SPHeRe is an ontology matching system that utilizes cloud infrastructure for matching large scale ontologies and focus on alignment representation to be stored in the Mediation Bridge Ontology (MBO). MBO is the mediation ontology that stores all the alignments generated between the matched ontologies and represents it in a manner that provides maximum metadata information. SPHeRe is a new initiative therefore it only participates in the large biomedical ontologies track of the OAEI 2013 campaign. The objectives of SPHeRe system participation in OAEI is to shift focus of ontology matching community towards areas such as cloud utilization, effective mapping representation, and flexible and extendable design of the matching system.
1.1
SPHeRe system target a complete package of a system with main objectives as accuracy, mapping representation, and flexible and extendible system. Its precision is on the higher side in large biomedical ontologies track, that shows its potential of improving the accuracy. It is based on different algorithms such as String Matching Bridge, Synonym Bridge, Child Based Structural Bridge (CBSB), Property Based Structural Bridge (PBSB), and Label Bridge. We plan to include further bridge algorithms in next version of the proposed system by incorporating new matching techniques.
Parallelism has been overlooked by ontology matching systems. SPHeRe avails this
opportunity and provides a solution by: (i) creating and caching serialized subsets of
candidate ontologies with single-step parallel loading; (ii) lightweight matcher-based
and redundancy-free subsets result in smaller memory footprints and faster load time;
and (iii) implementing data parallelism based distribution over subsets of candidate
ontologies by exploiting the multicore distributed hardware of cloud platform for parallel
ontology matching and execution [
Mapping representation is another aspect of SPHeRe system which is not covered in this paper. We have followed OAEI alignment representation format, but we consider mapping representation as an important dimension to be worked by ontology matching research community. The more expressive the alignments should be, the easy its expert verification and the more will be confidence level in transformation process. 1.2
SPHeRe system is based on bridge algorithms run in the parallel execution environment to generate alignments to be stored in the MBO as shown in Fig. 1. Matcher Library components stores all the bridge algorithms to be run on the parallel execution environment represented by Parallel Matching Framework. Communication between these two components is regulated by SPHeRe Execution Control module that behaves as a controller. The alignments are stored in the MBO; generated by the bridge algorithms stored in Matcher Library that are run by the Parallel Matching Framework.
String Matching Bridge
Synonym Bridge CBSB Label Bridge PBSB
Distributor Aggregator Parallel Hardware Interface
String Matching Bridge provides matching results by finding similar concepts based
on string matching techniques in the matching ontologies. Mainly the algorithm is based
on applying edit distance technique [
Label Bridge uses the labels of the source and target concepts for matching. Initially, concept labels are normalized e.g. using stop word elimination, then list of the source concept labels are matched with list of the target concept labels. The source and target concepts label list LabelListi and LabelListj are matched using ((LabelListi \ LabelListj ) 6= ). If any label in the lists matches, the source and target concepts are stored in the MBO as mappings.
Synonym Bridge is based on finding the similarity between concepts using
wordnet [
Child Based Structural Bridge (CBSB) bridge generates mappings between source and target ontologies based on matching children of the concepts. Initially, children of source Ci and target Cj concepts are accessed as lists ChildListi and ChildListj respectively. The number of common children in the lists is identified as M atchChildren. Finally the matching value SimScore is calculated and compared with the threshold T hreshold that is assigned value n. The matching value is calculated using SimScore
M atchedChildren / Average(ChildListi, ChildListj ). Property Based Structural Bridge (PBSB) uses String Matching Bridge techniques to match properties of source and target concepts for finding similar properties. This information is utilized as in CBSB for matching the source and target ontologies concepts based on their properties. These bridge algorithms are run on a parallel execution environment for better performance of the system.
Multiphase design of SPHeRe system is represented in Fig. 2(taken from [
Serialized subsets of source and target ontologies are loaded in parallel by
multithreaded ontology distribution interface (ODI). ODI is responsible for task distribution
of ontology matching over parallel threads (Matcher Threads). ODI currently
implements size-based distribution scheme to assign partitions of candidate ontologies to be
matched by matcher threads. In a single node, matcher threads correspond to the
number of available cores for the running instance. In multi-nodes, each node performs its
own parallel loading and internode control messages which are used to communicate
regarding the ontology distribution and matching algorithms. Matched results provided
by matcher threads are submitted to accumulation and delivery phase for the MBO
creation and delivery [
Ontology Aggregation Interface (OAI) accumulates matched results provided by
matcher threads. OAI is responsible for MBO creation by combining matched results as
mappings and delivering MBO via cloud storage platform. OAI provides a thread-safe
mechanism for all matcher threads to submit their matched results. After the completion
of all matched threads, OAI invokes MBO creation process which accumulates all the
matched results in a single MBO instance [
1.4
https://sites.google.com/a/bilalamin.com/sphere/results
SPHeRe is deployed in multi-node configuration on virtual instances (VMs) over a trinode private cloud equipped with commodity hardware. Each node is equipped with Intel(R) Core i7(R) CPU, 8GB memory with Xen Hypervisor. Jena API is utilized for its inferencing capabilities. As SPHeRe is using cloud infrastructure therefore initially we have only targeted large biomedical ontologies track. The results are as follows: 2.1
SPHeRe is a cloud based ontology matching system that provides the facility to user for matching large scale ontologies without changing their hardware specifications. Figure 3 shows the results of our proposed system in large biomedical ontologies track. It has shown better precision values in almost all the tracks except task 4, while the recall of the system needs to be improved.
Tasks Task 1 Task 2 Task 3 Task 4 Task 5 Task 6
Time (s) 16 8136 154 20664 2486 10584 2359 2610 1577 2338 9389 9776 #Mappings
Scores
Incoherence Analysis Precision Recall
F-Measure
Unsat. 0.960 0.846 0.916 0.614 0.924 0.881 0.772 0.753 0.162 0.160 0.469 0.466 0.856 0.797 0.275 0.254 0.623 0.610 367 1054 805 6523 ≥ 46256 ≥ 105,418
Degree 3.6% 0.7% 3.4% 3.2% ≥ 61.6% ≥ 55.7% Performance and precision are the strengths of our system. The design of proposed system also adds to its strength as it is a extendible and reusbale system. Recall is the main weakness of our system, but with the addition of new matching techniques as bridge algorithms can improve this aspect and therefore accuracy can be improved. Extendibility allows adoption of new bridge algorithms easily into the proposed system.
New bridge algorithms incorporating new matching techniques is the next line of plan for the proposed system. Object oriented and ontology alignment design patterns are to be implemented for matching different tracks of OAEI campaign. We also tend to include instance based matching, and incorporate change management techniques in the system. 4
SPHeRe system is a new initiative that relies on parallel execution of matcher bridge algorithms for achieving better performance and accuracy. The system is still working on improving the accuracy by incorporating more matcher bridge algorithms to increase the recall value of the system. Performance of the proposed system is better as compare to other system due to running large biomedical ontologies on a single system in appropriate time.