An entity linking method based on deep learning

Abstract

The invention discloses an entity linking method based on deep learning, which comprises the following steps: obtaining entity reference to be linked in a document to be identified, and extracting a candidate entity set corresponding to the entity reference in a library; calculating the conceptual similarity between each candidate entity and entity reference, and extracting the candidate entity with the highest conceptual similarity to entity reference from the candidate entity set as the entity to be linked; obtaining attribute information of the entity to be linked, judging whether the entity to be linked can be linked with the corresponding entity reference, and linking the entity to be linked that can be linked with the entity reference. The method can judge whether the candidate entity can be linked with the entity reference by combining the attribute information of the candidate entity and the concept similarity, the conceptual similarity between candidate entity and entity reference is determined, whether the candidate entity links the two processes of entity reference to carry out joint modeling, and whether the candidate entity can link with entity reference automaticallyin the process of model training, so as to make the judgment more accurate.

Description

technical field [0001] The invention relates to the field of databases, in particular to an entity linking method based on deep learning. Background technique [0002] With the rapid development of the Internet, digital resources can be seen everywhere, and the information carrier with the highest contact frequency is text information, such as news, blogs, comments, etc. At the same time, with the acceleration of the pace of life, users have a stronger demand for efficient reading. Digital resources contain a large number of text entities with clear semantic information. How to efficiently obtain entities in texts and make use of them is of practical significance. . In particular, entity linking is one of the most critical steps in the process of utilizing entities. For example, by analyzing the digital resources browsed or shared by users to extract entities and link them to the knowledge base, and use the linked entities as keywords or tags to model users' interests more...

AI Technical Summary

Problems solved by technology

For example, the traditional entity linking method only considers the conceptual similarity between the entity reference and the candidate entity when screening candidate entities, but if the entity reference background information given by the text to be recognized is insufficient, it cannot guarantee that the linked candidate entity is the correct entity ; When all candidate entities are not entities in the text to be recognized, the candidate entity with the highest conceptual similarity will still be linked to the entity of the document to be recognized

At the same time, in the judgment of concept similarity, keywords are generally only extracted through the topic model, and information on the semantic level is not obtained. If the keyword in the document to be recognized is the same word as the entity document, the correct overlap cannot be performed. degree; the topic model only considers the high-level semantic features at the entity topic level, but does not consider the low-level fine-grained word-level features, and cannot finely distinguish candidate entities with similar backgrounds

Images