One Size Does Not Fit All: Logic-based Clustering for On-the-fly Semantic Web Service Composition and Verification

Tracking #: 1252-2464

Huynh Khai
Tho Quan
Thang Bui

Responsible editor: 
Guest Editors Quality Management of Semantic Web Assets

Submission type: 
Full Paper
Recently, web service composition has been emerging widely since it is obviously hopeless to develop a specific web service which can single-handedly fulfill completely a requirement posed from clients. Preferably automatic, this task requires an efficient mechanism to semantically well-define the web services, which are perfectly fulfilled by the usage of ontology in semantic web services. Since ontological concepts are commonly comprehensive among computer-based systems, semantic web services can be not only composed precisely, but also verified efficiently from their functionality and QoS (Quality of Services) constraints. However, composition and verification tasks always suffer from huge computational cost, which make clustering approaches naturally considered. However, typical clustering techniques neither ensure the soundness nor completeness of a composition solution. In this paper, we suggest a logic-based approach for clustering of semantic web services. The clustering results are then further applied for service composition and verification in an on-the-fly manner. In theoretical aspect, our approach achieves both soundness and completeness. As for practical result, our metric of logic-based similarity generates more reasonable clusters, resulting in significant performance improvement enjoyed in our experiments.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
By Kevin Feeney submitted on 16/Dec/2015
Review Comment:

This is an interesting paper from the point of view of service clustering for faster exploration of service composition space. The main features of the research work that is presented in this paper is a logic based clustering algorithm, which is expressed in great detail and backed up with convincing statistical evidence to support its superiority over other web-service clustering methods. The significance of this contribution is enhanced by the authors demonstrating that the algorithm is both sound and complete - i.e. that it will always find a solution where one is possible.

In practice, what this means is that clusters of services are described by meaningful expressions which allow web-service composition engines to check cluster descriptions rather than the individual web-services, thereby reducing the need to check service descriptions.

The major problem with this paper is that it is quite strongly off topic with respect to this special issue. The work is focused on web service composition and has minimal use of semantic web technologies - the input / output datatypes are taken from the OWL-S service descriptions and a simple ontology is used to help calculate similarity between the classes that appear in these service descriptions, but these are relatively minor parts of the work described in the paper, the contribution of which is the logic based clustering algorithm.

Then when it comes to quality management, the paper does not appear to be related to the topic in any way. In fact it is quite the opposite - dealing with a situation where all services are accurately semantically described and always present. It is a paper about service description clustering in the abstract and it does not deal with the messiness of reality in any way.

Review #2
By Milan Dojchinovski submitted on 02/Feb/2016
Review Comment:

The paper describes an approach for clustering and composition of semantic web services. The main contribution of the paper is in the web service clustering by utilization of a logic-based computation of web service similarities. The logic-based similarity is computed as combination of a feature based similarity and ontology based similarity between the services. Number of examples are presented that demonstrate the similarity and clustering computation.

(1) Significance of the results: the authors claim better results in terms of visited/expanded states in the process of web service composition. No direct comparison with the existing approaches from the literature.

(2) Novelty: there is no significant novelty in the proposed approach. Ideas such as feature based or distance based similarities of concepts in ontologies are already known the Semantic Web/Linked Data community.

(3) Quality of writing: the paper is difficult to read due to the many formal definitions. The paper could benefit from simplifying those formalisms. Also, some parts of the paper are difficult to understand due to grammar/spelling mistakes, which could be improved by a native speaker.

Paper structure: well structured, however:
- Section 1. Introduction: instead of only motivating the work, it also provides unnecessary background information on already well-known terms by the community such as definition of the Semantic Web, Semantic Web Services, WSDL, etc. The introduction also unnecessarily provides too much info on Web Services clustering. The contributions could be better highlighted in the introduction.
- Section 2. Motivating examples are welcome, however this section is too long.
- Section 3. Preliminaries - Since the paper has been submitted to a Semantic Web Journal, general definition of what an ontology is and similar, is not needed.
- Section 3.2. Representation of Web Service as Model - you are not presenting "web service as a model", but actually a "web service composition model"
- Section 4. it is unclear how the ontology based similarity is integrated in the final logic based similarity.
- Section 5. Case Study - very important but very short section of the paper. The paper would benefit from better explanation of the case study. It is not clear, what is the input, what is the output and how it is computed.
- Section 6. Experiments - the approach is evaluated in three aspects: the number of expanded states, the number of visited states, and the execution time. The first two aspects are not well explained - why more/less expanded/visited states is better/worse? Also, the authors do not compare their approach to the other relevant approaches in the literature.
- Section 7. Related work - is in general well written. Nevertheless, I would prefer to have it as early 2nd or 3rd section, right after the introduction/motivation.
- Most of the figures in the paper are not well aligned with the text describing them.

There are also few untrue statements in the paper:
- "Functional properties are the input and output of a web service." - the functional properties are actually described as 1) a set of operations with a certain capabilities, or 2) functional category, or pointer to a classification taxonomy.
- "OWL-S is an upper ontology of OWL" - OWL-S can not be categorized as an upper ontology. By definition, an upper ontology is an ontology with very general concepts.

The paper has been submitted to the "Quality Management of Semantic Web Assets" SWJ special issue, however, the paper is out of the scope of this special issue. In other words, the focus of the paper has nearly no connection with quality management of semantic web assets.

Given the comments above, related to the paper originality, results significance, quality of writing, and the most important - the out-of-the-scope issue, I do not recommend acceptance of the paper.

Review #3
Anonymous submitted on 21/Feb/2016
Review Comment:

The language of the document should be improved. There are far too many language mistakes. For example just in the introduction: “Software was written by various programming languages …”, or “There have been much studies”. This is not proper English.

The notion of web service composition is introduced in section 1.1 and exemplified using a simple example. The authors don’t do the same with service verification. There is no explanation is understood by service verification, neither a simple example to illustrate the concept. It seems service verification is equivalent to service discovery?

According to the authors “Semantic Web Services is the software component that provides dynamic service discovery …”. Semantic web services are not a software component! They are about describing semantically the functional, non-functional and interface aspects of a service. Semantic Web Services platforms/implementations on the other hand are those that support service discovery, composition, invocation, etc.

The notion of ontology is introduced in Section 3. What is the rational of using both attributes and relations in the ontology. It is not clear when to use one and when the other.
The approached proposed in the paper is adopting OWL-S as a mechanism to model Semantic Web Services. What was the rational for choosing this approach for semantically modeling services? Would your approach for service composition and verification work with other semantic wen services frameworks? Why not using a more recent, lighter framework for semantic web service?

The model used for modeling QoS properties is extremely basic. It is using value pairs to described QoS aspects. How does this work when modeling QoS properties such as trust or security?
According to definition 3 a Semantic Web Service is an association between a web service and an ontology. Concepts from the ontology are used to describe the inputs and outputs. However it is not clear how a certain concept or set of concepts are attached to an input or output.

Example 4 describes the SightseeingCityService. Its web service has Sightseeing=1, City=1, respTime =5. How is this in accordance with Definition 2 where pre-conditions and effects need to be provided to mode a web service?

The introduced feature-based similarity is in my opinion not a good measurement for showing if two services are similar. Let’s take for example the following example: f = X and Y -> Z and g = X and Y -> not Z. Common part in this case is a=3, while different part is b=0. The feature-based similarity is in this case 3/3+0=1. The services have the highest similarity even though they are basically providing the exact opposition output of each other.

The ontology-based similarity is introduced but after, it is never used.

The logic-based similarity is also defined in a very controversial manner. Let’s take for example the following case: f = X1 and X2 and X3 … X10 -> Y and g = X1 and X2 and X3 … X10 -> Z. fg, which is defined as the rule having in the body the conjunction of all terms in the two constituent formulae and the disjunction of all terms in the constituent formulae will be in this case fg = X1 and X2 and X3 … X30 and X1 and X2 and X3 … X10 -> Y or Z which means fg = X1 and X2 and X3 … X30 -> Y or Z. The SemFe(f, fg) will have the common part a=10, while the different part will be b = |Y,Z|/2 = 2/2 = 1, and thus SemFe(f, fg) = 30/31. For SemFe(g, fg) we get the same value 10/11. Finally to compute SimLo(f,g) we get (SemFe(f, fg)/2 + SemFe(g, fg)/2 ) / 2 = 30/31 so basically 96% similarity. Now consider that Service 1 (formula f) attached is delivering a train ticket reservation given the user data as input. Service 2 (formula g) is about computing a credit score given the same data about the user. According to logic-based similarity the two services are 96% similar even though they do totally different things. One is about getting a train ticket, the other about computing a credit score.

In the conclusions part there is a strong statement that the proposed approach performs better in terms of expanded and visited nodes as well as processing time. However the evaluation section does not support this statement as there is no evaluation compared to related approaches.