Method and apparatus for updating mining model

• US 10,586,080 B2
• Filed: 03/07/2018
• Issued: 03/10/2020
• Est. Priority Date: 03/31/2017
• Status: Active Grant

• Current Assignee: Baidu Online Network Technology (Beijing) Co., Ltd (Baidu Incorporated)
• Original Assignee: Baidu Online Network Technology (Beijing) Co., Ltd (Baidu Incorporated)
• Inventors: Hu, Zejin
• Primary Examiner(s): Jackson, Jekieda R
• Application Number:
  US15/914,098
• Publication Number:
  US 20180285346A1
• Time in Patent Office: 734 Days
• Field of Search: 704 9
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/027 Frames

Abstract

The present disclosure provides a method and an apparatus for updating a mining model, which are applied to the mining model in a semantic analysis system. The method includes: acquiring a semantic analysis result obtained by analyzing a query through the semantic analysis system; generating a training format corpus according to the semantic analysis result, in which the training format corpus includes a first format corpus and a second format corpus; and performing an iterative updating on the mining model based on the first format corpus and the second format corpus, in which the first format corpus is configured to label a semantic feature in the query, and the second format corpus is configured to label a template feature in the query. With the present disclosure, a semantic analysis effect of the semantic analysis system may be improved effectively.

Claims

1. A method for updating a mining model, wherein the method is applied to the mining model in a semantic analysis system and comprises:

acquiring a semantic analysis result obtained by analyzing a query through the semantic analysis system;

generating a training format corpus according to the semantic analysis result, wherein the training format corpus comprises a first format corpus and a second format corpus;

performing an iterative updating on the mining model based on the first format corpus and the second format corpus,

wherein the first format corpus is configured to label a semantic feature in the query, and the second format corpus is configured to label a template feature in the query;

analyzing a plurality of test statements based on the mining model before the iterative updating so as to acquire a first analysis result;

analyzing the plurality of test statements based on the mining model after the iterative updating so as to acquire a second analysis result;

according to a preset model scoring rule, scoring the first analysis result so as to acquire a first score, and scoring the second analysis result so as to acquire a second score; and

regarding the mining model after the iterative updating as the mining model of the semantic analysis system when the second score is larger than the first score.

2. The method according to claim 1, wherein the first format corpus comprises:

first semantic contents corresponding to a field tag and an intention tag in the query respectively, a plurality of slot templates, and second semantic contents corresponding to the plurality of slot templates respectively.

3. The method according to claim 2, wherein generating a training format corpus according to the semantic analysis result comprises:

performing a segmentation on the query according to the semantic analysis result so as to acquire a plurality of segments;

regarding the plurality of segments as the first format corpus in the training format corpus.

4. The method according to claim 2, wherein the second format corpus comprises:

a plurality of combinations of at least two slot templates, and third semantic contents corresponding to the plurality of combinations respectively.

5. The method according to claim 4, wherein generating a training format corpus according to the semantic analysis result comprises:

extracting the at least two slot templates and the second semantic contents corresponding to the at least two slot templates from the first format corpus;

performing a permutation and combination on the at least two slot templates so as to acquire the plurality of combinations; and

regarding the plurality of combinations and the third semantic contents corresponding to the plurality of combinations as the second format corpus.

6. The method according to claim 1, wherein after the training format corpus is generated according to the semantic analysis result, the method further comprises:

screening the training format corpus.

7. An apparatus for updating a mining model, wherein the apparatus is applied to the mining model in a semantic analysis system and comprises:

a processor;

a memory for storing programs executable by the processor,

wherein the processor is configured to;

acquire a semantic analysis result obtained by analyzing a query through the semantic analysis system;

generate a training format corpus according to the semantic analysis result, wherein the training format corpus comprises a first format corpus and a second format corpus;

perform an iterative updating on the mining model based on the first format corpus and the second format corpus,

wherein the first format corpus is configured to label a semantic feature in the query, and the second format corpus is configured to label a template feature in the query;

analyze a plurality of test statements based on the mining model before the iterative updating so as to acquire a first analysis result;

analyze the plurality of test statements based on the mining model after the iterative updating so as to acquire a second analysis result;

according to a preset model scoring rule, score the first analysis result so as to acquire a first score, and score the second analysis result so as to acquire a second score; and

regard the mining model after the iterative updating as the mining model of the semantic analysis system when the second score is larger than the first score.

8. The apparatus according to claim 7, wherein the first format corpus comprises:

first semantic contents corresponding to a field tag and an intention tag in the query respectively, a plurality of slot templates, and second semantic contents corresponding to the plurality of slot templates respectively.

9. The apparatus according to claim 8, wherein the processor is configured to, generate a training format corpus according to the semantic analysis result, by acts of:

performing a segmentation on the query according to the semantic analysis result so as to acquire a plurality of segments;

regarding the plurality of segments as the first format corpus in the training format corpus.

10. The apparatus according to claim 8, wherein the second format corpus comprises:

a plurality of combinations of at least two slot templates, and third semantic contents corresponding to the plurality of combinations respectively.

11. The apparatus according to claim 10, wherein the processor is configured to, generate a training format corpus according to the semantic analysis result, by acts of:

extracting the at least two slot templates and the second semantic contents corresponding to the at least two slot templates from the first format corpus;

performing a permutation and combination on the at least two slot templates so as to acquire the plurality of combinations; and

regarding the plurality of combinations and the third semantic contents corresponding to the plurality of combinations as the second format corpus.

12. The apparatus according to claim 7, wherein the processor is further configured to:

screen the training format corpus.

13. A non-transitory computer-readable storage medium having stored therein instructions that, when executed by a processor of an apparatus, causes the apparatus to perform method for updating a mining model, the method comprising:

acquiring a semantic analysis result obtained by analyzing a query through the semantic analysis system;

generating a training format corpus according to the semantic analysis result, wherein the training format corpus comprises a first format corpus and a second format corpus;

performing an iterative updating on the mining model based on the first format corpus and the second format corpus,

wherein the first format corpus is configured to label a semantic feature in the query, and the second format corpus is configured to label a template feature in the query;

analyzing a plurality of test statements based on the mining model before the iterative updating so as to acquire a first analysis result;

analyzing the plurality of test statements based on the mining model after the iterative updating so as to acquire a second analysis result;

according to a preset model scoring rule, scoring the first analysis result so as to acquire a first score, and scoring the second analysis result so as to acquire a second score; and

regarding the mining model after the iterative updating as the mining model of the semantic analysis system when the second score is larger than the first score.

14. The non-transitory computer-readable storage medium according to claim 13, wherein the first format corpus comprises:

first semantic contents corresponding to a field tag and an intention tag in the query respectively, a plurality of slot templates, and second semantic contents corresponding to the plurality of slot templates respectively.

15. The non-transitory computer-readable storage medium according to claim 14, wherein generating a training format corpus according to the semantic analysis result comprises:

performing a segmentation on the query according to the semantic analysis result so as to acquire a plurality of segments;

regarding the plurality of segments as the first format corpus in the training format corpus.

16. The non-transitory computer-readable storage medium according to claim 14, wherein the second format corpus comprises:

a plurality of combinations of at least two slot templates, and third semantic contents corresponding to the plurality of combinations respectively.

17. The non-transitory computer-readable storage medium according to claim 16, wherein generating a training format corpus according to the semantic analysis result comprises:

extracting the at least two slot templates and the second semantic contents corresponding to the at least two slot templates from the first format corpus;

performing a permutation and combination on the at least two slot templates so as to acquire the plurality of combinations; and

regarding the plurality of combinations and the third semantic contents corresponding to the plurality of combinations as the second format corpus.