Family atlas automatic construction method based on multi-task joint neural network model

Abstract

The invention discloses a family atlas automatic construction method based on a multi-task joint neural network model. The method comprises the following steps: firstly, establishing the multi-task joint neural network model; training an end-to-end joint neural network model; optimizing the end-to-end joint neural network model to obtain a final model; and finally, constructing a family atlas through the final model. According to the multi-task joint model disclosed by the invention, relationship classification between every two entities is avoided, and the redundancy of model output is greatly reduced. The joint extraction method provided by the invention can also be applied to other data fields, and can reflect good time efficiency and accuracy on an information extraction task. According to the method, the performance of entity extraction and relationship classification tasks is improved from two aspects of efficiency and precision, and the method can be flexibly expanded to other tasks needing entity extraction except for family atlas construction tasks and performing relationship classification.

Description

technical field [0001] The invention belongs to the field of natural language processing, relates to a method for joint extraction of entity relations based on a multi-task deep neural network, and relates to a method for automatically constructing family graphs based on a multi-task joint neural network model, and is mainly applied to families with obituaries as data sources The map is automatically constructed. Background technique [0002] Family map, also known as genealogy, genealogy, etc., is a form of genealogy, a kind of precious humanistic data, which records the figures and deeds of the same ancestral blood group, and is useful for historical folk customs, public security information, and society. In-depth research in fields such as economics and genetic analysis has its unique and irreplaceable functions. Traditional genealogists need to collect, analyze, and organize information distributed in various documents in order to assemble a relatively complete family t...

AI Technical Summary

Problems solved by technology

However, the idea of ​​extracting entities and relationships separately has the following three disadvantages: First, because the relationship classification task requires the output information of entity recognition, errors generated when identifying entities will affect the subsequent relationship classification results, and ultimately affect the accuracy of the overall result sex

Secondly, the traditional method uses two independent models to process entity extraction and relationship classification, ignoring the correlation between each subtask, resulting in the blindness of the entity extraction model in judging, unable to use the information of the relationship type between entities

Finally, the pipelined multi-task method needs to compare all the extracted entities when classifying the relationship. Since there is no relationship between any two entities, and the relationship between entities is sparse in most cases, so The traditional pipelined multi-tasking method will generate a large number of redundant relationship type judgments

In summary, the traditional extraction scheme has obvious deficiencies, and the extraction method needs to be improved urgently

Images