MTab: Matching Tabular Data to Knowledge Graph with Probability Models Phuc Nguyen 1 2 Natthawut Kertkeidkachorn 0 Ryutaro Ichise 0 1 2 Hideaki Takeda 1 2 National Institute of Advanced Industrial Science and Technology , Japan National Institute of Informatics , Japan SOKENDAI (The Graduate University for Advanced Studies) , Japan

This paper presents the design of our system, namely MTab, for Semantic Web Challenge on Tabular Data to Knowledge Graph Matching (SemTab 2019). MTab combines the voting algorithm and the probability model to solve critical bottlenecks of the matching task. Results on SemTab 2019 show MTab obtains the promising performance. Tabular Data to Knowledge Graph Matching (SemTab 2019) 4 is a challenge on matching semantic tags from table elements to knowledge bases (KBs), especially DBpedia. Fig. 1 depicts the three sub-tasks for SemTab 2019. Given a table data, CTA (Fig. 1a) is the task of assigning a semantic type (e.g., a DBpedia class) to a column. In CEA (Fig. 1b), a cell is linked to an entity in KB. The relation between two columns is assigned to a property in KB in CPA (Fig. 1c).

 Introduction

To address the three tasks of the challenge, we designed our system (MTab) by the 4-steps pipeline as shown in Fig. 2.

Step 1 is to pre-process a table data by predicting languages of the table with fasttext [ 1 ], correcting spelling, predicting data types (e.g., number or text), and searching relevant entities in DBpedia. Due to the heterogeneous problem, we utilize entity searching on many services including DBpedia Lookup, DBpedia 4 http://www.cs.ox.ac.uk/isg/challenges/sem-tab/

Copyright 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

Phuc Nguyen et al. endpoint. Also, we search relevant entities on Wikipedia and Wikidata by redirected links to DBpedia to increase the possibility of nding the relevant entities. We assume that cells in a column have the same type. We then use information from Step 1 to estimate the probability of the types for the column in Step 2. The type candidate which has the highest probability is the result for CTA task. In Step 3, the result of Step 2 and information of Step 1 are used to estimate the probability for entities. Similarly, the entity candidate which has the highest probability is the result for CEA task. In Step 4, we use the result from Step 3 to estimate the property between two entities, and then, adopt the voting technique to estimate the probability for all rows of two columns. The result for CPA is the highest probability of property candidate in Step 4. We repeatedly execute Step 2, 3 and 4 to nd the best candidates for columns, cells, and the relation between two columns. 3

 Results and Conclusion 1. Joulin , A. , Grave , E. , Bojanowski , P. , Mikolov , T. : Bag of tricks for e cient text classi cation . In: EACL 2017 . pp. 427 { 431 . ACL (April 2017 )