NAMED ENTITY EXTRACTION DEVICE, NAMED ENTITY EXTRACTION METHOD, NAMED ENTITY EXTRACTION MODEL AND PROGRAM

To provide a named entity extraction device, a named entity extraction method, a named entity extraction model, and a program that realize highly accurate named entity extraction.SOLUTION: A named entity extraction device 100 includes: an input unit that accepts information to be extracted; a proces...

Full description

Saved in:
Bibliographic Details
Main Authors HIAI SATOSHI, NAGAYAMA SHOJI, KOJIMA JUNJI
Format Patent
LanguageEnglish
Japanese
Published 12.03.2024
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:To provide a named entity extraction device, a named entity extraction method, a named entity extraction model, and a program that realize highly accurate named entity extraction.SOLUTION: A named entity extraction device 100 includes: an input unit that accepts information to be extracted; a processing unit that inputs the information to be extracted into a named entity extraction model, and obtains a classification vector indicating a confidence level of a classification class of a named entity in the information to be extracted from the named entity extraction model; and an output unit that outputs classification information indicating the classification class of the named entity based on the classification vector. The named entity extraction model includes: a plurality of machine learning models that, upon accepting the information to be extracted, outputs a plurality of hidden vectors from one or more different intermediate layers of a specific machine learning model; and an ensemble layer that performs ensemble processing on output values based on the plurality of hidden vectors and outputs the classification vector.SELECTED DRAWING: Figure 4 【課題】高精度の固有表現抽出を実現する固有表現抽出装置、固有表現抽出方法、固有表現抽出モデル及びプログラムを提供する。【解決手段】固有表現抽出装置100は、抽出対象の情報を受け付ける入力部と、前記抽出対象の情報を固有表現抽出モデルに入力し、前記固有表現抽出モデルから前記抽出対象の情報における固有表現の分類クラスの確信度を示す分類ベクトルを取得する処理部と、前記分類ベクトルに基づいて、前記固有表現の分類クラスを示す分類情報を出力する出力部と、を有し、前記固有表現抽出モデルは、前記抽出対象の情報を受け付けると、特定の機械学習モデルの異なる1つ以上の中間層からの複数の隠れベクトルを出力する複数の機械学習モデルと、前記複数の隠れベクトルに基づく出力値に対してアンサンブル処理を実行し、前記分類ベクトルを出力するアンサンブル層と、を有する。【選択図】図4
Bibliography:Application Number: JP20220135736