Construction method and device of chemical engineering knowledge graph and intelligent question answering method and device

1 . An intelligent question answering method for chemical knowledge, comprising:

obtaining a question inputted by a user through a human-computer interaction interface, wherein the question is related to an abnormal operating condition in a chemical process;

inputting the question into a pre-trained deep learning module to obtain a question entity data and a question property data related to the chemical knowledge thereof, and identifying an intention of the question, wherein the deep learning model is trained on question samples according to the chemical knowledge;

using a question analysis and inference module to determine a first graph entity data associated with the question entity data from each graph entity data of a chemical knowledge graph;

when the intention of the question is to retrieve a corresponding property according to a first entity and a second entity, performing the following:

selecting all second graph entity data that are once related or twice related to the first graph entity data according to a standard knowledge representation in the chemical knowledge graph, wherein a relationship of once related refers to the first graph entity data that can be associated with the second graph entity data through one standard knowledge representation, and a relationship of twice related refers to the first graph entity data that can be associated with the second graph entity data through two standard knowledge representations; and

respectively combining the first graph entity data with each of selected second graph entity data to obtain a plurality of candidate paths;

using a question post-processing module to perform the following:

inputting the question into a word vector model pre-trained on a basis of chemical knowledge samples to obtain a first vector of the question;

respectively inputting the plurality of candidate paths into the word vector model to obtain a second vector of each;

respectively calculating a cosine value between each second vector and each first vector to serve as a text matching degree between each of the plurality of candidate paths and the question, and selecting a candidate path with a highest text matching degree as an optimal search path; and

searching the chemical knowledge graph according to the optimal search path to obtain an answer corresponding to the question, wherein the answer provides a cause, a consequence, or a safety precaution associated with the abnormal operating condition; and

returning the answer in a standard form through the human-computer interaction interface.

2 . The intelligent question answering method according to claim 1 , wherein the pre-trained deep learning module comprises an entity link dictionary and a property dictionary, steps of identifying the question entity data and the question property data related to the chemical knowledge thereof further comprise:

inputting the obtained question entity data into the entity link dictionary, and mapping the question entity data to data consistent with a description of the chemical knowledge graph according to synonyms and/or machine learning fuzzy matching; and

inputting the obtained question property data into the property dictionary, and mapping the question property data to data consistent with the description of the chemical knowledge graph according to synonyms and/or machine learning fuzzy matching.

3 . The intelligent question answering method according to claim 1 , wherein steps of identifying the intention of the question comprise:

in response to identifying the question entity data related to the chemical knowledge and a corresponding question property data from the question, determining that the intention of the question is to retrieve a corresponding second entity according to a first entity and a property; and

in response to identifying two question entity data related to the chemical knowledge from the question, determining that the intention of the question is to retrieve a corresponding property according to the first entity and the second entity.

4 . The intelligent question answering method according to claim 1 , wherein steps of searching the chemical knowledge graph according to the optimal search path to obtain the answer corresponding to the question comprise:

searching the chemical knowledge graph according to the optimal search path to determine a corresponding standard knowledge representation;

determining a position of the answer in the standard knowledge representation according to the intention of the question; and

organizing the answers according to the question to obtain the answer in a standard form.

5 . The intelligent question answering method according to claim 4 , further comprising a following step:

returning the answer in the standard form to the user.

6 . The intelligent question answering method according to claim 1 , wherein before obtaining the question raised by the user, the intelligent question answering method further comprises following steps:

obtaining a knowledge data in a chemical field;

pre-processing the knowledge data to obtain an entity data and a property data related to the chemical knowledge;

determining a preliminary knowledge representation according to the entity data and the property data;

performing entity alignment on the preliminary knowledge representation to obtain the standard knowledge representation; and

constructing the chemical knowledge graph according to the standard knowledge representation.

7 . The intelligent question answering method according to claim 6 , wherein the knowledge data comprises a structured data, a semi-structured data and/or an unstructured data, and steps of pre-processing the knowledge data comprise:

integrating the structured data to obtain the entity data and the property data related to the chemical knowledge; and/or

performing knowledge extraction on the semi-structured data and/or the unstructured data to obtain the entity data and the property data related to the chemical knowledge.

8 . The intelligent question answering method according to claim 7 , wherein the property data comprises a numeric property data and a relation property data, wherein the numeric property data is used to describe a property value of one entity data in one preliminary knowledge representation, and the relation property data is used to describe a relationship between two entity data in the one preliminary knowledge representation.

9 . The intelligent question answering method according to claim 8 , wherein step of determining the preliminary knowledge representation according to the entity data and the property data comprises:

constructing the preliminary knowledge representation in a triplet form according to the obtained entity data and property data in a form of entity-numeric property-property value, or first entity-relation property-second entity.

10 . The intelligent question answering method according to claim 6 , wherein steps of performing entity alignment on the preliminary knowledge representation to obtain the standard knowledge representation comprise:

analyzing a plurality of the preliminary knowledge representations to determine a plurality of different entity data indicating one chemical entity; and

decomposing the plurality of different entity data indicating the one chemical entity into the same entity data to obtain the standard knowledge representation that utilizes the same entity data to indicate the same chemical entity.

11 . The intelligent question answering method according to claim 6 , wherein steps of constructing the chemical knowledge graph according to the standard knowledge representation comprise:

performing knowledge discovery according to the entity data and the property data of the plurality of standard knowledge representations to obtain at least one high-credibility standard knowledge representation;

performing knowledge inference according to the entity data and the property data of the plurality of standard knowledge representations to obtain a plurality of standard knowledge representations with unknown credibility;

evaluating a quality of the plurality of standard knowledge representations with unknown credibility to determine the high-credibility standard knowledge representation thereof, and

constructing the chemical knowledge graph according to each of the high-credibility standard knowledge representations.

12 . The intelligent question answering method according to claim 11 , wherein steps of evaluating the quality of the plurality of standard knowledge representations with unknown credibility comprise:

performing text matching between the plurality of standard knowledge representations with unknown credibility and the knowledge data in the chemical field to obtain the text matching degree of each standard knowledge representation; and

determining the standard knowledge representation with the text matching degree that is higher than a preset matching degree threshold as a high-credibility standard knowledge representation.

13 . An intelligent question answering device for chemical knowledge, comprising:

a memory having computer-readable instructions stored thereon; and

a processor that is connected to the memory,

wherein the instructions, when executed by the processor, allow the processor to obtain a question inputted by a user through a human-computer interaction interface,

wherein the question is related to an abnormal operating condition in a chemical process;

input the question into a pre-trained deep learning to obtain a question entity data and a question property data related to the chemical knowledge thereof, and identifying an intention of the question, wherein the deep learning model is trained on question samples according to the chemical knowledge;

use a question analysis and inference module to

determine a first graph entity data associated with the question entity data from each graph entity data of a chemical knowledge graph;

when the intention of the question is to retrieve a corresponding property according to a first entity and a second entity, performing the following:

select all second graph entity data that are once related or twice related to the first graph entity data according to a standard knowledge representation in the chemical knowledge graph, wherein a relationship of once related refers to the first graph entity data that can be associated with the second graph entity data through one standard knowledge representation, and a relationship of twice related refers to the first graph entity data that can be associated with the second graph entity data through two standard knowledge representations; and

respectively combine the first graph entity data with each of selected second graph entity data to obtain a plurality of candidate paths;

use a question post-processing module to perform the following:

input the question into a word vector model pre-trained on a basis of chemical knowledge samples to obtain a first vector of the question;

respectively input the plurality of candidate paths into the word vector model to obtain a second vector of each;

respectively calculate a cosine value between each second vector and each first vector to serve as a text matching degree between each of the plurality of candidate paths and the question, and selecting a candidate path with a highest text matching degree as an optimal search path; and

search the chemical knowledge graph according to the optimal search path to obtain an answer corresponding to the question, wherein the answer provides a cause, a consequence, or a safety precaution associated with the abnormal operating condition; and

return the answer in a standard form through the human-computer interaction interface.

14 . A non-transitory computer-readable storage medium having computer instructions therein, wherein the computer instructions are configured to cause a computer to at least:

obtain a question inputted by a user through a human-computer interaction interface, wherein the question is related to an abnormal operating condition in a chemical process;

input the question into a pre-trained deep learning module to obtain a question entity data and a question property data related to the chemical knowledge thereof, and identifying an intention of the question, wherein the deep learning model is trained on question samples according to the chemical knowledge;

use a question analysis and inference module to

determine a first graph entity data associated with the question entity data from each graph entity data of a chemical knowledge graph;

when the intention of the question is to retrieve a corresponding property according to a first entity and a second entity, performing the following:

select all second graph entity data that are once related or twice related to the first graph entity data according to a standard knowledge representation in the chemical knowledge graph, wherein a relationship of once related refers to the first graph entity data that can be associated with the second graph entity data through one standard knowledge representation, and a relationship of twice related refers to the first graph entity data that can be associated with the second graph entity data through two standard knowledge representations; and

respectively combine the first graph entity data with each of selected second graph entity data to obtain a plurality of candidate paths;

use a question post-processing module to perform the following:

input the question into a word vector model pre-trained on a basis of chemical knowledge samples to obtain a first vector of the question;

respectively input the plurality of candidate paths into the word vector model to obtain a second vector of each;

respectively calculate a cosine value between each second vector and each first vector to serve as a text matching degree between each of the plurality of candidate paths and the question, and selecting a candidate path with a highest text matching degree as an optimal search path; and

search the chemical knowledge graph according to the optimal search path to obtain an answer corresponding to the question, wherein the answer provides a cause, a consequence, or a safety precaution associated with the abnormal operating condition; and

return the answer in a standard form through the human-computer interaction interface.