We demonstrate Wikipedia Category Comprehensiveness Checker (WC3) based on the DBpedia metadata database. This system supports to check comprehensiveness and consistency of Wikipedia category annotation using DBpedia database. This system is available online.
Error (other retrieved articles). Precision, recall, and f-measures are calculated by Found/(Found+Error), Found/(Found+NotFound), and the harmonic mean of precision and recall, respectively.
The system selects the SPARQL query that has the highest f-measure among all candidates. The following is a candidate SPARQL query for \People from Tokyo"1.
SELECT DISTINCT ?s WHERE {?s rdf:type dbo:Person . ?s dbp:birthPlace dbr:Tokyo MINUS { ?s dbo:wikiPageRedirects ?o . }}
There are two problems with the previous WC3. The rst is computation time. Because SPARQL query generation requires generating many SPARQL candidates and evaluating their quality, it takes a long time (around 1 min) to obtain the nal result. The other is related to freshness. WC3 uses local DBpedia archives, so when the editors modify Wikipedia articles based on WC3's suggestions, WC3 cannot con rm the appropriateness of the editing results.
To solve these problems with the previous WC3, we implemented a newer version by adding the following two modules. This system is available online and the URL of the system is https://wnews.ist.hokudai.ac.jp/wc3/. It also has links to a demonstration movie and a detailed help page.
SPARQL query database SPARQL queries generated for Wikipedia categories are cached in the database, and the user can modify a query and store the modi ed query when it is better than an existing one. Retrieved results are also cached in the database to allow checking the comprehensiveness of the Wikipedia category annotation.
Access to DBpedia Live The system can send the same SPARQL query to DBpedia Live (http://live.dbpedia.org/) instead of the local DBpedia server to evaluate the appropriateness of the editing results.
The system has the following functions. 1. Retrieve SPARQL query database
When the user inputs the name of the Wikipedia category, the system returns SPARQL queries and pre-computed retrieved result are shown as a result. The user can modify the SPARQL query and compare the retrieved results of the original query and the new query. The user can also send the same SPARQL query to DBpedia Live to check the appropriateness of any edits conducted after the preparation of the current DBpedia archive. 2. Check the appropriateness of the Wikipedia category annotation for the page The system summarizes the results of the Wikipedia category analysis for the page. 3. SPARQL query construction
There are two SPARQL query construction methods in the system. One is the automatic SPARQL query construction method used by the previous WC3. The other is the construction of new SPARQL queries based 1 \dbo," and \dbp," are abbreviations for \http://dbpedia.org/ontology," \http://dbpedia.org/property," respectively on SPARQL queries for sibling categories. For example, SPARQL queries for \People from Nagoya" are constructed by using sibling categories \People from Tokyo.". In this case string \Tokyo" in the query are replaced by \Nagoya."
This system uses latest DBpedia database (2016-10 version). Target categories for WC3 are all Wikipedia categories with the following conditions: 1) subcategories of \Categories by parameter" with set-and-topic-style excluding stubs; 2) at least one article directly belongs to the category. There are 655,937 target categories in this version of DBpedia. In this experiment, we selected target categories that have at least 10 articles that directly belong to the category for initial database construction (231,148 categories).
Figure 1 shows a screen-shot of the system. First, the user enters a category name using the \load" button. Then, information about a stored SPARQL query is loaded if one exists. If no stored information is available, the user can generate a query using the automatic SPARQL query construction method of WC3 or by using information from sibling categories. To highlight the quality of the SPARQL query, when the recall and precision are larger than 0.7 or lower than 0.3, the corresponding elements are highlighted in green or red, respectively. !"#$%&'()%*+,-.'")/*' :1"B3'%,'56789:';$*-.'1"0,-/)%1,"' )2,$%'312<1"+'C)%*+,-1*3' !"0,-/)%1,"')2,$%'3%,-*4'56789:';$*-.' 0,-',-1+1")<'4)%)2)3*'=",%'0,-'>?#*41)'<1@*A'
D*E'56789:';$*-.' 0,-'C,/#)-13,"' 8*<)%*4'F1B1#*41)'#)+*3')-*'
C)%*+,-1G*4'1"%,'H'C)%*+,-1*3'
Fig. 1. Screen-shot of WC3 interface
The user can check information about a page in Wikipedia or DBpedia page information stored in the database by clicking the corresponding links. When the user adds a check-mark to the \Check Candidate Categories for Errors" box, a list of candidate categories is shown as additional columns (Figure 2). For example, candidate categories for \Adam Jezierski" and \Adil Ibragimov" are \1990 births" and \1989 births"2. In both cases, the suggested candidate categories seem to be appropriate for those pages according to the information in the text and the infobox information for the page.
There are many errors for birth year
Wikipedia category annotation The user can also check the appropriateness of the edits conducted after the most recent DBpedia database construction by checking the \Compare with DBpedia Live" box and clicking on the \Compare" button. Comparison results are shown with the same categories for the \load" case. \+" and \{" show pages that can be categorized by using DBpedia Live only or by the original database, respectively. Results from DBpedia Live are also stored for checking the comprehensiveness of the Wikipedia page edits.
In this paper, we have introduced a new WC3 system that uses the SPARQL query database and DBpedia Live. This system solves the problems of the previous WC3 (computation time and freshness of the database). This system is available online and supports volunteer editors in maintaining Wikipedia categories. Updating Wikipedia categories based on this framework is also bene cial for knowledge engineers who would like to utilize Wikipedia as a knowledge resource.
This work was partially supported by JSPS KAKENHI Grant Number 18H03338.