Error correction method and apparatus, and computer readable medium

1. An error correction method, implemented by a computer, comprising:

determining a plurality of target candidate entities from a preset dictionary tree based on a query request;

for each target candidate entity, obtaining a first probability that the target candidate entity is a legitimate entity by calculating the target candidate entity based on a language model;

evaluating, based on an evaluation model, each target candidate entity to obtain an evaluation result, a target candidate entity corresponding to an evaluation result; and

determining a real intent entity corresponding to the query request based on the first probability and the evaluation result;

wherein evaluating each target candidate entity comprises:

extracting attribution information corresponding to respective target candidate entities;

determining weights for the attribution information corresponding to respective target candidate entities; and

evaluating each target candidate entity based on the weight;

wherein determining the real intent entity corresponding to the query request based on the first probability and the evaluation result comprises:

for each target candidate entity, weighting the first probability and the evaluation result corresponding to the target candidate entity, to obtain a weighted value corresponding to the target candidate entity; and

determining a target candidate entity corresponding to a maximum weighted value as the real intent entity.

2. The error correction method of claim 1 , further comprising:

calculating a second probability that the query request is inputted correctly;

wherein determining the plurality of target candidate entities from the preset dictionary tree based on the query request comprises:

determining the plurality of target candidate entities from the preset dictionary tree based on the query request in response to that the second probability is smaller than a first threshold.

3. The error correction method of claim 1 , further comprising:

extracting a plurality of entities from a preset knowledge base; and

taking each entity as a node of the preset dictionary tree, taking an entity item of the entity as a child node of the node, and assigning a unique identifier to the entity item, to obtain the preset dictionary tree.

4. An error correction apparatus, comprising:

a processor;

a memory, having computer programs executable by the processor;

wherein when the computer programs are executed by the processor, the processor is caused to perform followings:

determining a plurality of target candidate entities from a preset dictionary tree based on a query request;

for each target candidate entity, obtaining a first probability that the target candidate entity is a legitimate entity by calculating the target candidate entity based on a language model;

evaluating, based on an evaluation model, each target candidate entity to obtain an evaluation result, a target candidate entity corresponding to an evaluation result; and

determining a real intent entity corresponding to the query request based on the first probability and the evaluation result;

wherein evaluating each target candidate entity comprises:

extracting attribution information corresponding to respective target candidate entities;

determining weights for the attribution information corresponding to respective target candidate entities; and

evaluating each target candidate entity based on the weight;

wherein determining the real intent entity corresponding to the query request based on the first probability and the evaluation result comprises:

for each target candidate entity, weighting the first probability and the evaluation result corresponding to the target candidate entity, to obtain a weighted value corresponding to the target candidate entity; and

determining a target candidate entity corresponding to a maximum weighted value as the real intent entity.

5. The error correction apparatus of claim 4 , wherein the processor is caused to further perform calculating a second probability that the query request is inputted correctly;

wherein determining the plurality of target candidate entities from the preset dictionary tree based on the query request comprises: determining the plurality of target candidate entities from the preset dictionary tree based on the query request in response to that the second probability is smaller than a first threshold.

6. The error correction apparatus of claim 4 , wherein the processor is caused to further perform:

extracting a plurality of entities from a preset knowledge base; and

taking each entity as a node of the preset dictionary tree, to take an entity item of the entity as a child node of the node, and to assign a unique identifier to the entity item, to obtain the preset dictionary tree.

7. A non-transitory computer readable storage medium having computer programs stored thereon, wherein when the computer programs are executed by a processor, the processor is caused to perform:

determining a plurality of target candidate entities from a preset dictionary tree based on a query request;

for each target candidate entity, obtaining a first probability that the target candidate entity is a legitimate entity by calculating the target candidate entity based on a language model;

evaluating, based on an evaluation model, each target candidate entity to obtain an evaluation result, a target candidate entity corresponding to an evaluation result; and

determining a real intent entity corresponding to the query request based on the first probability and the evaluation result;

wherein evaluating each target candidate entity comprises:

extracting attribution information corresponding to respective target candidate entities;

determining weights for the attribution information corresponding to respective target candidate entities; and

evaluating each target candidate entity based on the weight;

wherein determining the real intent entity corresponding to the query request based on the first probability and the evaluation result comprises:

for each target candidate entity, weighting the first probability and the evaluation result corresponding to the target candidate entity, to obtain a weighted value corresponding to the target candidate entity; and

determining a target candidate entity corresponding to a maximum weighted value as the real intent entity.

8. The error correction method of claim 1 , wherein, determining the plurality of target candidate entities from the preset dictionary tree based on the query request comprises:

performing a calculation on a character string in the query request and the preset dictionary tree, to obtain a plurality of original candidate entities; and

selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on a second threshold.

9. The error correction apparatus of claim 4 , wherein determining the plurality of target candidate entities from the preset dictionary tree based on the query request comprises:

performing a calculation on a character string in the query request and the preset dictionary tree, to obtain a plurality of original candidate entities; and

selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on a second threshold.

10. The error correction method of claim 1 , wherein the language model comprises a NGRAM language model.

11. The error correction method of claim 1 , wherein evaluating each target candidate entity employs a neural network model.

12. The error correction method of claim 1 , wherein a learning to rank model is utilized in the evaluating each target candidate entity.

13. The error correction method of claim 8 , wherein selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on the second threshold comprises:

calculating a first distance between each original candidate entity and the query request;

comparing each first distance with the second threshold; and

determining the original candidate entity corresponding to the first distance smaller than or equal to the second threshold as the target candidate entity.

14. The error correction method of claim 8 , wherein when the second threshold comprises a third threshold and the fourth threshold, selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on the second threshold comprises:

calculating a second distance between each original candidate entity and the query request based on a first algorithm;

comparing each second distance with the third threshold;

extracting a first candidate entity corresponding to the second distance smaller than or equal to the third threshold from the plurality of original candidate entities;

calculating a third distance between each first candidate entity and the query request based on a second algorithm; and

taking the first candidate entity corresponding to the third distance smaller than or equal to the fourth threshold as the target candidate entity.

15. The error correction method of claim 14 , wherein

when the first algorithm is an edit distance algorithm, the second algorithm is a Jaccard distance algorithm; and

when the first algorithm is the Jaccard distance algorithm, the second algorithm is the edit distance algorithm.

16. The error correction apparatus of claim 9 , wherein selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on the second threshold comprises:

calculating a first distance between each original candidate entity and the query request;

comparing each first distance with the second threshold; and

determining the original candidate entity corresponding to the first distance smaller than or equal to the second threshold as the target candidate entity.

17. The error correction apparatus of claim 9 , wherein when the second threshold comprises a third threshold and the fourth threshold, selecting the plurality of target candidate entities corresponding to the query request from the plurality of original candidate entities based on the second threshold comprises:

calculating a second distance between each original candidate entity and the query request based on a first algorithm;

comparing each second distance with the third threshold;

extracting a first candidate entity corresponding to the second distance smaller than or equal to the third threshold from the plurality of original candidate entities;

calculating a third distance between each first candidate entity and the query request based on a second algorithm; and

taking the first candidate entity corresponding to the third distance smaller than or equal to the fourth threshold as the target candidate entity.

18. The error correction apparatus of claim 17 , wherein

when the first algorithm is an edit distance algorithm, the second algorithm is a Jaccard distance algorithm; and

when the first algorithm is the Jaccard distance algorithm, the second algorithm is the edit distance algorithm.