Review Comment:
This manuscript was submitted as 'full paper' and should be reviewed along the usual dimensions for research contributions which include (1) originality, (2) significance of the results, and (3) quality of writing.
In this review I use the following notation to refer to specific lines: 2:1:44-47: second page, first column, lines 44 to 47
The aim of this paper is to present an approach that combines knowledge representation and reasoning and probability modeling to recommend types and quantities of fertilizers for citrus orchards.
The claimed contributions are:
1. a fertilization ontology that can be leveraged to do semantic reasoning for proper fertilizer types
2. probability modeling to estimate the expected fertilizer quantity
These contributions are meant to be evaluated in a laboratory setting where an automated fertilization machine takes automatic decisions on what and how much fertilizers to spread over samples of four different real-world orchards.
Although the approach is original and the integration with IoT-enabled automation systems would be timely and relevant for the Semantic Web Journal, the paper suffers from many issues that I detail below. I don't think these issues may be solved given the two-round publication policy of this journal. Hence, I recommend rejection, and the authors to submit a new article one these issues are solved.
The main issues of the paper can be formulated as follows:
1. the writing is very problematic making the paper hard to follow. Many typos and grammatical errors must be corrected. We can clearly identify that different non-native English speakers wrote different sections. The structure of the paper must be reworked. Many repetitions (including a whole paragraph repeated twice). Many imprecisions.
2. the originality of the paper is not demonstrated, several related references on precision agriculture are missing.
3. the proposed ontology cannot be assessed properly as it is not published online. It is not developed nor evaluated following the best practices of the Semantic Web community.
4. the paper fails at explaining how the recommended fertilizer quantities are computed. Apart from stating that a specific fertilizer is lacking/in excess, the quantities are automagically computed..
5. even though the excess/lack of raw fertilizers N, P, K, is determined, the paper does not explain how the recommendation of complex organic/non-organic fertilizers is computed (ex. rape cake, cattle urine, plantash)
6. these recommendations of complex organic/non-organic fertilizers is the main addition to the state of the art that is claimed in this paper. However, the evaluation is made only for recommendation of quantities of raw fertilizers. Also, the evaluation does not compare the actual system from the most related work [4], but only the proposed approach "with, and without" reasoning. Therefore:
- the experiment does not evaluate the proposed approach
- the experiment does not compare the proposed approach with the most related work [4].
The Introduction justifies the need for automatic irrigation/fertilization machines for precision agriculture. It is well written. It is kind of a shock to me to learn that many growers are short of expert knowledge to determine types and quantities of fertilizers. But that's off topic.
Section 2 on background and related work is probably lacking important references to past work of the authors, and other pioneer work on Semantic Web + IoT approaches for precision agriculture:
Andreas Kamilaris, Feng Gao, Francesc X. Prenafeta-Boldú and Muhammad Intizar Ali. Agri-IoT: A Semantic Framework for Internet of Things-enabled Smart Farming Applications. In Proc. of the IEEE World Forum on Internet of Things (WF-IoT), Reston, VA, USA, December 2016.
I could find many other related work, none of which are which are cited by this paper:
Sriswasdi, W., Luengsrisagoon, S., Lorsuwansiri, N., Wuttilerdcharoenwong, S., Khunthong, V., Suksaengsri, T., ... & Pusittigul, A. (2008). A smart mobilized fertilizing expert system: 1-2-3 personalized fertilizer. In World conference on agricultural information and IT, IAALD AFITA WCCA 2008, Tokyo University of Agriculture, Tokyo, Japan, 24-27 August, 2008 (pp. 397-404). Tokyo University of Agriculture.
Ramamurthy, V., Naidu, L. G. K., Kumar, S. R., Srinivas, S., & Hegde, R. (2009). Soil-based fertilizer recommendations for precision farming. Current Science, 641-647.
Yuan, Y., Zeng, W., & Zhang, Z. (2013, August). A semantic technology supported precision agriculture system: a case study for citrus fertilizing. In International Conference on Knowledge Science, Engineering and Management (pp. 104-111). Springer, Berlin, Heidelberg.
Wang, Y., Wang, Y., Yuan, Y., Guo, Y., Zhang, Z., Deng, L., & Li, L. (2014). A decision support system for fertilization and irrigation management of citrus based on semantic ontology. Transactions of the Chinese Society of Agricultural Engineering, 30(9), 93-101.
Wang, Y., & Wang, Y. (2018). Citrus ontology development based on the eight-point charter of agriculture. Computers and Electronics in Agriculture, 155, 359-370.
Boonbrahm, S., & Ruangrajitpakorn, T. (2017, December). An Expert System Using Ontology as Knowledge Base for Personalized Rice Cultivation Suggestion. In Recent Advances and Future Prospects in Knowledge, Information and Creativity Support Systems: Selected Revised Papers from the Tenth International Conference on Knowledge, Information and Creativity Support Systems (KICSS 2015), 12-14 November 2015, Phuket, Thailand (Vol. 685, p. 126). Springer.
Papajorgji, P. J., & Pardalos, P. M. (2010). advances in Modeling agricultural systems.
Section 3.2 presents the ontology. The ontology is not online nor documented, therefore it is impossible to assess this contribution. Other parts of the paper bring insight on what the ontology contains (ex. Table 3 page 11). The development methodology of the ontology is not clearly described. The contents of the ontology is not clearly described. The section contains many imprecisions, such as:
- 3:2:21 We have included more knowledge on frequently used fertilizers: what knowledge, what fertilizes, quantify.
- 3:2:24 our ontology contains many OWL definitions [...]
- 3:2:36 the newly constructed ontology consists of a large quantity of expert knowledge ...
The sentence in 3:2:28-30 has nothing to do in a paper submitted to this journal.
The first paragraphs of 3.2.2 is repeated five lines later!
The sentence 5:1:41-44 contradicts Figure 2, where other fertilizing nutrients than N, P, K, are considered.
Section 3.3 describes how the ontology is augmented by knowledge specific to a given orchard.
Step 1 add information about the 37 common types of fertilizers (list?) --> this should be moved to the previous section. It is not specific to one orchard.
Step 2 add three classes corresponding to:
(C1) the set of fertilizers that are applicable to the orchard, given its citrus plants.
(C2) the set of fertilizers that should not be applied to to orchard, given its soil and other conditions (what are these other conditions?)
(C3) the class C3 = C1-C2.
It is very unfortunate that classes C1 and C2 are constructed manually. It would be possible to use reasoning to directly infer the applicable fertilizers from the description of an orchard in terms of its citrus plants and soil. The way things are made as described in this paper still leaves too much to be explicited by the domain expert.
Section 3.4 describes the use of a Naive Bayes classifier to determine the class of the orchard given observed symptoms among nine (lack, normal amount, excess, of N, P, K). Three major issues with this sections are:
- why a Naive Bayes classifier? why not another technique? Compare?
- there should actually be 27 (3^3) classes
- The output says nothing about the actual quantity of fertilizer that should be used. And even less does it says about the quantity of one or more of the 37 frequently used fertilizers. The only mention to this is in 6:1:37-38, but (1) the knowledge base is not accessible, and (2) the reasoning process is not described.
Section 4 provides a description of four orchards that serve as the evaluation of the recommender system
7:1:49 states that the upper limit of recommended quantity of nitrogen is supplied -> what is this upper limit, is it described in the Knowledge Base?
In other places the article states that a quantity "a bit below the recommended quantity", or "a bit more than the recommended quantity" is recommended. This is extremely imprecise, nothing explains how the actual recommended quantity is obtained, nor what exact quantity "a bit below" or "a bit more" refers to.
Section 4.3 selects the most related work for comparison ([4]), and describes the evaluation methodology. This section is not well written.
The evaluation is made only for recommendation of quantities of raw fertilizers. Also, the evaluation does not compare the actual system from the most related work [4], but only the proposed approach "with, and without" reasoning. Therefore:
- the experiment does not evaluate the proposed approach
- the experiment does not compare the proposed approach with the most related work [4].
Section 5 describes the deployment and case study in a laboratory setting where an automated fertilization machine takes automatic decisions on what and how much fertilizers to spread over samples of four different real-world orchards.
The automatic fertilization machine cannot not spread any of the 37 common fertilizers. So the experiment is flawed anyways as it does not evaluate the main claims of the paper.
Table 2 provides the results of the recommendation on the three orchards, given their semantic description, reasoning, classification, and computation of the quantities of raw fertilizers (N, P, K), and computation of quantities of some of the 37 common fertilizers after what is called "reasoning". This reasoning step does not correspond to what has been described in the previous sections (quantities). Also, these quantities are "magically computed", and the experiment actually isn't led with these fertilizers.
The results in Fig 4. show that the proposed approach leads a 100% accuracy in all cases. The fact that the authors claim to have a recommendation/expectation ratio of 100% in all cases is really suspicious.. In the absence of the raw data, the program, the knowledge base, it is impossible for the readers to reproduce the experiments. I don't want to just "trust" the authors on these figures. So I prefer to believe the experiment is flawed.
Minor comments:
Abstract: The Ontology Web Language -> *The Web Ontology Language
Introduction:
Due to lacking of -> due to the lack of
2:1:44 specially -> especially?
2:2:38 and 2:2:39: why these references? totally out of scope. Later in the paper reference [23] is used, and it is perfectly adapted to defining what an ontology is.
3.1.23 stating that adopting semantic knowledge is a "trend" is not a valid scientific argument.
3:2:23-38 needs to be rephrased
3:2:35-39 needs to be rephrased
3:2:49-51 ontology has already been defined at this point of the paper!
4:1:18-29 needs to be rephrased
Fig 2: what is the difference authors make between OWL classes and ontology classes?
4:1:48-50 needs to be rephrased
4:1:51 RDF triples -> if OWL and Description Logics is considered, use the terminology of OWL consistently. (axioms. Not RDF triples)
Fig 3: blocks have the same colors if they have the same ancestor class.... all classes have OWL Thing as a common ancestor. All blocks should share the same color.
The sentence 5:1:50 - 5:2:30 is repeated again and again throughout the paper. It is the sign of a paper that is not well structured
5:2:50 multifarious -> multivarious
6:1:16 now we don't know. This is domain expert knowledge
6:1:28 which selected semantic reasoner? anyways all DL reasoners use the open-world assumption
6:2:3-8 not clear.
6:2:3-8 that deduced by the observed symptoms -> that is deduced?
6:2:14-15 each feature has its own irrelevant values -> What does this mean?
Equation 1: notations not defined
6:2:29 not clear
Equation 2: product is on ij? i is not defined nor used
7:1:17 oblivious -> obvious
8:2:3 weighs -> does not exist
8:2:8 "and other prerequisites" -> too imprecise for a scientific paper
|