METHODS, DEVICES, AND SYSTEMS FOR CONSTRUCTING INTELLIGENT KNOWLEDGE BASE

• US 20190332672A1
• Filed: 07/08/2019
• Published: 10/31/2019
• Est. Priority Date: 12/31/2015
• Status: Active Grant

• Current Assignee: Shanghai Xiaoi Robot Technology Co., Ltd.
• Original Assignee: Shanghai Xiaoi Robot Technology Co., Ltd.
• Inventors: ZENG, Yongmei, LI, Bo, YAO, Gongzhi, ZHU, Pinpin
• Granted Patent: US 11,017,178 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A device for constructing an intelligent knowledge base, comprising an abstract semantic expression obtaining unit to obtain a plurality of abstract semantic expressions from an abstract semantic database, wherein each of the plurality of abstract semantic expressions comprises a semantic-lacking element; a receiving unit to receive an initial request message; an abstract semantic recommending module to obtain abstract semantic expressions corresponding to the initial request message by performing an abstract semantic recommending process on the initial request message based on the plurality of abstract semantic expressions; a filling unit to extract from the initial request message an element corresponding to a semantic-lacking element of the obtained abstract semantic expressions, and to fill the extracted element into the semantic-lacking element to obtain specific semantic expressions corresponding to the initial request message; and a storing unit to store the initial request message and the specific semantic expressions into the intelligent knowledge base.

Claims

1. A device for constructing an intelligent knowledge base, comprising:

an abstract semantic expression obtaining unit to obtain a plurality of abstract semantic expressions from an abstract semantic database, wherein each of the plurality of abstract semantic expressions comprises a semantic-lacking element;

a receiving unit to receive an initial request message;

an abstract semantic recommending module, coupled to the abstract semantic expression obtaining unit and the receiving unit, to obtain one or more abstract semantic expressions corresponding to the initial request message by performing an abstract semantic recommending process on the initial request message based on the plurality of abstract semantic expressions;

a filling unit, coupled to the abstract semantic recommending module, to extract from the initial request message an element corresponding to a semantic-lacking element of the obtained one or more abstract semantic expressions, and to fill the extracted element into the semantic-lacking element to obtain one or more specific semantic expressions corresponding to the initial request message; and

a storing unit to store the initial request message and the one or more specific semantic expressions into the intelligent knowledge base.

2. The device according to claim 1, wherein the abstract semantic recommending module comprises:

a word segmentation unit to perform a word segmentation process on the initial request message to obtain one or more single words;

a rule word identification unit to identify each of the one or more single words as a semantic rule word or a non-semantic rule word;

a part-of-speech tagging unit to perform a part-of-speech tagging process on each non-semantic rule word to obtain part-of-speech information of the non-semantic rule word;

a wordclass determination unit to perform a wordclass determination process on each semantic rule word to obtain wordclass information of the each semantic rule word;

a searching unit to perform a searching process on the obtained one or more abstract semantic expressions to acquire an abstract semantic candidate set relevant to the initial request message, wherein the abstract semantic candidate set comprises several abstract semantic expressions; and

a matching unit to derive an abstract semantic expression corresponding to the initial request message by performing a matching process on the several abstract semantic expressions in the abstract semantic candidate set based on the part-of-speech information and the wordclass information.

3. The device according to claim 1, wherein the abstract semantic recommending module comprises:

a word segmentation unit to perform a word segmentation process on the initial request message to obtain one or more single words;

a part-of-speech tagging unit to perform a part-of-speech tagging process on each of the one or more single words to obtain part-of-speech information of each of the one or more single words;

a wordclass determination unit to perform a wordclass determination process on each of the one or more single words to obtain its wordclass information;

a searching unit to perform a searching process on the obtained one or more abstract semantic expressions to acquire an abstract semantic candidate set relevant to the initial request message, wherein the abstract semantic candidate set comprises several abstract semantic expressions; and

a matching unit to derive an abstract semantic expression corresponding to the initial request message by performing a matching process on the several abstract semantic expressions in the abstract semantic candidate set based on the part-of-speech information and the wordclass information.

4. The device according to claim 3, wherein the abstract semantic recommending module further comprises:

a number determination unit to determine, before the matching process, whether a number of the several abstract semantic expressions in the abstract semantic candidate set is within a predetermined range;

a removing unit to remove, when the number of the several abstract semantic expressions in the abstract semantic candidate set is above the predetermined range, one or more of the several abstract semantic expressions; and

a supplementing unit to supplement, when the number of the several abstract semantic expressions in the abstract semantic candidate set is under the predetermined range, one or more of abstract semantic expressions from a default set.

5. The device according to claim 2, wherein:

the derived abstract semantic expression comprises a semantic rule word; and

at least one of semantic rule words of the several abstract semantic expressions in the abstract semantic candidate set is the same or belongs to a same wordclass as one of the single words of the initial request message.

6. The device according to claim 5, wherein the derived abstract semantic expression corresponding to the initial request message satisfies the following conditions:

a part-of-speech corresponding to the semantic-lacking element comprises a part-of-speech of the filling element;

a keyword of the initial request message except the filling element is the same or belongs to a same wordclass as the semantic rule word of the derived abstract semantic expression; and

the derived abstract semantic expression has a same order as the initial request message.

7. The device according to claim 3, further comprising:

a scoring unit to;

when a number M of the obtained one or more abstract semantic expressions corresponding to the initial request message is greater than a number N of the obtained one or more specific semantic expressions, perform a scoring process on each abstract semantic expression corresponding to the initial request message;

wherein the filling unit further;

fills the extracted element into semantic-lacking elements of the N abstract semantic expressions having higher scores; and

obtains N specific semantic expressions corresponding to the initial request message.

8. The device according to claim 7, wherein the scoring unit performs a scoring process according to one or more methods selected from a group comprising:

setting a higher score to an abstract semantic expression corresponding to the initial request message if a number of the matched semantic-lacking elements is higher;

defining a semantic-lacking element of the abstract semantic expression as a core semantic element, and setting a higher score to an abstract semantic expression if its semantic-lacking element is closer to the core semantic element;

setting a higher score to an abstract semantic expression if a confidence value of its part-of-speech is higher;

setting a higher score to an abstract semantic expression if a priority level is higher, wherein the priority level is pre-assigned to the abstract semantic expression; and

/or

setting a higher score to an abstract semantic expression if a probability is higher, wherein the probability is determined based on a natural language model, and corresponds to whether data information obtained by filling segmentation words of a large amount of crawled corpus data into the abstract semantic expression has correct semantic information.

9. The device according to claim 1, further comprising:

an answer providing unit to provide an answer corresponding to the initial request message, wherein the storing unit stores the initial request message along with the provided answer into the intelligent knowledge base.

10. A computer-implemented method for constructing an intelligent knowledge base, comprising:

providing an abstract semantic database, wherein the abstract semantic database comprises a plurality of abstract semantic expressions and each of the plurality of abstract semantic expressions comprises a semantic-lacking element;

receiving, via an intelligent-knowledge-base constructing device, an initial request message, which needs an abstract semantic recommendation;

performing an abstract semantic recommending process on the initial request message based on the abstract semantic database;

obtaining, via the intelligent-knowledge-base constructing device, one or more abstract semantic expressions corresponding to the initial request message;

extracting, from the initial request message, an element corresponding to a semantic-lacking element of the one or more abstract semantic expressions;

filling the extracted element into the semantic-lacking element of the one or more abstract semantic expressions to obtain one or more specific semantic expressions corresponding to the initial request message; and

storing the initial request message and the one or more specific semantic expressions into the intelligent knowledge base.

11. The method according to claim 10, wherein the abstract semantic recommending process comprises:

performing a word segmentation process on the initial request message to obtain one or more single words;

identifying each of the one or more single words as a semantic rule word or a non-semantic rule word;

performing a part-of-speech tagging process on each non-semantic rule word to obtain part-of-speech information of the non-semantic rule word;

performing a wordclass determination process on each semantic rule word to obtain wordclass information of the semantic rule word;

obtaining an abstract semantic candidate set relevant to the initial request message by performing a searching process on the abstract semantic database, wherein the abstract semantic candidate set comprises several abstract semantic expressions; and

deriving an abstract semantic expression corresponding to the initial request message by performing a matching process on the several abstract semantic expressions in the abstract semantic candidate set based on the part-of-speech information and the wordclass information.

12. The method according to claim 10, wherein the abstract semantic recommending process comprises:

performing a word segmentation process on the initial request message to obtain one or more single words;

performing a part-of-speech tagging process on each single word to obtain part-of-speech information of the each single word;

performing a wordclass determination process on each single word to obtain wordclass information of the each single word;

performing a searching process on the abstract semantic database to obtain an abstract semantic candidate set relevant to the initial request message, wherein the abstract semantic candidate set comprises several abstract semantic expressions; and

deriving an abstract semantic expression corresponding to the initial request message by performing a matching process on the several abstract semantic expressions in the abstract semantic candidate set based on the part-of-speech information and the wordclass information.

13. The method according to claim 12, further comprising:

determining, before the matching process, whether a number of the several abstract semantic expressions in the abstract semantic candidate set is within a predetermined range;

removing, when the number of the several abstract semantic expressions in the abstract semantic candidate set is above the predetermined range, one or more of the several abstract semantic expressions; and

supplementing, when the number of the several abstract semantic expressions in the abstract semantic candidate set is under the predetermined range, one or more of abstract semantic expressions from a default set.

14. The method according to claim 11, wherein:

the derived abstract semantic expression comprises a semantic rule word; and

at least one of semantic rule words of the several abstract semantic expressions in the abstract semantic candidate set is the same or belongs to a same wordclass as one of the single words of the initial request message.

15. The method according to claim 14, wherein the derived abstract semantic expression corresponding to the initial request message satisfies the following conditions:

a part-of-speech corresponding to the semantic-lacking element comprises a part-of-speech of the filling element;

a keyword of the initial request message except the filling element is the same or belongs to a same wordclass as the semantic rule word of the derived abstract semantic expression; and

the derived abstract semantic expression has a same order as the initial request message.

16. The method according to claim 12, further comprising:

determining a number M of the obtained one or more abstract semantic expressions corresponding to the initial request message is greater than a number N of the specific semantic expressions;

performing, based on the determination, a scoring process on each abstract semantic expression corresponding to the initial request message;

extracting, from the initial request message, elements corresponding to semantic-lacking elements of N abstract semantic expressions having higher scores;

filling the extracted elements into the corresponding semantic-lacking elements of the N abstract semantic expressions having higher scores; and

obtaining N specific semantic expressions corresponding to the initial request message.

17. The method according to claim 16, wherein the scoring process employs one or more methods selected from a group comprising:

setting a higher score to an abstract semantic expression corresponding to the initial request message if a number of the matched semantic-lacking elements is higher;

defining a semantic-lacking element of the abstract semantic expression as a core semantic element, and setting a higher score to an abstract semantic expression if its semantic-lacking element is closer to the core semantic element;

setting a higher score to an abstract semantic expression if a confidence value of its part-of-speech is higher;

setting a higher score to the abstract semantic expression if a priority level is higher, wherein the priority level is pre-assigned to the abstract semantic expression; and

/or

setting a higher score to an abstract semantic expression if a probability is higher, wherein the probability is determined based on a natural language model, and corresponds to whether data information obtained by filling segmentation words of a large amount of crawled corpus data into the abstract semantic expression has correct semantic information.

18. The method according to claim 10, further comprising:

when an abstract semantic expression corresponding to the initial request message cannot be obtained, prompting to manually add an extended question corresponding to the initial request message into the intelligent knowledge base.

19. The method according to claim 18, further comprising:

when the initial request message is stored into the intelligent knowledge base, providing an answer corresponding to the initial request message; and

storing the answer along with the initial request message into the intelligent knowledge base.

20. A non-transitory computer-readable-medium storing instructions that, when executed, cause a computer to perform a method for constructing an intelligent knowledge base according to any one of the above method claims.