=Paper=
{{Paper
|id=Vol-1766/om2016_poster2
|storemode=property
|title=Annotating web tables through ontology matching
|pdfUrl=https://ceur-ws.org/Vol-1766/om2016_poster2.pdf
|volume=Vol-1766
|authors=Vasilis Efthymiou,Oktie Hassanzadeh,Mohammad Sadoghi,Mariano Rodriguez-Muro
|dblpUrl=https://dblp.org/rec/conf/semweb/EfthymiouHSR16
}}
==Annotating web tables through ontology matching==
https://ceur-ws.org/Vol-1766/om2016_poster2.pdf
Annotating Web Tables Through
Ontology Matching
Vasilis Efthymiou1,2 , Oktie Hassanzadeh1 , Mohammad Sadoghi1 , and
Mariano Rodriguez-Muro1
1 2
IBM T.J. Watson Research Center University of Crete, Greece
{vefthymi,hassanzadeh,msadoghi,mrodrig}@us.ibm.com
1 Introduction
Web tables have been proven to constitute valuable sources of information for
applications, ranging from Web search, to data discovery in spreadsheet software
and KB augmentation [1]. A requirement for those applications is to understand
the semantics of Web tables and potentially match their contents with existing
URIs in the Web of Data, a process known as Web table annotation [4].
Recent works on Web table annotation follow an iterative approach between
instance- and schema-level refinements, until convergence [6, 7]. In this work,
we annotate Web tables using ontology matching. As this field has solid tools
and benchmarks3 , we design a framework that provides the required input to
any ontology matching tool, resulting in Web table annotations. Moreover, our
blocking enables even the less scalable ontology matching tools provide annota-
tions to large-scale KBs, such as DBpedia. The contributions of our work are:
– We introduce a generic and scalable framework for Web table annotation
using existing ontology alignment systems.
– We evaluate our framework and compare the results against state-of-the-art
Web table annotation tools, with promising results.
– Our framework can be extended as a benchmark for ontology matching tools.
2 Matching Framework
Model. We assume that each table row describes a real-world entity, and each
column represents a property. Each cell of the header row defines the name of
a property, except the cell of the label column, which defines the name of the
table’s class. All the entities in the table are instances of this class. The values of
a column can be either literals, or references to other entities, corresponding to
dataype, or object properties, respectively. To make this distinction, we sample
the data types of each column, also identifying the label column, as in [6]. In
a second scan, we create a new instance of the table class for each row, whose
property values are the cell contents of this row for the respective column.
Blocking. To enable ontology matching tools that do not scale well be
applicable in this framework, and to improve the efficiency of matching tools that
do scale, we have applied a pre-processing step of candidate mappings selection,
known as blocking [2]. Specifically, we retain from DBpedia, the target ontology,
only those instances whose labels match with the labels of our table’s instances.
3
http://oaei.ontologymatching.org/
Finally, we call an ontology matching tool with the table ontology and the
DBpedia ontology after blocking, as input, and return the mapping results.
Evaluation. We evaluate our approach using the instance mappings of the
T2D gold standard4 and LogMap [5], one of the most efficient ontology matching
tools [3]. Our MapReduce-based framework annotates and evaluates the whole
corpus in less than 4 minutes. Table 1 presents the micro-averaged recall, pre-
cision, and F-measure results, against T2K [6] and two baselines: DBpedia
lookup. For each entity label in our table, we use top-1 DBpedia lookup5 re-
sult as annotation. DBpedia lookup refined. We keep the type of the top-1
lookup result for each cell in a first scan of the table, and then the top-5 most
frequent types for each column as acceptable types. Then, we perform a second
lookup, restricting the results to the acceptable types, and use the top-1 result
as the annotation.
Table 1. Results over T2D gold standard. Blocking results in parentheses.
Method Recall Precision F-measure
DBpedia lookup 0.73 0.79 0.76
DBpedia lookup refined 0.76 0.86 0.81
T2K 0.76 0.90 0.82
Ontology matching 0.57 (0.71) 0.89 (0.32) 0.70 (0.44)
The results show that our framework, using LogMap, suggests a good number
of correct results, with high precision. In the future, we plan to improve blocking
and extend our model to provide a first alignment, which can be utilized by many
ontology matching tools. Our goal is to provide an ontology matching benchmark
for instance-, class- and property-mappings, that can result in a new track in
the upcoming OAEI campaigns.
References
1. S. Balakrishnan, A. Y. Halevy, B. Harb, H. Lee, J. Madhavan, A. Rostamizadeh,
W. Shen, K. Wilder, F. Wu, and C. Yu. Applying WebTables in Practice. In CIDR,
2015.
2. V. Christophides, V. Efthymiou, and K. Stefanidis. Entity Resolution in the Web
of Data. Synthesis Lectures on the Semantic Web: Theory and Technology. Morgan
& Claypool Publishers, 2015.
3. E. Daskalaki, G. Flouris, I. Fundulaki, and T. Saveta. Instance matching bench-
marks in the era of linked data. Web Semantics: Science, Services and Agents on
the World Wide Web, 39:1–14, 2016.
4. O. Hassanzadeh, M. J. Ward, M. Rodriguez-Muro, and K. Srinivas. Understanding
a large corpus of web tables through matching with knowledge bases: an empirical
study. In ISWC, pages 25–34, 2015.
5. E. Jiménez-Ruiz and B. Cuenca Grau. Logmap: Logic-based and scalable ontology
matching. In ISWC, pages 273–288, 2011.
6. D. Ritze, O. Lehmberg, and C. Bizer. Matching HTML tables to dbpedia. In WIMS,
pages 10:1–10:6, 2015.
7. Z. Zhang. Towards efficient and effective semantic table interpretation. In ISWC,
pages 487–502, 2014.
4
http://webdatacommons.org/webtables/goldstandard.html
5
http://wiki.dbpedia.org/projects/dbpedia-lookup