Introduction

Many valuable tabular resources have been made available on the Internet and Open Data Portals, thanks to the Open Data movement. However, the usage of the tabular data is very limited in applications due to lacking or insufficient data descriptions, various data formats, vocabulary issues. Tabular data usually do not have a description, or the description does not cover data content. Tabular data also lack specification on table structure, and layout. Moreover, many tables do not use a standard vocabulary such as expressed in non-English, abbreviation, ambiguous or contain many misspellings, encoding problems. It is crucial to have a tabular data annotation system that could provide explicit information about table content to improve tabular data usability.

Previous studies addressed many tabular data annotation tasks such as structural annotations [6], [9] or semantic annotations as the participant systems in the Semantic Web Challenge on Tabular Data to Knowledge Graph Matching:

SemTab 2019 [3], and SemTab 2020 [4]. Unfortunately, most solutions or systems are not available to use, or require extensive configuration, setup, high computing power, or high time complexity [10].

This paper introduces MTab, a public service that generates structural and semantic annotations for tabular data. The structural annotations provide information about table headers, the table core attribute. The semantic annotations offer table elements matching knowledge graph concepts: cell-entity (CEA task), column-type (CTA task), and CPA task where the relation between core attribute to another column is annotated with a property. We also provide a graphical interface to visualize the annotation results.

The major advantages of MTab compared to other systems are as follows.

-Effectiveness: MTab tool is the best performance system in SemTab 2019 [5], [3] and SemTab 2020 [7], [4]. The key success of MTab is on the entity search modules with multilingual support (a keyword search with BM25 algorithm, a fuzzy search with edit distances, and an aggregation search with weighted fusion of keyword search and fuzzy search). The fuzzy search could support up to six edits (on the low-budget mac mini M1 2021), while most other systems only support two edits. As a result, MTab could address a higher level of noisiness compared to other systems. The entity search module achieves 87.98% on average of the top 1 accuracy (the top 1000 accuracy is 99.7%) [8] on Semtab 2020 [4] and Tough Tables datasets [1 MTab's repository, API documents, and other information could be accessed at https://github.com/phucty/mtab_tool; the demonstration video is available at https://youtu.be/0ibTWeObWaA.

MTab

Knowledge Graphs

We build a WikiGraph from the dump data of Wikidata, Wikipedia, and DBpedia as the target knowledge graph the annotation tasks. Wikidata is the central knowledge graph because it has the largest number of entities among the three graphs. With the dump data on 1 January 2021, we extracted 91.2 million entities and 249.3 million entity labels in multilingual, including entity labels, aliases, other names, redirect entity labels, and disambiguation entities. We also extracted 3.5 billion triples in WikiGraph. Additionally, WikiGraph will be updated frequently based on the future released dumps of knowledge graphs (Wikidata, Wikipedia, and DBpedia).

Entity Search Modules

Entity Search on a Cell We introduce the search modes1 as follows [8].

-Keyword search with BM25 algorithm: We use the hyper-parameters as b = 0.75, k 1 = 1.2. -Fuzzy search with edit distance: We use Damerau-Levenshtein distance as the edit distance for fuzzy search. We also perform candidate filtering and hashing with pre-calculating entity label deletes as the Symmetric Delete algorithm [2] to reduce the number of operations on pairwise edit distance calculation. Overall, MTab could support the fuzzy search up to six edits. -Aggregation search: This module is a weighted fusion of the keyword search and the fuzzy search results.

Statement Search on Two Cells

This module is built on the assumption that there is a logical relation between two cells of a table row, equivalent to a knowledge graph triple. We only keep the candidates of the two cells that have equivalent statements in the WikiGraph. We implement this statement search with a sparse matrix of 91 million entities and around 500 million edges.

Table Annotation: Use Case and Demo

MTab demonstration is available at https://mtab.app. Users could submit table files in various table formats, expressed in any language to MTab API, or copy data content and paste it to the interface. Then, users could tap to the "Annotate" button to get the annotation results. MTab will perform the following steps.

The annotation procedure2 are as the following steps: The photo on the right is the annotation results. The table header is in the first row, and the core attribute is in the first column. Entity annotations are in red and located below the table cell value. The type annotation is in green and located in the "Type" column. Finally, the relations between the core attribute and other columns are in blue and located in the property column.

Conclusions

This paper presents a demonstration of the MTab toolkit for table annotation with knowledge graphs of Wikidata, DBpedia, and Wikipedia. MTab is effective, efficient, and easy to use.

In the future work, we will focus on building downstream applications based on MTab's annotations such as question answering, and data analysis.