Review Comment:
While deep learning technologies have been used in many out-of-hospital health monitoring scenarios, they are suffering from the pain of lack of training data. Collecting data from patients is time and resource consuming, and it may also have some ethical issues. This paper provides a novel stream reasoning based approach to overcome the lack of training data, and it is really appreciated that the authors are taking an empirical approach to the problem.
The paper is well-motivated, well written and well designed. Firstly, the paper clearly identifies the current research gaps on live health IoT data: lack of data interoperability, context consideration and training data. Then, this work proposes an ontology to represent the patient's medical information, activities, events and the sensors observing particular properties of the patient. The paper also illustrates constraints and situations concepts to interpret the values measured by sensors in regard to the patient's static context. The evaluation of the ontology is based on experimental analysis of real body temperature. I suggest the following improvements:
1. Section 3.4: Consider the subtitle "ontology validation" instead of "ontology evaluation". Ideally, ontology evaluation for real-world applications focuses more on completeness and expressiveness. OOPS! (OntOlogy Pitfall Scanner!) is a very good tool to detect the ontology pitfalls, but it is designed for validating the structure, architecture and design of the ontology.
2. Section 5.1: Figure 3 is a good example of the ontological representation of EMR. However, one specific example may not be enough to evidence the coverage of the ontology. It may need a quantitive analysis to make it more convincing. For example, "We experiment with the ontology on XXX number of EMR, and it covers an average XX% of the concepts."
3. Section 5.2: It could be better to clarify constraints, context and situation. Ideally, the situation refers to a high level of context and is inferred from the aggregation (fusion) of multiple pieces of context. For example, in Table 1 and Table 2, C1-C5 refer to the temperature context, while C6-C8 refer to the time context. In this case, the situation is a high-level abstraction from temperature and time context. Compared with constraint->situation, a more logical flow is constraint->context->situation.
4. Section 5.3: The expressiveness of the ontology needs a short paragraph. It could be a comparison of ontological stream reasoning with other techniques mentioned in Section 2.1. The results will be very engaging if stream reasoning shows (1) significant time reduction, (2) better performance on limited training data, and (3) successfully detect the personalisation and provide customised alert service.
Additional comments:
1. Section 3.3: The relationship (i.e., object property) of the ontology can be integrated into Figure 1, which could save some space for other sections.
2. Be careful about the use of words and punctuations. For example, on Page 4 Line 31, use "his/her" or "their" instead of "her"; On Page 1 Line 46, use ``’’ instead of "" (LaTex specific).
3. Add one short paragraph to discuss the severity of the importance. From a practical perspective, a health monitoring system should be able to distinguish the severity of the importance, such as Critical, Important, Moderate and Low, etc.
|
