We Divide, You Conquer:

Abstract

Large ontologies still pose serious challenges to state-of-the-art ontology alignment systems. In this paper we present an approach that combines a lexical index, a neural embedding model and locality modules to effectively divide an input ontology matching task into smaller and more tractable matching subtasks. We have conducted a comprehensive evaluation using the datasets of the Ontology Alignment Evaluation Initiative. The results are encouraging and suggest that the proposed methods are adequate in practice and can be integrated within the workflow of state-of-the-art systems.

Introduction

Large-scale ontology matching tasks still pose serious challenges to ontology alignment systems. For example, only 6 out of 10 systems participating in the OAEI 2017 largebio track were able to complete the largest tasks [ 2 ]. OAEI systems are typically able to cope with small and medium size ontologies, but fail to complete large tasks in a given time frame and/or with the available resources (e.g., memory). Prominent examples across the OAEI campaigns are: (i) YAM++ version 2011 [ 3 ] (best results in conference track, but failed to complete the anatomy task); (ii) CODI version 2011.5 [ 4 ] (best results in anatomy but could not cope with the largebio track); (iii) MAMBA version 2015 [ 5 ] (top system in the conference track but could not complete the anatomy track); (iv) FCA-Map version 2016 [ 6 ] (completed both anatomy and phenotype tasks but did not complete the largest largebio tasks); and (v) POMap version 2017 [ 7 ] (one of the top systems in anatomy but could not finish the largest largebio tasks).

In this paper we propose a novel method to effectively divide the matching task into several (independent) smaller subtasks. This method relies on an efficient lexical index (as in LogMap [ 8 ]), a neural embedding model [ 9 ] and locality modules [ 10 ]. Unlike other state-of-the-art approaches, our method provides guarantees about the preservation of the coverage of the relevant ontology alignments as defined in Section 2.2. In this section we introduce the background concepts that are used throughout the paper. ? An extended version of this paper is available in arXiv.org [ 1 ]. 2.1

Basic definitions

A mapping (also called match or correspondence) between entities1 of two ontologies2 O1 and O2 is typically represented as a 4-tuple he1; e2; r; ci where e1 and e2 are entities of O1 and O2, respectively; r 2 fv; w; g is a semantic relation; and c is a confidence value, usually, a real number within the interval (0; 1]. In our approach we simply consider mappings as a pair he1; e2i. An ontology alignment is a set of mappings M between two ontologies O1 and O2.

An ontology matching task MT is composed of a pair of ontologies O1 (typically called source) and O2 (typically called target) and possibly an associated reference alignment MRA. The objective of a matching task is to discover an (implicit) overlapping of O1 and O2 in the form of an alignment M. The size or search space of a matching task is typically bound to the size of the Cartesian product between the entities of the input ontologies: jSig(O1)j jSig(O2)j being Sig(O) the signature (i.e., entities) of the ontology O.

An ontology matching system is a program that, given as input the ontologies O1 and O2 of a matching task, generates an ontology alignment MS .

The standard evaluation measures for an alignment MS are precision (P), recall (R) and f-measure (F) computed against a reference alignment MRA as follows: P = jM

S \ M

S jM j

RA j ; R = jM

S jM \ M

RA j RA j ; F = 2

P R P + R 2.2

Matching subtasks and quality measures: size ratio and coverage

We denote division of an ontology matching task MT , composed by the ontologies O1 and O2, as the process of finding matching subtasks MTi = hO1; O2ii (with i=1,. . . ,n), i where O1i O1 and O2i O2. The size of the matching subtasks aims at being smaller than the original task in terms of search space. Let DMnT = fMT1; : : : ; MTng be the result of dividing a matching task MT . The size ratios of the matching subtasks MTi and DMnT are computed as follows:

SizeRatio(MTi; MT ) = jSig(O1i)j jSig(O1)j jSig(O2i)j jSig(O2)j SizeRatio(DMnT ; MT ) = n X SizeRatio(MTi; MT ) i=1 (1) (2) (3)

The ratio SizeRatio(MTi; MT ) is expected to be less than 1:0 while the aggregation Pn

i=1 SizeRatio(MTi; MT ), being n the number of matching subtasks, can be greater than 1:0 (as matching subtasks may overlap).

The coverage of the matching subtask aims at providing guarantees about the preservation of the (potential) outcomes of the original matching task (i.e., information loss). That is, it indicates if the relevant ontology alignments in the original matching task can still be computed with the matching subtasks. The coverage is calculated with respect to a relevant alignment M, possibly the reference alignment MRA of the matching task if it exists. The formal notion of coverage is given in Definitions 1 and 2. 1 We refer to (OWL 2) classes, data and object properties and named individuals as entities. 2 We assume ontologies are expressed in OWL 2.

Definition 1 (Coverage of a matching task). Let MT = hO1; O2i be a matching task and M an alignment. We say that a mapping m = he1; e2i 2 M is covered by the matching task if e1 2 Sig(O1) and e2 2 Sig(O2). The coverage of MT w.r.t. M (denoted as Coverage(MT ; M)) represents the set of mappings M0 M covered by MT .

Definition 2 (Coverage of the matching task division). Let the result of dividing a

matching task MT be DMnT = fMT1; : : : ; MTng and M an alignment. We say that a mapping m 2 M is covered by DMT if m is at least covered by one of the matching subtask MTi (with i=1,. . . ,n) as in Definition 1. The coverage of DMT w.r.t. M (denoted as Coverage(DMT ; M)) represents the set of mappings M0 M covered by DMT . The coverage is often given as a ratio with respect to the (covered) alignment: CoverageRatio(DMnT ; M) = jCoverage(DMT ; M)j jMj (4)

2.3 Locality-based modules in ontology alignment

Logic-based module extraction techniques compute ontology fragments that capture the meaning of an input signature with respect to a given ontology. In this paper we rely on bottom-locality modules [ 10 ], which will be referred to as locality-modules or simply as modules. Locality modules play an important role in ontology alignment tasks. For example, they provide the context, i.e., sets of semantically related entities [ 10 ], for the entities in a given mapping or set of mappings as formally presented in Definition 3. Definition 3 (Context of a mapping and an alignment). Let m = he1; e2i be a mapping between two ontologies O1 and O2. We define the context of m (denoted as Context(m; O1; O2)) as a pair of modules O10 O1 and O20 O2, where O10 and O20 include the semantically related entities to e1 and e2, respectively [ 10 ]. Similarly, the context for an alignment M between two ontologies O1 and O2 is denoted as Context(M; O1; O2) = hO10; O20i, where O10 and O20 are modules including the semantically related entities for the entities e1 2 Sig(O1) and e2 2 Sig(O2) in each mapping m = he1; e2i 2 M.

2.4 Context as matching task

The context of an alignment between two ontologies represents the (explicit) overlapping of these ontologies with respect to the aforesaid alignment. Intuitively, the ontologies in the context of an alignment cover all the mappings in that alignment. Definition 4 formally presents the context of an alignment as the overlapping matching task to discover that alignment.

Definition 4 (Overlapping matching task). Let M be an alignment between O1 and O2, and Context(M; O1; O2) = hO10; O2i the context of M. We define MTOM1-O2 = 0 hO10; O20i as the overlapping matching task for M. A matching task MT = hO1; O2i can be reduced to the task MTOM1-O2 = hO10; O20i without information loss in terms of finding M.

A matching system should aim at computing M with both the original matching task MT and the reduced task MTOM1-O2 . For example, in the small OAEI largebio tasks [ 2 ] systems are given, instead of the original matching task (e.g., whole FMA and NCI ontologies), the context of the reference alignment as a (reduced) overlapping

RA RA matching task (e.g., MTfma-nci = Context(Mfma-nci; OFMA; ONCI) = hOF0MA; ON0CIi). The approach presented in this paper relies on an ‘inverted’ lexical index (we will refer to this index as LexI), commonly used in information retrieval applications, and also used in ontology alignment systems like LogMap [ 8 ]. LexI encodes the labels of all entities of the input ontologies O1 and O2, including their lexical variations (e.g., preferred labels, synonyms), in the form of pairs key-value where the key is a set of words and the value is a set of entity identifiers3 such that the set of words of the key appears in (one of) the entity labels. Table 1 shows a few example entries of LexI for two input ontologies.

LexI is created as follows. (i) Each label associated to an ontology entity is split into a set of words; for example, the label “Lunate facet of hamate” is split into the set f“lunate”, “facet”, “of”, “hamate”g. (ii) Stop-words are removed, for example,“of” is removed from the set of words (i.e., f“lunate”, “facet”, “hamate”g). (iii) Stemming techniques are applied to each word (i.e., f“lunat”, “facet”, “hamat”g). (iv) Combinations of (sub)set of words serve as keys in LexI; for example, f“lunat”, “facet”g, f“hamat”, “lunat”g and so on.4 (v) Entities leading to the same (sub)set of words are associated to the same key in LexI, for example, the entity O1:Lunate facet of hamate with numerical identifier 55518 is associated to the LexI key f“hamat”, “lunat”g (see Table 1). Finally, (vi) entries in LexI pointing to entities of only one ontology are not considered (see last two rows of LexI in Table 1). Note that a single entity label may lead to several entries in LexI, and each entry in LexI points to one or many entities.

Each entry in LexI, after discarding entries pointing to only one ontology, is a source of candidate mappings. For instance the example in Table 1 suggests that there is a (potential) mapping m = hO1:Serous acinus; O2:Liver acinus; ; ci since the entities O1:Serous acinus and O2:Liver acinus are associated to the same entry in LexI facinusg. These mappings are not necessarily correct but link lexically-related entities, that is, those entities sharing at least one word among their labels (e.g., “acinus”). Given a subset of entries of LexI (i.e., l LexI), the function Mappings(l) = Ml provides the set of mappings derived from l. We refer to the set of all (potential) mappings suggested by LexI (i.e., Mappings(LexI)) as MLexI. Note that MLexI represents a manageable subset of the Cartesian product between the entities of the input ontologies. 3 The indexation module associates unique numerical identifiers to entity URIs. 4 In order to avoid a combinatorial blow-up, the number of computed subsets of words is limited.

Most of the state-of-the-art ontology matching systems rely, in one way or another, on lexical similarity measures to either discover or validate candidate mappings [ 11 ]. Thus, mappings outside MLexI will rarely be discovered by standard matching systems.

3.2 Creation of matching subtasks from LexI

Considering all entries in LexI (i.e., one cluster) may lead to a very large number of candidate mappings MLexI. The context of MLexI leads to (two) large overlapping modules O1LexI and O2LexI that, although smaller than the input ontologies O1 and O2, may still be challenging for many ontology matching systems. A solution is to divide the entries in LexI in more than one cluster.

Definition 5 (Matching subtasks from LexI). Let MT = hO1; O2i be a matching task, LexI the lexical index of the ontologies O1 and O2, and fc1; : : : ; cng n clusn = ters of entries in LexI. We denote the set of matching subtasks from LexI as DMLeTxI = fMT1LexI; : : : ; MTnLexIg where each cluster ci leads to the matching subtask MTi hO1i; O2ii, such that Mappings(ci) = MiLexI is the set of mappings suggested by the LexI entries in ci and O1i and O2i represent the context of MiLexI w.r.t. O1 and O2. Hypothesis 1 If MT = hO1; O2i is a matching task and MS the mappings computed for MT by a lexical-based matching system, then, with independence of the clustering strategy of LexI and the number of subtasks n, DMnT = fMT1LexI; : : : ; MTnLexIg will cover (almost) all the mappings in MS (i.e., CoverageRatio(DMnT ; MS ) 1:0).

Hypothesis 1 suggests that a matching system will unlikely discover mappings with MT = hO1; O2i that cannot be discovered with DMnT = fMT1LexI; : : : ; MTnLexIg . This intuition is supported not only by the observation that most of the ontology matching systems rely on lexical similarity, but also by the use of the notion of context (see Definition 3 and Definition 4) in the creation of the matching subtasks.

Intuitively each cluster of LexI leads to a smaller set of mappings MiLexI (with respect to MLexI) and to a smaller matching task MTiLexI (with respect to both MT LexI and MT ) in terms of search space. Hence SizeRatio(MTiLexI; MT ) is expected to be smaller than 1:0, as mentioned in Section 2.2. Reducing the search space in each matching subtask MTiLexI has the potential of enabling the use of systems that can not cope with the original matching task MT in a given time-frame or with (limited) computational resources. The aggregation of ratios may be greater than 1:0 and will depend on the clustering strategy. 5 The number of clusters n is a parameter given as input. See Section 6 for a discussion of possibles ways of automatically obtaining n.

Hypothesis 2 Given a matching task MT and an ontology matching system that fails to complete MT under a set of given computational constraints, there exists a division of the matching task DMnT = fMT1LexI; : : : ; MTnLexIg for which that system is able to compute an alignment of the individual matching subtasks MT1LexI; : : : ; MTnLexI under the same constraints. 3.3

Clustering strategies

We have implemented two clustering strategies which we refer to as: naive and neural embedding. Both strategies receive as input the index LexI and the number of desired clusters n, and provide as output a set of clusters fc1; : : : ; cng from LexI. As in Definition 5, these clusters lead to the matching subtasks in DMnT = fMT1LexI; : : : ; MTnLexIg.

The choice of strategy, according to Hypothesis 1, will not have an impact on the coverage; but it may influence the size of the matching subtasks. Note that, neither of the strategies aims at computing optimal clusters of the entries in LexI, but clusters that can be efficiently computed.

Naive strategy. This strategy implements a very simple algorithm that randomly splits the entries in LexI into a given number of clusters of the same size. The matching tasks resulting from this strategy are expected to have a high overlapping as different entries in LexI leading to similar set of mappings may fall into different clusters. Although the overlapping of matching subtasks will impact the global search space, it is still expected to be smaller than in the original matching task.

Neural embedding strategy. This strategy aims at identifying more accurate clusters, leading to matching tasks with less overlapping, and thus, reducing the global size of the computed division of the matching task DMnT . It relies on StarSpace toolkit6 and its neural embedding model [ 9 ], which aims at learning entity embeddings. Each entity7 is described by a finite set of discrete features (bag-of-features). The model is trained by assigning a d-dimensional vector to each of the discrete features in the set that we want to embed directly. Applied to the lexical index LexI, the neural embedding model would learn vector representations for the individual words in the index keys, and for the individual entity identifiers in the index values. Since an index key is a set of words (see Table 1), we use the mean vector representation of the vectors associated to each word. Based on these aggregated neural embeddings we then perform standard clustering with the K-means algorithm.

Hypothesis 3 There exists a number of clusters or matching subtasks ‘n’ for which the clustering strategies can compute DMnT = fMT1LexI; : : : ; MTnLexIg for a given matching task MT such that SizeRatio(DMnT ; MT ) < 1:0.

Hypothesis 3 suggests that there exists a division DMnT of MT such that the size (or search space) of DMnT is smaller than MT , and DMnT can be computed by the proposed naive and neural embedding strategies. In this section we support Hypothesis 1-3 (Section 3). We rely on the datasets of the Ontology Alignment Evaluation Initiative (OAEI) [ 2 ], more specifically, on the matching tasks provided in the anatomy, largebio and phenotype tracks (see Table 2).

The methods have been implemented in Java8 and Python9 and were tested on a Ubuntu Laptop with an Intel Core i7-4600U CPU@2.10GHz (4 cores). Up to 15 Gb of RAM was allocated. The next sections present the performed experiments.10 n We have evaluated the adequacy of the clustering strategies to compute divisions DMT = fMT1LexI; : : : ; MTnLexIg for each of the matching tasks in Table 2 with respect to the available reference alignments. We report results in terms of coverage (as in Equation 4) and size (as in Equation 3) of the resulting division DMnT of the matching tasks.

We have compared the two strategies for different number of clusters or resulting matching subtasks n 2 f2; 5; 10; 20; 50; 100; 200g. For the naive strategy, as a random split of LexI is performed, we run 10 experiments for each of the values of n to evaluate the effect of different random selections. The variations in the size of the obtained matching tasks was negligible. Results represent the average of the 10 experiments Coverage ratio. Figure 2 shows the coverage of the different divisions DMnT of the matching task for the naive (left) and neural embedding (right) strategies. The coverage ratio is very good, being 0:927 in the worst case (n = 200 in SNOMED-NCI) and 0:99 in the best case (n = 2 in FMA-NCI). This means that, in the worst case, almost 93% of the available reference mappings are covered by the matching subtasks in DMnT . The differences in terms of coverage between the naive and neural embedding strategies are minimal, with the neural embedding strategy providing slightly better results on average. These results reinforce Hypothesis 1 as the coverage with respect to system-generated mappings is expected to be even better. 6 StarSpace: https://github.com/facebookresearch/StarSpace 7 Note that in the context of neural embedding models the term entity refers to objects of different kind, e.g., a word, a sentence, a document or even an ontology entity. 8 Java codes: https://github.com/ernestojimenezruiz/logmap-matcher 9 Python codes: https://github.com/plumdeq/neuro-onto-part 10 Extended evaluation material in [ 1 ] and https://doi.org/10.5281/zenodo.1214149 Size ratio. The results in terms of the size (i.e., search space) of the selected divisions DMnT are presented in Figure 3 for the naive (left) and neural embedding (right) strategies. The results with the neural embedding strategy are extremely positive, while the results of the naive strategy, although slightly worse as expected, are surprisingly very competitive. Both strategies improve the search space with respect to the original MT for all cases with the exception of the naive strategy in the AMA-NCIA case with n < 50, and the SNOMED-NCI case with n > 20, which validates Hypothesis 3. SNOMED-NCI confirms to be the hardest case in the largebio track. Here the size ratio increases with the number of matching subtasks n and gets stable with n > 100. Size of the source and target modules. The scatter plots in Figures 4 and 5 visualize the size of the source modules against the size of the target modules for the n . For instance, the (orange) triangles represent pmoaitncthsinjgStiags(kOs 1i)nj;ejaScihg(dOiv2ii)sjionbeDinMg TO1i and O2i the source and target modules (with i i=1,. . . ,5) in the matching subtasks of DM5T . Figure 4 shows the plots for the AMANCIA case while Figure 5 for the FMA-NCI case, using the naive (left) and neural embedding (right) strategies. The naive strategy leads to rather balanced an similar tasks (note differentiated cloud of points) for each division DMnT for both cases. The neural embedding strategy has more variability in the size of the tasks within a given division DMnT . In the FMA-NCI case the tasks generated by the neural embedding strategy are also less balanced and the target module tends to be larger than the source mod0 500 Si1z0e0s0ource onto1l5o0g0y 2000 0 500 Si1z0e0s0ource onto1l5o0g0y 2000 (a) Naive strategy (b) Neural embedding strategy Fig. 4: Source and target module sizes in the computed subtasks for AMA-NCIA. ule. Nonetheless, on average, the (aggregated) size of the matching tasks in the neural embedding strategy are significantly reduced as shown in Figure 3.

Computation times. The time to compute the divisions of the matching task is tied to the number of locality modules to extract, which can be computed in polynomial time relative to the size of the input ontology [ 10 ]. The creation of LexI does not add an important overhead, while the training of the neural embedding model in the advance strategy ranges from 21s in AMA-NCI to 224s in SNOMED-NCI. Overall, for example, the required time to compute the division with 50 matching subtasks ranges from 2s in AMA-NCIA to 413s in SNOMED-NCI with the naive strategy, and from 24s (AMANCIA) to 647s (SNOMED-NCI) with the neural embedding strategy. 4.2

Evaluation of OAEI systems

In this section we support Hypothesis 2 by showing that the division of the alignment task enables systems that, given some computational constraints, were unable to complete an OAEI task. We have selected the following five systems from the latest OAEI campaigns: MAMBA [ 5 ], GMap [ 13 ], FCA-Map [ 6 ], KEPLER [ 14 ], and POMap [ 7 ]. MAMBA and GMap failed to complete the OAEI 2015 Anatomy track [ 2 ] with 8Gb of allocated memory, while FCA-Map, KEPLER and POMap could not complete the largest tasks in the largebio track within a 12 hours time-frame (with 16Gb of allocated

Comparison with OAEI results. There are baseline results in the OAEI for the selected systems [ 2 ], with the exception of MAMBA where the results are novel for the anatomy track. GMap, if 14Gb were allocated, was able to complete the anatomy task and obtained an f-measure of 0:861. KEPLER, POMap and FCA-Map completed the OAEI task involving small fragments of FMA-NCI (i.e., the overlapping matching task as in Definition 4) with an f-measure of 0:891, 0:861 and 0:935, respectively. The f-measure using the divisions of the matching task is slightly lower for GMap. The results are much lower for the cases of KEPLER, POMap and FCA-Map, but they cannot be fully comparable as systems typically reduce their performance when dealing with the whole largebio ontologies [ 2 ]. The authors of FCA-Map have also recently reported results for an improved version of FCA-Map [ 15 ]. They completed the FMA-NCI task in near 11 In a preliminary evaluation round a 4 hours time-frame was given, which was later extended. 7 hours, with a precision of 0:41, a recall of 0:87 and a f-measure of 0:56. The results obtained with DM20 T 50 are thus very positive, since both strategies lead to much better numbers in teramnsdoDf McoTmputation times and f-measure. 5

Related work

Partitioning has been widely used to reduce the complexity of the ontology alignment task. In the literature there are two major categories of partitioning techniques, namely: independent and dependent. Independent techniques typically use only the structure of the ontologies and are not concerned about the ontology alignment task when performing the partitioning. Whereas dependent partitioning methods rely on both the structure of the ontology and the ontology alignment task at hand. Although our approach does not compute (non-overlapping) partitions of the ontologies, it can be considered a dependent technique.

Prominent examples of ontology alignment systems including partitioning techniques are Falcon-AO [ 16 ], COMA++ [ 17 ] and TaxoMap [ 18 ]. COMA++ and FalconAO perform independent partitioning where the clusters of the source and target ontologies are independently extracted. Then pairs of similar clusters (i.e., matching subtasks) are aligned using standard techniques. TaxoMap [ 18 ] implements a dependent technique where the partitioning is combined with the matching process. TaxoMap proposes two methods, namely: PAP (partition, anchor, partition) and APP (anchor, partition, partition). The main difference of these methods is the order of extraction of (preliminary) anchors to discover pairs of partitions to be matched (i.e., matching subtasks).

The above approaches, although they present interesting results, did not provide any guarantees about the coverage (as in Definition 2) of the discovered partitions. In [ 19 ] we performed a preliminary study with the PBM method of Falcon-OA, and the PAP and APP methods of TaxoMap. The results in terms of coverage with the largebio tasks were very low, which directly affected the results of the evaluated systems. These rather negative results encouraged us to work on the approach presented in this paper.

Our dependent approach, unlike traditional partitioning methods, computes overlapping self-contained modules (i.e., locality modules). Locality modules guarantee the extraction of all semantically related entities for a given signature, which enhances the coverage results and enables the inclusion of the relevant information required by an alignment system. It is worth mentioning that the need of self-contained and covering modules was also highlighted in a preliminary work by Paulheim [ 20 ]. 6

Conclusions and future work

We have developed a novel framework to split the ontology alignment task into several matching subtasks based on a lexical index and locality modules. We have also presented two clustering strategies of the lexical index. One of them relies on a simple splitting method, while the other relies on a fast (log-linear) neural embedding model. We have performed a comprehensive evaluation of both strategies. The achieved high coverage (i.e., minimal information loss) in combination with the reduction of the search space and the small computation times suggests that the computed divisions based on LexI are suitable in practice. The division of the matching task allowed us to obtain results for five systems which failed to complete these OAEI matching tasks in the past.

Both the naive and the neural embedding strategies require the size of the number of matching subtasks or clusters as input. The (required) matching subtasks may be known before hand if, for example, the matching tasks are to be run in parallel in a number of available CPUs. For the cases where the resources are limited or where a matching system is known to cope with small ontologies, we plan to design an algorithm to estimate the number of clusters so that the size of the matching subtasks in the computed divisions is appropriate to the system and resource constraints.

As immediate future we plan to extend the conducted evaluation to better understand the impact of the division over different ontology alignment systems. We also aim at studying different notions of context tailored to the ontology alignment task. Acknowledgements. EJR was funded by the Centre for Scalable Data Access (SIRIUS), the RCN project BigMed, and The Alan Turing project AIDA.

References

1. Jimenez-Ruiz , E. , Agibetov , A. , Samwald , M. , Cross , V. : Breaking-down the Ontology Alignment Task with a Lexical Index and Neural Embeddings . arXiv ( 2018 ) Available from: https://arxiv.org/abs/ 1805 .12402.

2. (Ontology Alignment Evaluation Initiative) : http://oaei.ontologymatching.org/.

3. Ngo , D. , Bellahsene , Z. , Coletta , R.: YAM+ + results for OAEI 2011 . In: 6th International Workshop on Ontology Matching. ( 2011 )

4. Huber , J. , Sztyler , T. , No¨ßner, J., Meilicke , C. : CODI: combinatorial optimization for data integration: results for OAEI 2011 . In: 6th Int'l Workshop on Ontology Matching . ( 2011 )

5. Meilicke , C. : MAMBA - results for the OAEI 2015 . In: 10th International Workshop on Ontology Matching. ( 2015 ) 181 - 184

6. Zhao , M. , Zhang , S.: FCA-Map results for OAEI 2016 . In: 11th International Workshop on Ontology Matching. ( 2016 )

7. Laadhar , A. , Ghozzi , F. , Megdiche , I. , Ravat , F. , Teste , O. , Gargouri , F. : POMap results for OAEI 2017 . In: 12th International Workshop on Ontology Matching. ( 2017 ) 171 - 177

8. Jime ´nez- Ruiz , E. , Cuenca Grau , B. , Zhou , Y. , Horrocks , I. : Large-scale interactive ontology matching: Algorithms and implementation . In: European Conf. Artif. Intell. (ECAI) . ( 2012 )

9. Wu , L. , Fisch , A. , Chopra , S. , Adams , K. , Bordes , A. , Weston , J.: StarSpace: Embed All The Things! arXiv ( 2017 )

10. Cuenca Grau , B. , Horrocks , I. , Kazakov , Y. , Sattler , U. : Modular reuse of ontologies: Theory and practice . J. Artif. Intell. Res . 31 ( 2008 ) 273 - 318

11. Euzenat , J. , Shvaiko , P. : Ontology Matching, Second Edition . Springer ( 2013 )

12. Harrow , I. , Jime´nez- Ruiz , E. , et al.: Matching disease and phenotype ontologies in the ontology alignment evaluation initiative . J. Biomedical Semantics 8 ( 1 ) ( 2017 )

13. Li , W. , Sun , Q. : GMap: results for OAEI 2015 . In: 10th International Workshop on Ontology Matching. ( 2015 ) 150 - 157

14. Kachroudi , M. , Diallo , G. , Yahia , S.B. : OAEI 2017 results of KEPLER . In: 12th International Workshop on Ontology Matching. ( 2017 ) 138 - 145

15. Zhao , M. , Zhang , S. , Li , W. , Chen , G.: Matching biomedical ontologies based on formal concept analysis . J. Biomedical Semantics 9 ( 1 ) ( 2018 ) 11 : 1 - 11 : 27

16. Hu , W. , Qu , Y., Cheng, G.: Matching large ontologies: A divide-and-conquer approach . Data Knowl. Eng . 67 ( 2008 ) 140 - 160

17. Algergawy , A. , Massmann , S. , Rahm , E.: A clustering-based approach for large-scale ontology matching . In: ADBIS . ( 2011 ) 415 - 428

18. Hamdi , F. , Safar , B. , Reynaud , C. , Zargayouna , H.: Alignment-based partitioning of largescale ontologies . In: Advances in Knowledge Discovery and Management . ( 2009 ) 251 - 269

19. Pereira , S. , Cross , V. , Jime´nez- Ruiz , E.: On partitioning for ontology alignment . In: International Semantic Web Conference (Posters & Demonstrations) . ( 2017 )

20. Paulheim , H.: On Applying Matching Tools to Large-scale Ontologies . In: OM. ( 2008 )