Knowledge graph entity concept description generation system

Abstract

The invention provides a knowledge graph entity concept description generation system, which is used for generating a concept description text of an entity according to an information box which corresponds to the entity in a knowledge graph and contains attributes and values, and is characterized by comprising a word template generation module used for storing afirst sequence-to-sequence model which is alreadlypre-trained; the first sequence-to-sequence model comprises an information frame encoder and a template decoder, the information frame encoder is used for reconstructing an information frame corresponding to a to-be-processed entity into a word sequence and encoding the word sequence into a first hidden state, and the template decoder is used for outputting a template sequence according to the first hidden state; a concept description text generation module stores a pre-trained second sequence-to-sequence model, the second sequence-to-sequence model comprises a template encoder and a concept description decoder, the template encoder is used for encoding a template sequence into a second hidden state, and the concept description decoder is used for decoding the template sequence into a second hidden state; a concept description decoder is used for outputting a concept description text according to the first hidden state and the second hidden state.

Description

technical field [0001] The invention belongs to the field of natural language generation, and in particular relates to a form-to-text generation system guided by a priori grammar template. Background technique [0002] With the rise of large-scale open cross-domain knowledge graphs, knowledge graph technology has attracted more and more attention from researchers in academia and industry. However, although such an open domain knowledge graph has very rich structured information, the entities in it often lack a concise conceptual description text. In the knowledge graph, the conceptual description of an entity is a class of noun phrases that can reflect entity classification information. Concept description has a wide range of application scenarios, including question answering (for example: "Q: Who is "Jay Chou"? Answer: Musicians from Taiwan, China"), named entity disambiguation (for example: "Apple (fruit) and Apple (technology company)" ), information retrieval (for exa...

AI Technical Summary

Problems solved by technology

Although these types describe the ontology types of entities, their coarse-grainedness limits their use in some applications; generation-based methods cannot distinguish the modifiers in the concept description from the head words during the generation process. , so the output tends to lose the grammatical structure required for conceptual description, resulting in the generated text often having grammatical or factual errors

[0005] Therefore, the above prior art often cannot guarantee the following effects when generating concept descriptions:

[0006] 1) The concept description must be grammatically correct, because a small mistake can lead to a serious grammatical error, such as street with Paris, France is obviously wrong;

Images