Question generating method and apparatus, inquiring diagnosis system, and computer readable storage medium
AssigneeBoe Technology Group Co., Ltd.
InventorsZhenzhong Zhang, Xue Chen
Life of this patent
Estimating the term…
Abstract
The present disclosure relates to a question generating method and apparatus, an inquiring diagnosis system, and a computer readable storage medium. The question generating apparatus includes at least one processor, the at least one processor being configured to: acquire a candidate question set Q; calculate an information value of each candidate question in the candidate question set Q; and generate at least one question according to the information value of each candidate question.
Claims18
A question generating apparatus, comprising at least one processor, the at least one processor being configured to:
- acquire a candidate question set Q according to a dialog context, wherein the candidate question set Q comprises a first selected number of candidate questions qi, of which correlation degree with the dialog context satisfies a threshold, where i is a positive integer less than or equal to the first selected number;
- acquire an answer aj corresponding to a candidate question in the candidate question set Q to obtain an answer set A, where j is a positive integer less than or equal to the first selected number, and j=i indicates that the answer a corresponds to the candidate question qi;
- calculate an information value of each candidate question in the candidate question set Q, comprising:
- obtaining a predicted answer to the candidate question qi by using a first recurrent neural network;
- calculating a probability of generating an answer aj for the dialog context and the candidate question qi according to a similarity between the candidate questions qi and each of the other candidate questions;
- calculating an information amount brought by each candidate question qi by using a second recurrent neural network according to the probability of generating the answer aj for the dialog context and the information amount brought by each candidate question qi;
- and
- generate at least one question according to the information value of each candidate question,
- wherein the at least one processor is further configured to:
- based on a corpus and a loss function, perform training by using a stochastic gradient descent method to form the first recurrent neural network and the second recurrent neural network, and wherein the corpus comprises a dialog context for training Tcontext, a candidate question set for training TQ, and an answer set for training TA.
A question generating method, comprising:
- acquiring a candidate question set Q according to a dialog context, wherein the candidate question set Q comprises a first selected number of candidate questions qi, of which correlation degree with the dialog context satisfies a threshold, where i is a positive integer less than or equal to the first selected number;
- acquiring an answer aj corresponding to a candidate question in the candidate question set Q to obtain an answer set A, where j is a positive integer less than or equal to the first selected number, and j=i indicates that the answer aj corresponds to the candidate question qi;
- calculating an information value of each candidate question in the candidate question set Q, comprising:
- obtaining a predicted answer to the candidate question qi by using a first recurrent neural network;
- calculating a probability of generating an answer aj for the dialog context and the candidate question qi according to a similarity between the candidate questions qi and each of the other candidate questions;
- calculating an information amount brought by each candidate question qi by using a second recurrent neural network according to the probability of generating the answer aj for the dialog context and the information amount brought by each candidate question qi;
- and
- generating at least one question based on the information value of each candidate question,
- wherein the at least one processor is further configured to:
- based on a corpus and a loss function, perform training by using a stochastic gradient descent method to form the first recurrent neural network and the second recurrent neural network, and wherein the corpus comprises a dialog context for training Tcontext, a candidate question set for training TQ, and an answer set for training TA.
Dependencies are inferred from the claim text.