DIALOG GENERATION METHOD, DEVICE, ELECTRONIC EQUIPMENT, AND MEDIUM

• Main Authors: SUN SHUQI, JIAO ZHENYU, HUANG LEI, GUO HONGJIE, SUN KE, LI TINGTING
• Format: Patent
• Language: English; Japanese
• Published: 26.07.2021
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

To provide a dialog generation method, a device, electronic equipment, and a medium that can prevent deterioration of accuracy in question understanding due to a sample quantity.SOLUTION: A method includes: inputting an acquired question into a small shot learning model and a deep learning model to generate a first feature and a second feature; combining the first feature and the second feature to generate a feature sequence; and inputting the feature sequence into a fusion model to generate dialog information corresponding to question information.EFFECT: Generating dialog information corresponding to a question by connecting a small shot learning model, a deep learning model, and a fusion model enables the models to acquire a better effect both in the cases of small and large sample quantities, eliminates the need for selecting the small shot learning model or the deep learning model by setting a threshold, and prevents deterioration of accuracy in question understanding associated with an increase in sample quantity, thus improving model stability.SELECTED DRAWING: Figure 1 【課題】サンプル量により質問理解の精度が低下することを防ぐことが可能な対話生成方法、装置、電子機器及び媒体を提供する。【解決手段】方法は、取得した質問をそれぞれ少数ショット学習モデルと深層学習モデルに入力して第１特徴と第２特徴を生成し、第１特徴と第２特徴を組み合わせて特徴シーケンスを生成し、特徴シーケンスを融合モデルに入力して質問情報に対応する対話情報を生成することを含む。【効果】少数ショット学習モデル、深層学習モデル及び融合モデルを連携して質問に対応する対話情報を生成することで、モデルはサンプルが少ない場合とサンプルが多い場合の両方でより良い効果を取得し、閾値を設けて少数ショット学習モデルと深層学習モデルを選択する必要がなく、サンプル量の増加に伴って質問理解の精度が低下しにくく、モデルの安定性が向上する。【選択図】図１