Text error correction method and apparatus based on recurrent neural network of artificial intelligence

1. A text error correction method based on a recurrent neural network of artificial intelligence, comprising:

acquiring text data to be error-corrected by:

determining, using a language model, that a language model score corresponding to the input text data is less than a preset value; and

determining, using a preset classification model, that the input text data is the text data to be error-corrected; and

performing error correction on the text data to be error-corrected by using a trained recurrent neural network model so as to generate error-corrected text data, wherein the trained recurrent neural network model comprises an input layer having an input vector {x 1 , x 2 . . . x n }, an output layer having an output vector {o 1 , o 2 . . . o n }, and a hidden layer having a node of h t =f(Ux t +Wh t-1 ), where t is a positive integer, U represents a parameter weight matrix for connecting the input layer to the hidden layer, W represents a parameter weight matrix for connecting a node of the hidden layer to another, and f represents a non-linear activation function, wherein h t and h t-1 are each a state of a hidden layer node in a t th step and a (t−1) th step, respectively;

wherein before performing error correction on the text data to be error-corrected, the method further comprises:

acquiring a corpus of text pairs to be trained, each of the text pairs comprising an erroneous text and an accurate text corresponding to the erroneous text; and

training a preset recurrent neural network model by using the corpus of the text pairs so as to determine the trained recurrent neural network model;

wherein acquiring the corpus of text pairs to be trained comprises:

receiving N inputs in succession from a user within a preset time, wherein the N inputs comprise speech or text, and N≥2; and

determining identification results of an Nth input and an (N−1) th input of the N inputs as a first text pair of the corpus of text pairs, wherein the (N−1) th input is the erroneous text and the Nth input is the accurate text corresponding to the erroneous text.

2. The method of claim 1 , wherein training the preset recurrent neural network model by using the corpus of the text pairs comprises:

training a long short-term memory recurrent neural network model by using the corpus of the text pairs; or

training a bi-directional recurrent neural network model by using the corpus of the text pairs.

3. A text error correction apparatus based on a recurrent neural network of artificial intelligence, comprising:

one or more processors; and

a memory,

wherein one or more software modules are stored in the memory and executable by the one or more processors, and comprising:

a first acquiring module configured to acquire text data to be error-corrected by:

determining, using a language model, that a language model score corresponding to the input text data is less than a preset value; and

determining, using a preset classification model, that the input text data is the text data to be error-corrected; and

a processing module configured to perform error correction on the text data to be error-corrected by using a trained recurrent neural network model so as to generate error-corrected text data, wherein the trained recurrent neural network model comprises an input layer having an input vector {x 1 , x 2 . . . x n }, an output layer having an output vector {o 1 , o 2 . . . o n }, and a hidden layer having a node of h t =f(Ux t +Wh t-1 ), where t is a positive integer, U represents a parameter weight matrix for connecting the input layer to the hidden layer, W represents a parameter weight matrix for connecting a node of the hidden layer to another, and f represents a non-linear activation function, wherein h t and h t-1 are each a state of a hidden layer node in a t th step and a (t−1) th step, respectively;

wherein the one or more software modules further comprises:

a second acquiring module configured to acquire a corpus of text pairs to be trained, each of the text pairs comprising an erroneous text and an accurate text corresponding to the erroneous text; and

a determining module configured to train a preset recurrent neural network model by using the corpus of the text pairs so as to determine the trained recurrent neural network model;

wherein the second acquiring module is specifically configured to receive N inputs in succession from a user within a preset time, wherein the N inputs comprise speech or text, and N≥2, and determine identification results of an Nth input and an (N−1) th input of the N inputs as a first text pair of the corpus of text pairs, wherein the (N−1) th input is the erroneous text and the Nth input is the accurate text corresponding to the erroneous text.

4. The apparatus of claim 3 , wherein the determining module is specifically configured to:

train a long short-term memory recurrent neural network model by using the corpus of the text pairs; or

train a bi-directional recurrent neural network model by using the corpus of the text pairs.

5. A non-transitory computer-readable storage medium having stored therein computer programs that, when executed by a processor, causes the processor to perform a text error correction method based on a recurrent neural network of artificial intelligence, the method comprising:

acquiring text data to be error-corrected by:

determining, using a language model, that a language model score corresponding to the input text data is less than a preset value; and

determining, using a preset classification model, that the input text data is the text data to be error-corrected; and

performing error correction on the text data to be error-corrected by using a trained recurrent neural network model so as to generate error-corrected text data, wherein the trained recurrent neural network model comprises an input layer having an input vector {x 1 , x 2 . . . x n }, an output layer having an output vector {o 1 , o 2 . . . o n }, and a hidden layer having a node of h t =f(Ux t +Wh t-1 ), where t is a positive integer, U represents a parameter weight matrix for connecting the input layer to the hidden layer, W represents a parameter weight matrix for connecting a node of the hidden layer to another, and f represents a non-linear activation function, wherein h t and h t-1 are each a state of a hidden layer node in a t th step and a (t−1) th step, respectively;

wherein before performing error correction on the text data to be error-corrected, the method further comprises:

acquiring a corpus of text pairs to be trained, each of the text pairs comprising an erroneous text and an accurate text corresponding to the erroneous text; and

training a preset recurrent neural network model by using the corpus of the text pairs so as to determine the trained recurrent neural network model;

wherein acquiring the corpus of text pairs to be trained comprises:

receiving N inputs in succession from a user within a preset time, wherein the N inputs comprise speech or text, and N≥2; and

determining identification results of an Nth input and an (N−1) th input of the N inputs as a first text pair of the corpus of text pairs, wherein the (N−1) th input is the erroneous text and the Nth input is the accurate text corresponding to the erroneous text.

6. The non-transitory computer-readable storage medium of claim 5 , wherein training the preset recurrent neural network model by using the corpus of the text pairs comprises:

training a long short-term memory recurrent neural network model by using the corpus of the text pairs; or

training a bi-directional recurrent neural network model by using the corpus of the text pairs.