging problem. Moreover, the huge amount of raw data has implied the insufficiency of manual and semi-automatic annotation and leads to an increase of the research of automatic semantic annotation. However, many of the existing semantic annotation mechanisms require many joint conditions that could generate redundant processing of transitional results for annotating the sensor data using SPARQL queries. In this paper, we present an Incremental Clustering Driven Automatic Annotation for IoT Streaming Data (IHC-AA-IoTSD) using SPARQL to improve the annotation efficiency. The processes and corresponding algorithms of the incremental hierarchical clustering driven automatic annotation mechanism are presented in detail, including data classification, incremental hierarchical clustering, querying the extracted data, semantic data annotation, and semantic data integration. The IHC-AA-IoTSD has been implemented and experimented on three healthcare datasets and compared with leading approaches namely-Agent-based Text Labelling and Automatic Selection (ATLAS), Fuzzy-based Automatic Semantic Annotation Method (FBASAM), and an Ontology-based Semantic Annotation Approach (OBSAA), yielding encouraging results with Accuracy of 86.67%, Precision of 87.36%, Recall of 85.48%, and F-score of 85.92% at 100k triple data.

Data management in digital twins: a systematic literature review

“…In the former group, (ZHANG et al, 2017) proposes the integration of data from different stakeholders (e.g., hos-pitals, pharmaceuticals, patients) to leverage the creation of new services and applications, and (ALHUMUD; HOSSAIN; MASUD, 2016) proposes a solution for managing the hospital and patient data based on the easier data exchange among different systems. As for the studies focused on the treatment of patients, there are POCs targeted at analyzing data on heart disease and diabetes (NÚÑEZ-VALDEZ et al, 2020) and predicting the risk of stroke (HUSSAIN; PARK, 2021).…”

“…(ANSARI; GLAWAR; NEMETH, 2019; HUSSAIN; PARK, 2021) created a semantic knowledge base based on domain ontologies to enrich already processed data with more domain meaning, enabling data integration. (BRECHER et al, 2021;NÚÑEZ-VALDEZ et al, 2020;RYBNYTSKA et al, 2020;ZONZINI et al, 2020) used semantic metadata annotation methods to optimize data integration by identifying new classes, relationships, or domain descriptions.…”

An intelligent edge enabled 6G‐flying ad‐hoc network ecosystem for precision agriculture

“…Greenhouse monitoring, UAV‐based crop imaging is the primary focus in their works. In the work Economic data analytic AI technique on IoT edge devices for health monitoring of agriculture machines (Gupta et al, 2020; Núñez‐Valdez et al, 2020), the authors propose. The implementation of green IoT for the agriculture machine has been conceptualized.…”

Scalable Teacher Forcing Network for Semi-Supervised Large Scale Data Streams

“…That is, it forces the weight vector to be as close as possible to its prior values before receiving pseudo labels. It is seen from the negative sign of the update formula in (23). In other words, the regularization strategy freezes the parameters of important rules.…”

“…This issue leads to an availability of only a small fraction of labelled samples (termed as scarcely labelled samples [16]), or, in the extreme case, labelled samples are only available during the warm-up phase, termed as the infinite delay problem [22]. This issue prohibits the use of fully supervised learning algorithm and requires a label enrichment mechanism via pseudo-label generation [23] to self-annotate unlabelled samples and the label augmentation approach to perturb labelled samples without changing their labels. The underlying challenge exists in the accumulation of own classification errors, which typically causes loss of generalization performance, because wrong labels are fed into the model update algorithm, thereby changing the shape of the decision boundary onto a wrong direction.…”