Form entity linking method based on multiple knowledge bases

Abstract

The invention discloses a form entity linking method based on multiple knowledge bases. The method is mainly used for solving the problem of entity linking in a form. The method includes the steps that firstly, a candidate entity is generated for a character string in each cell in a given form, and the candidate entities are extracted from the self-given knowledge bases; then a universal probability propagation algorithm based on a map is provided to rank the candidate entities corresponding to the character strings in the cells. The method can act on any single knowledge base. According to the ranking result of the candidate entities based on the different single knowledge bases, by means of the equivalence relation among the entities from the different knowledge bases, the ranked candidate entities, extracted from the different knowledge bases, corresponding to the character strings in the cells are divided. Finally, three heuristic rules are used for finally determining the entities, existing in the different knowledge bases, needing to be linked with the character strings in the cells, and therefore tasks of form entity linking based on the multiple knowledge bases are completed.

Description

technical field [0001] The invention belongs to the field of entity linking and relates to a table entity linking method based on multiple knowledge bases. Background technique [0002] There are a large number of HTML tables with high-quality relational data in the current World Wide Web, and these tables are regarded as important sources for knowledge extraction from the World Wide Web. In order to realize the vision of the Semantic World Wide Web, many works try to mine the latent semantic information in the tables, and represent the content in a given table as RDF triples. The first step in semantic information mining of table content is entity linking. Entity linking is to identify the true meaning of the strings in each cell in the table, and link these strings to entities in a given knowledge base. If the potential entities in the table cannot be correctly identified, it will be difficult to mine the correct RDF triples from the content of the given table, so linking...

AI Technical Summary

Problems solved by technology

However, there are two main problems in existing methods and systems for linking table entities: 1) Many methods or systems rely on features based on specific information, such as column headings and entity types in the knowledge base, but most of them are extracted from the World Wide Web There are no column headers in the tables, and many knowledge bases do not have semantic information such as entity types, which makes these methods and systems not universal and less practical; 2) All current methods and systems are based on a single knowledge base. Table entity links, but this does not guarantee the quality of table entity links. Many entities in tables do not exist in a single knowledge base, so it is unreasonable to link entities only for a single knowledge base

TabEL is more advanced than LIEGE, because TabEL can perform entity linking based on any single knowledge base for tables with multiple rows and columns, but the system still cannot complete the task of linking table entities based on multiple knowledge bases, because many strings should be Linked entities do not exist in a given single knowledge base, resulting in unsatisfactory quality of tabular entity linking using the TabEL system

In addition, the system relies on prior probabilities calculated from different sources, and each source has its own emphasis, resulting in unobjective prior probabilities obtained and easily affecting the quality of table entity links

Images