Sentence-meaning recognition method, sentence-meaning recognition device, sentence-meaning recognition apparatus and storage medium

1. A sentence-meaning recognition method, comprising:

performing a character-based similarity calculation on a first sentence to be recognized and a second sentence obtained based on the first sentence to obtain a first similarity and a first sentence-meaning recognition result;

determining whether the first similarity triggers an enhanced recognition according to an enhanced recognition condition;

outputting the first sentence-meaning recognition result if the enhanced recognition condition is not satisfied, and if the enhanced recognition condition is satisfied, performing a Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence to obtain a second similarity and a second sentence-meaning recognition result; and

generating and outputting a third sentence-meaning recognition result according to the first sentence-meaning recognition result and the second sentence-meaning recognition result,

wherein the step of performing Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence, to obtain the second similarity and the second sentence-meaning recognition result comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing a vectorization processing on the obtained Chinese-phonetic-alphabets, respectively, and performing feature extraction on the vectorized Chinese-phonetic-alphabets by using a second deep learning model to obtain a third extraction result and a fourth extraction result;

performing a similarity calculation on the third extraction result and the fourth extraction result to obtain the second similarity; and

generating the second sentence-meaning recognition result according to the second similarity, and

wherein the step of converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing the vectorization processing on the obtained Chinese-phonetic-alphabets comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets;

dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of the Chinese-phonetic-alphabet of each character, and dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of a window with a fixed length; and

vectorizing the Chinese-phonetic-alphabets of each group.

2. The method of claim 1 , wherein the enhanced recognition condition comprises the first similarity being less than a first threshold and greater than a second threshold;

wherein when the first similarity is not greater than the second threshold, the first sentence-meaning recognition result is null.

3. The method of claim 1 , wherein the step of performing the character-based similarity calculation on the first sentence to be recognized and the second sentence obtained based on the first sentence to obtain the first similarity and the first sentence-meaning recognition result comprises:

performing a character-vectorization processing on the first sentence and the second sentence respectively, and extracting features from the vectorized first sentence and the vectorized second sentence by using a first deep learning model to obtain a first extraction result and a second extraction result;

performing a similarity calculation on the first extraction result and the second extraction result to obtain the first similarity; and

generating the first sentence-meaning recognition result according to the first similarity.

4. The method of claim 3 , wherein the step of performing the similarity calculation on the first extraction result and the second extraction result to obtain the first similarity comprises:

performing a similarity calculation on each of second extraction results extracted from multiple second sentences and the first extraction result respectively, to obtain a plurality of candidate similarities; and

determining the highest candidate similarity among the candidate similarities as the first similarity.

5. A sentence-meaning recognition method, comprising:

performing a character-based similarity calculation on a first sentence to be recognized and a second sentence obtained based on the first sentence to obtain a first similarity and a first sentence-meaning recognition result;

determining whether the first similarity triggers an enhanced recognition according to an enhanced recognition condition;

outputting the first sentence-meaning recognition result if the enhanced recognition condition is not satisfied, and if the enhanced recognition condition is satisfied, performing a Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence to obtain a second similarity and a second sentence-meaning recognition result; and

generating and outputting a third sentence-meaning recognition result according to the first sentence-meaning recognition result and the second sentence-meaning recognition result wherein

the step of generating and outputting the third sentence-meaning recognition result according to the first sentence-meaning recognition result and the second sentence-meaning recognition result comprises:

determining whether the first sentence-meaning recognition result and the second sentence meaning recognition result are the same;

wherein when the first sentence-meaning recognition result and the second sentence meaning recognition result are determined to not be the same, the third sentence-meaning recognition result is configured to be null and is output; and

wherein when the first sentence-meaning recognition result and the second sentence meaning recognition result are determined to be the same, the first sentence-meaning recognition result is output as the third sentence meaning recognition result.

6. The method of claim 1 , further comprising:

performing a voice recognition on voice information to obtain the first sentence.

7. The method of claim 6 , further comprises:

before the step of performing the voice recognition on voice information to obtain the first sentence, collecting the voice information.

8. The method of claim 1 , wherein the second sentence is obtained from a segmentation domain library based on the first sentence.

9. A sentence-meaning recognition apparatus, comprising:

a first recognition device that performs a character-based similarity calculation on a first sentence to be recognized and a second sentence obtained based on the first sentence to obtain a first similarity and a first sentence-meaning recognition result;

a determining device that determines whether the first similarity triggers an enhanced recognition according to an enhanced recognition condition;

a first output device that outputs the first sentence-meaning recognition result when the determination result is that the enhanced recognition is not triggered;

a second recognition device that, when the determination result is that the enhanced recognition is triggered, performs a Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence to obtain a second similarity and a second sentence-meaning recognition

result; and

a second output device that generates and output a third sentence-meaning recognition result according to the first sentence-meaning recognition result and the second sentence-meaning recognition result,

wherein the step of performing Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence, to obtain the second similarity and the second sentence-meaning recognition result comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing a vectorization processing on the obtained Chinese-phonetic-alphabets, respectively, and performing feature extraction on the vectorized Chinese-phonetic-alphabets by using a second deep learning model to obtain a third extraction result and a fourth extraction result;

performing a similarity calculation on the third extraction result and the fourth extraction result to obtain the second similarity; and

generating the second sentence-meaning recognition result according to the second similarity, and

wherein the step of converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing the vectorization processing on the obtained Chinese-phonetic-alphabets comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets;

dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of the Chinese-phonetic-alphabet of each character, and dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of a window with a fixed length; and

vectorizing the Chinese-phonetic-alphabets of each group.

10. The sentence-meaning recognition apparatus of claim 9 , wherein the second recognition device comprises:

an extraction element that converts the first sentence and the second sentence into Chinese-phonetic-alphabets and perform a vectorization processing on the obtained Chinese-phonetic-alphabets, respectively, and perform feature extraction on the vectorized Chinese-phonetic-alphabets by using a second deep learning model to obtain a third extraction result and a fourth extraction result;

a calculation element performs a similarity calculation on the third extraction result and the fourth extraction result to obtain the second similarity;

a generation element that generates the second sentence-meaning recognition result according to the second similarity.

11. The sentence-meaning recognition apparatus of claim 10 , wherein the extraction element further:

converts the first sentence and the second sentence into Chinese-phonetic-alphabets;

divide the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of the Chinese-phonetic-alphabet of each character, and divides the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of a window with a fixed length; and

vectorizes the Chinese-phonetic-alphabets of each group.

12. The sentence-meaning recognition apparatus of claim 9 , wherein the enhanced recognition condition comprises the first similarity being less than a first threshold and greater than a second threshold;

wherein when the first similarity is not greater than the second threshold, the first sentence-meaning recognition result is null.

13. The sentence-meaning recognition apparatus of claim 9 , further comprising:

a voice recognition device configured to perform a voice recognition on voice information to obtain the first sentence.

14. The sentence-meaning recognition apparatus of claim 12 , further comprising:

a voice recognition device that performs a voice recognition on voice information to obtain the first sentence.

15. The sentence-meaning recognition apparatus of claim 13 , further comprising: a collection device that collects the voice information.

16. The sentence-meaning recognition apparatus of claim 9 , further comprising:

a storage device that stores a segmentation domain library, wherein the second sentence is obtained from the segmentation domain library based on the first sentence.

17. A sentence-meaning recognition equipment, comprising:

one or more processors;

a memory that stores one or more programs,

wherein the one or more processors perform the method of claim 1 when the one or more programs are executed by the one or more processors.

18. A non-transitory storage medium storing a computer program, wherein the computer program is executed to implement the method of A sentence-meaning recognition method, comprising:

performing a character-based similarity calculation on a first sentence to be recognized and a second sentence obtained based on the first sentence to obtain a first similarity and a first sentence-meaning recognition result;

determining whether the first similarity triggers an enhanced recognition according to an enhanced recognition condition;

outputting the first sentence-meaning recognition result if the enhanced recognition condition is not satisfied, and if the enhanced recognition condition is satisfied, performing a Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence to obtain a second similarity and a second sentence-meaning recognition result; and

generating and outputting a third sentence-meaning recognition result according to the first sentence-meaning recognition result and the second sentence-meaning recognition result,

wherein the step of performing Chinese-phonetic-alphabet-based similarity calculation on the first sentence and the second sentence, to obtain the second similarity and the second sentence-meaning recognition result comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing a vectorization processing on the obtained Chinese-phonetic-alphabets, respectively, and performing feature extraction on the vectorized Chinese-phonetic-alphabets by using a second deep learning model to obtain a third extraction result and a fourth extraction result;

performing a similarity calculation on the third extraction result and the fourth extraction result to obtain the second similarity; and

generating the second sentence-meaning recognition result according to the second similarity, and

wherein the step of converting the first sentence and the second sentence into Chinese-phonetic-alphabets and performing the vectorization processing on the obtained Chinese-phonetic-alphabets comprises:

converting the first sentence and the second sentence into Chinese-phonetic-alphabets;

dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of the Chinese-phonetic-alphabet of each character, and dividing the Chinese-phonetic-alphabets of the first sentence and the second sentence into groups on the basis of a window with a fixed length; and

vectorizing the Chinese-phonetic-alphabets of each group.

19. The method of claim 5 , wherein the enhanced recognition condition comprises the first similarity being less than a first threshold and greater than a second threshold;

wherein when the first similarity is not greater than the second threshold, the first sentence-meaning recognition result is null.

20. The method of claim 5 , wherein the step of performing the character-based similarity calculation on the first sentence to be recognized and the second sentence obtained based on the first sentence to obtain the first similarity and the first sentence-meaning recognition result comprises:

performing a character-vectorization processing on the first sentence and the second sentence respectively, and extracting features from the vectorized first sentence and the vectorized second sentence by using a first deep learning model to obtain a first extraction result and a second extraction result;

performing a similarity calculation on the first extraction result and the second extraction result to obtain the first similarity; and

generating the first sentence-meaning recognition result according to the first similarity.

21. The method of claim 20 , wherein the step of performing the similarity calculation on the first extraction result and the second extraction result to obtain the first similarity comprises:

performing a similarity calculation on each of second extraction results extracted from multiple second sentences and the first extraction result respectively, to obtain a plurality of candidate similarities; and

determining the highest candidate similarity among the candidate similarities as the first similarity.

22. The method of claim 5 , further comprising:

performing a voice recognition on voice information to obtain the first sentence.

23. The method of claim 22 , further comprises:

before the step of performing the voice recognition on voice information to obtain the first sentence, collecting the voice information.

24. The method of claim 5 , wherein the second sentence is obtained from a segmentation domain library based on the first sentence.