Utilization of DBpedia mappings in cross language wikipedia infobox completion

Abstract

Wikipedia plays an important role in the web as one of the biggest knowledge source due to its large coverage of information that came from various domains. As for today, Wikipedia covers articles from 282 different languages with more than 5 million articles and the number keep expanding. Each language version of Wikipedia covers different range of articles completeness and is maintained independently by the community. Consequently, the problem of missing information among cross-language Wikipedia articles has emerged. Infobox is a small box, which is located inside a Wikipedia page and contains summary of the topic in the semi-structured manner. Since, infoboxes are often useful for Wikipedia data extraction, it is important to maintain their information quality as well. Several studies have been done in alignment and generation of new entries for Wikipedia infoboxes. [5] developed an information extractor that extracts all possible infobox attribute-value pairs from Wikipedia text by using CRF and generate new infoboxes from the result. A different approach was used by [4] who built a binary classifier to predict the similarity of cross language attribute pairs to align two infoboxes. Other infobox alignment approach was discussed in [15] and [16]. We proposed an approach to fix information gap in tbetween cross language Wikipedia articles by utilizing the existing DBpeda mappings. Our goal was to add new information from the infoboxes of Korean Wikipedia articles to their corresponding English Wikipedia articles. To determine attribute-value pairs that we should generate, we tried to find two attributes which are likely to have similar meaning by looking at their mapped DBpedia property. In addition, we also used instance-based attribute alignment method [2] to expand our aligned attribute list. The results showed that we could expand up to 38% of the existing Wikipedia attribute-value pairs from our datasets with 61% of accuracy as well as automatically creating new Wikipedia-DBpedia mappings.