CN111767453A

CN111767453A - Query instruction generation method, device, equipment and storage medium based on semantic network

Info

Publication number: CN111767453A
Application number: CN202010566674.3A
Authority: CN
Inventors: 殷嘉珩; 顾根
Original assignee: Shanghai Synyi Medical Technology Co ltd
Current assignee: Shanghai Synyi Medical Technology Co ltd
Priority date: 2020-06-09
Filing date: 2020-06-19
Publication date: 2020-10-13

Abstract

The application provides a query instruction generation method, a query instruction generation device, query instruction generation equipment and a storage medium based on a semantic network, and query input information is acquired; converting each word into a representation vector according to semantic web association and associated concept information of context information corresponding to each word in the input information; converting the characterization vectors corresponding to the words into word slot labels, and obtaining intentions according to the word slot labels corresponding to all the words; producing corresponding instructions according to the intents. The method and the device convert the words into the representation vectors according to the context of the words in the input information and the knowledge information of the words, and convert the representation vectors into the intentions. Compared with a static instruction generation system according to keywords, the same word can be converted into different representation vectors according to the context information and background knowledge of the word, so that richer information is provided, the representation vectors can be converted into more accurate intentions, and the overall performance of the instruction generation system is improved.

Description

Query instruction generation method, device, equipment and storage medium based on semantic network

Technical Field

The invention relates to the technical field of semantic recognition, in particular to a query instruction generating method, a query instruction generating device, query instruction generating equipment and a storage medium based on a semantic network.

Background

At present, many query instruction generation methods based on natural language processing technology exist, which can help people to obtain more needed contents in a system through generated instructions, for example, a hundred-degree search can automatically correct wrong words and wrong words in user query keywords and return more ideal search results, and search requirements of most users on the internet can be met. However, in some situations, particularly in the vertical domain subdivision, the existing algorithms still cannot meet the requirements, for example, in the process of querying medical data, a large amount of information is not obtained through keywords, and semantic and knowledge information behind the keywords needs to be understood. It makes sense how to correctly generate a more accurate query instruction based on a small amount of keyword information.

Disclosure of Invention

In view of the above-mentioned shortcomings in the prior art, it is an object of the present application to provide a method, apparatus, device and storage medium for generating a semantic web-based query instruction, so as to solve at least one problem in the prior art.

To achieve the above and other related objects, the present application provides a query instruction generating method based on semantic web, including: acquiring input information of query; converting each word into a representation vector according to semantic web association and associated concept information of context information corresponding to each word in the input information; converting the characterization vectors corresponding to the words into word slot labels, and obtaining intentions according to the word slot labels corresponding to all the words; producing corresponding instructions according to the intents.

In an embodiment of the application, the converting each word into a representation vector according to semantic web association and associated concept information of context information corresponding to each word in the input information includes: modeling the semantics and background knowledge of each word based on the context information and semantic web information of each word in the input information to form a graph embedding model; and training semantic web association according to the context information and associating the concept information through the graph embedding model to form a characterization vector.

In an embodiment of the present application, the method includes: the graph embedding model is trained by using associated concept information of the semantic net in an unsupervised mode, so that the graph embedding model can use a large amount of semantic net data in the open field.

In an embodiment of the present application, the graph embedding model includes: any one of a Deepwalk model, a Node2Vec model, a HARP model, and a Walklets model.

In an embodiment of the application, the converting the characterization vector corresponding to each word into a word slot label, and obtaining an intention according to the word slot labels corresponding to all the words includes: converting the characterization vector into the word slot tag sequence and the intention sequence through a discriminant model; the distinguishing model obtains the semantics and background knowledge of each word through the graph embedding model so as to model and reason words, and converts the representation vectors corresponding to the words into corresponding word slot labels and corresponding intentions according to the modeling and the reasoning of the distinguishing model.

In an embodiment of the present application, the discriminant model includes: a word slot label classifier and an intent classifier to convert the token vector into a word slot label and an intent, respectively.

To achieve the above and other related objects, the present application provides a semantic web-based query instruction generating apparatus, including: the acquisition module is used for acquiring the input information of the query; the first conversion module is used for converting each word into a representation vector according to semantic web association of context information corresponding to each word in the input information and associated concept information; the second conversion module is used for converting the representation vectors corresponding to the words into word slot labels and obtaining intentions according to word slot transitions corresponding to all the words; and the instruction generating module is used for producing a corresponding instruction according to the intention.

To achieve the above and other related objects, the present application provides a computer apparatus, comprising: a memory, and a processor; the memory is to store computer instructions; the processor executes computer instructions to implement the method as described above.

To achieve the above and other related objects, the present application provides a computer readable storage medium storing computer instructions which, when executed, perform the method as described above.

In summary, the query instruction generation method, device, equipment and storage medium based on the semantic network provided by the present application obtains the query input information; converting each word into a representation vector according to semantic web association and associated concept information of context information corresponding to each word in the input information; converting the characterization vectors corresponding to the words into word slot labels, and obtaining intentions according to the word slot labels corresponding to all the words; producing corresponding instructions according to the intents.

Has the following beneficial effects:

the method and the device convert the words into the representation vectors according to the context of the words in the input information and the knowledge information of the words, and convert the representation vectors into the intentions. Compared with a static instruction generation system according to keywords, the same word can be converted into different representation vectors according to the context information and knowledge of the word, so that richer information is provided, the representation vectors can be converted into more accurate intentions, and the overall performance of the instruction generation system is improved.

Drawings

FIG. 1 is a flowchart illustrating a semantic Web-based query instruction generation method according to an embodiment of the present application.

FIG. 2 is a block diagram of a semantic Web based query instruction generating apparatus according to an embodiment of the present invention.

Fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application is provided by way of specific examples, and other advantages and effects of the present application will be readily apparent to those skilled in the art from the disclosure herein. The present application is capable of other and different embodiments and its several details are capable of modifications and/or changes in various respects, all without departing from the spirit of the present application. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.

It should be noted that the drawings provided in the following embodiments are only schematic and illustrate the basic idea of the present application, and although the drawings only show the components related to the present application and are not drawn according to the number, shape and size of the components in actual implementation, the type, quantity and proportion of the components in actual implementation may be changed at will, and the layout of the components may be more complex.

At present, a plurality of query instruction generation methods based on natural language processing technology exist, but in some occasions, particularly in the vertical subdivision field, the existing algorithm still cannot meet the requirements, for example, in the query process of medical data, a large amount of information is not obtained through keywords, and semantic and knowledge information behind the keywords also needs to be understood. It makes sense how to correctly generate a more accurate query instruction based on a small amount of keyword information. To this end, the present application provides a query instruction generation method, apparatus, device and storage medium based on a semantic web, so as to solve at least one of the above problems.

Fig. 1 is a flow chart illustrating a semantic web-based query instruction generation method according to an embodiment of the present application. As shown, the method comprises:

step S101: acquiring input information of query;

in this embodiment, the input information includes instruction information for querying input by the user, and the instruction information is preferably text instruction information, or text instruction information converted from picture information or voice information. In some real-time instances, the queried content encompasses medical data where semantic and knowledge information is more complex.

Step S102: and converting each word into a representation vector according to the semantic web association of the context information corresponding to each word in the input information and the associated concept information.

In this embodiment, the probability of each word in the input information is usually related to one or more words before the word and one or more words after the word, and related words have related semantic relations in the semantic web, that is, each word is related to the context information of the word in the input information and the concept relations in the related semantic web.

The semantic web described in this application is a data-centric network in which information can be understood and processed by machines. The semantic network is an intelligent network which can judge according to semantics, and barrier-free communication between people and computers is realized. Compared with a huge brain, the intelligent degree is extremely high, and the coordination capability is very strong. Every computer connected on the semantic network can understand not only words and concepts, but also the logic relation between the words and the concepts, and can do the work of people. It will free human from the heavy work of searching related web pages and turn the user into a full-energy god. Computers in the semantic web can find needed information in mass resources on the world wide web by using own intelligent software, so that existing information islands are developed into a huge database. The establishment of the semantic Web is a part of the artificial intelligence field, which is not in line with the idea of the Web 3.0 intelligent network, so the initial implementation of the semantic Web is also one of the important features of the Web 3.0, but the realization of the semantic Web to become the super brain on the network requires long-term research, which means that the related implementation of the semantic Web occupies an important part of the network development process, and the semantic Web continues to be gradually converted into the "intelligent network" in several network times. Various web technologies are currently possible to apply to semantic web (in the sense of semantic web), such as: DOM document object model, a set of standard interfaces that access XML and HTML document components. XPath, XLink, XPointerXINCLude XML fragment XML query languages XHTML XML Schema, RDF (resource DescriptionFramework) XSL, XSLT Extensible Stylesheet Language SVG (Scalable vector graphics) SMIL DTD microformat metadata concepts.

In this embodiment, the step S102 is mainly implemented by using a graph embedding model, and the specific method includes:

A. modeling the semantics and background knowledge of each word based on the context information and semantic web information of each word in the input information to form a graph embedding model;

B. and training semantic web association according to the context information and associating the concept information through the graph embedding model to form a characterization vector.

In this embodiment, words are converted into characterization vectors according to the words in the input information and semantic gateways of their context concepts and associated concept information through a graph embedding model. The graph embedding model provides a source of the characterization vector, and can model the semantics and background knowledge of each word based on the context information and semantic web information of the word in the input information, so as to generate the characterization vector.

In addition, it should be noted that, for the graph embedding model of the present application, it is not required to train using labeled training samples, and the graph embedding model may be trained using associated concept information of the semantic web in an unsupervised manner, so that a large amount of semantic web data in the public domain may be used in the training of the graph embedding model, and therefore, the graph embedding model may learn the common sense regularity in the language understanding process from a large amount of training samples, and may migrate the capability of the regularity to the second conversion module, so that the instruction generation system of the present embodiment has stronger generalization, and further improves the performance of the instruction generation system.

Generally, social networks, word coexistence networks, and communication networks are widely available in various real-world applications. Through the analysis of the above, we can understand the social structure, language and different communication modes, so the graph is always the hot point of the research in the academic world. Graph analysis tasks can be roughly abstracted into the following four categories: (a) node classification, (b) link prediction, (c) clustering, and (d) visualization. Wherein the node classification aims at determining labels (also called vertices) of the nodes based on the otherwise labeled nodes and the network topology. Link prediction refers to the task of predicting missing links or links that may occur in the future. Clustering is used to find a subset of similar nodes and group them together; finally, visualization helps to gain insight into the network structure.

It should be noted that the goal of graph embedding is to find a low-dimensional vector representation of a high-dimensional graph, and it is very difficult to obtain a vector representation of each node in the graph, and has several challenges, which have been driving research in the field: attribute selection, scalability, embedded dimensionality. Over the past decade, there has been a great deal of research in the field of graphics embedding, with an emphasis on designing new embedding algorithms. As developed to the present, these embedding methods can be broadly classified into three major categories: (1) factorization (matrix factorization) -based methods, (2) random walk-based methods, and (3) deep learning-based methods. In the present application, a random walk-based method is collected.

Specifically, the graph embedding model in the present application includes: any one of a Deepwalk model, a Node2Vec model, a HARP model, and a Walklets model.

Deepwalk: inspired by word2vec, a certain specific point is selected as a starting point, a sequence of points is obtained by random walking, then the obtained sequence is regarded as a sentence, and word2vec is used for learning to obtain a representation vector/embedding of the point. Deep walk acquires local context information of a point in the graph through random walk, so that the learned representation vector reflects the local structure of the point in the graph, and the more adjacent points (or higher-order adjacent points) shared by two points in the graph, the shorter the distance between the corresponding two vectors/embeddings is.

node2 vec: similar to deep walk, node2vec maintains higher order proximity between nodes by maximizing the probability of occurrence of nodes in a sequence resulting from random walks. The biggest difference with deep walk is that node2vec uses biased random walk to make a trade-off between breadth-first (bfs) and depth-first (dfs) graph search, resulting in higher quality and more information content embedding than deep walk.

Harp (hierarchical representation for networks): deepwalk and node2vec random initialization node embedding (random initialization node embedding matrix, as model parameters, trained) to train the model. Such initialization is likely to fall into local optima since their objective function is non-convex. HARP introduces a strategy to improve the solution and avoid local optima through better weight initialization. To this end, HARP creates a hierarchy of nodes by coarsening nodes in a layer above the aggregate hierarchy using a graph. It then generates the embedding of the coarsest graph and initializes the node embedding (one of the hierarchies) of the refined graph with the learned embedding. It propagates this embedding through the hierarchy to obtain the embedding of the original graph. Therefore, HARP can be used in combination with random walk based methods (such as DeepWalk and node2vec) to obtain better optimization function solutions.

Walklets: the sequences generated by Deepwalk and node2vec by random walks implicitly preserve higher order proximity between nodes, which due to their randomness, will result in connected nodes of different distances. On the other hand, factorization based methods, such as GF and HOPE, explicitly preserve the distance between nodes by modeling the nodes in the objective function. Walklets combine explicit modeling with the idea of random walks. The model modifies the random walk strategy used in Deepwalk by jumping certain nodes in the graph. This is performed for a number of scales of jump lengths, similar to decomposing A in GraRep^kAnd the sequence of a set of points obtained by random walking is used to train a model similar to Deepwalk.

In this embodiment, the semantic web association and associated concept information respectively corresponding to each word and context information in the input information may be, for example:

the words include: the 'apple and release party' is context information corresponding to the 'apple', and a semantic gateway according to the context information is connected with: the related concept information corresponding to the background knowledge can be corresponding to a 'scientific and technological product', and then the term 'apple' can be obtained, namely the term 'fruit' or 'scientific and technological product' can be represented by the term 'scientific and technological product' to select the 'scientific and technological product' as the related concept, but not select the 'fruit' as the related concept.

Step S103: converting the characterization vectors corresponding to the words into word slot labels, and obtaining intentions according to the word slot labels corresponding to all the words;

in this embodiment, the step S103 is implemented by a discriminant model, and the method specifically includes: converting the characterization vector into the word slot tag sequence and the intention sequence through a discriminant model; the distinguishing model obtains the semantics and background knowledge of each word through the graph embedding model so as to model and reason words, and converts the representation vectors corresponding to the words into corresponding word slot labels and corresponding intentions according to the modeling and the reasoning of the distinguishing model.

For example, in the above example, according to the semantics of "products and technologies" corresponding to "release party" and the background knowledge of "technology products", it can be determined that the word slot corresponding to "apple" should be "technology products" or "technology manufacturers" instead of the concept of "fruits" according to "release party" as the "apple" context information.

Specifically, the discriminant model includes: a word slot label classifier and an intent classifier to convert the token vector into a word slot label and an intent, respectively.

In the embodiment, the converted characterization vectors are input into a word slot label classifier, and corresponding word slot label classification results are output; the slot label classifier is obtained by training each word according to semantic web correlation of context information corresponding to each word in the input information and a characterization vector converted by correlation concept information. The intention classifier may be similar to the word slot label classifier, and some preset intention classifications may also be input for training to obtain the intention classification result of the specified category, for example, the intention is preset to be classified into a purchase intention, a repair intention, a return intention, a sell intention, a complaint intention, and the like.

For example, as in the context of a shopping inquiry, the word slot labels may include "technology products" and "events" and the like, while the intent may include "purchases" and "returns" and the like.

It should be noted that, in the word representation stage of the instruction generation system, the graph embedding model provides a source of a representation vector, and can model the semantics and background knowledge of each word based on the context information and semantic web information of the word in the input information, so as to generate the representation vector and provide richer information for the discrimination model, and the discrimination model can convert each representation vector into a more accurate word slot label and intention, so as to improve the overall performance of the instruction generation system.

Step S104: producing corresponding instructions according to the intents.

In this embodiment, the preset instructions may be produced according to the obtained intention. For example, if the intention is determined as a purchase intention, then the input information includes information of a scientific and technical product, i.e., apple, and then the corresponding instruction at least includes: instructions related to purchasing apple technology products, such as: providing a link to a purchase website, or providing product information and quotation, etc.

To further facilitate understanding of the query instruction generation method based on semantic web described in the present application, the following examples are given:

for example, the input information is text instruction information input by the user: "apple publishes the meeting today, i want to buy a new cell phone". Firstly, words are converted into characterization vectors through semantic web association of the words in input information and context information thereof and association of concept information. Secondly, converting the characterization vectors into word slot labels and intents through a discriminant model; where the word slot and intent are determined from input information, in the context of the shopping inquiry described above, the word slot labels may include "technology products" and "events" and the like, and the intent may include "purchases" and "returns" and the like.

The present application uses the IOB (in/O u t/B e g I n) form to represent word slot labels, i.e. a plurality of token vectors can be converted into the sequence of word slot labels "O, B-event, I-event, O, B-technology product, I-technology product", where O represents that the word does not belong to any word slot, B represents that the word belongs to and is the first word of the word slot, I represents that the word belongs to and is not the first word of the word slot, and a plurality of token vectors can be converted into the intent "purchase". Finally, instructions are generated based on the word slot labels and the intents. Wherein, based on the word slot label and the intention, it is determined that the instruction generated by the input information is "purchase" and the corresponding scientific and technical product is "iPhone", so that the instruction can be sent to the query system of the shopping website for the operation of "putting the iPhone into the shopping cart".

For another example, suppose "open apple release meeting today, i want to buy a new mobile phone" appears in the discrimination scene of the discrimination model, since apple has two meanings in practical application: 1. "apple (science and technology company)", 2 "apple (fruit)", so that the traditional semantic recognition method is difficult to correctly recognize.

It follows that the recognition capability of the model is very dependent on the background knowledge, and therefore the model performance is poor in scenes lacking background knowledge. In the method, because the graph embedding model can use the associated concept information of the semantic web, the similarity of the graph embedding models of the semantic web of apple (science and technology company) and release party can be utilized, so that the discriminant model also obtains the background knowledge information after the graph embedding model is added, and words which cannot be distinguished literally can be correctly modeled and inferred, namely, the generalization capability of the discriminant model is improved by combining the graph embedding model.

In summary, compared with the static mapping relationship of 'word to vector' and the mapping relationship of only considering the context and not considering the semantic web information, the method of the application converts the word into the representation vector according to the word and the semantic gateway link and the associated concept information of the context concept thereof, can provide richer information, and can convert the representation vector into more accurate word slot labels and intentions, thereby improving the overall performance of the instruction generation system.

Fig. 2 is a schematic block diagram of a semantic web-based query instruction generating apparatus according to an embodiment of the present application. As shown, the apparatus 200 includes:

an obtaining module 201, configured to obtain input information of a query;

a first conversion module 202, configured to convert each word into a representation vector according to semantic web association of context information corresponding to each word in the input information and associated concept information;

the second conversion module 203 is configured to convert the representation vectors corresponding to the words into word slot labels, and obtain intentions according to word slot transitions corresponding to all the words;

and the instruction generating module 204 is used for producing a corresponding instruction according to the intention.

It should be noted that, because the contents of information interaction, execution process, and the like between the modules/units of the apparatus are based on the same concept as the method embodiment described in the present application, the technical effect brought by the contents is the same as the method embodiment of the present application, and specific contents may refer to the description in the foregoing method embodiment of the present application, and are not described herein again.

It should be further noted that the division of the modules of the above apparatus is only a logical division, and the actual implementation may be wholly or partially integrated into one physical entity, or may be physically separated. And these units can be implemented entirely in software, invoked by a processing element; or may be implemented entirely in hardware; and part of the modules can be realized in the form of calling software by the processing element, and part of the modules can be realized in the form of hardware. For example, the first conversion module 202 may be a separate processing element, or may be implemented by being integrated into a chip of the apparatus, or may be stored in a memory of the apparatus in the form of program code, and a processing element of the apparatus calls and executes the functions of the first conversion module 202. Other modules are implemented similarly. In addition, all or part of the modules can be integrated together or can be independently realized. The processing element described herein may be an integrated circuit having signal processing capabilities. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in the form of software.

For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or one or more microprocessors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), among others. For another example, when one of the above modules is implemented in the form of a Processing element scheduler code, the Processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

Fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present invention. As shown, the computer device 300 includes: a memory 301, and a processor 302; the memory 301 is used for storing computer instructions; the processor 302 executes computer instructions to implement the method described in fig. 1.

In some embodiments, the number of the memories 301 in the computer device 300 may be one or more, the number of the processors 302 may be one or more, and fig. 3 illustrates one example.

In an embodiment of the present application, the processor 302 in the computer device 300 loads one or more instructions corresponding to processes of an application program into the memory 301 according to the steps described in fig. 1, and the processor 302 executes the application program stored in the memory 301, thereby implementing the method described in fig. 1.

The Memory 301 may include a Random Access Memory (RAM), and may also include a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. The memory 301 stores an operating system and operating instructions, executable modules or data structures, or a subset thereof, or an expanded set thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic services and for handling hardware-based tasks.

The Processor 302 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the device can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component.

In some specific applications, the various components of the computer device 300 are coupled together by a bus system that may include a power bus, a control bus, a status signal bus, etc., in addition to a data bus. But for clarity of explanation the various buses are referred to in figure 3 as a bus system.

In an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the method described in fig. 1.

The computer-readable storage medium, as will be appreciated by one of ordinary skill in the art: the embodiment for realizing the functions of the system and each unit can be realized by hardware related to computer programs. The aforementioned computer program may be stored in a computer readable storage medium. When the program is executed, the embodiment including the functions of the system and the units is executed; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

The application effectively overcomes various defects in the prior art and has high industrial utilization value.

The above embodiments are merely illustrative of the principles and utilities of the present application and are not intended to limit the invention. Any person skilled in the art can modify or change the above-described embodiments without departing from the spirit and scope of the present application. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present application.

Claims

1. A query instruction generation method based on a semantic network is characterized by comprising the following steps:

acquiring input information of query;

converting each word into a representation vector according to semantic web association and associated concept information of context information corresponding to each word in the input information;

converting the characterization vectors corresponding to the words into word slot labels, and obtaining intentions according to the word slot labels corresponding to all the words; producing corresponding instructions according to the intents.

2. The method according to claim 1, wherein converting each word into a characterization vector according to semantic web association and associated concept information of context information corresponding to each word in the input information comprises:

modeling the semantics and background knowledge of each word based on the context information and semantic web information of each word in the input information to form a graph embedding model;

and training semantic web association according to the context information and associating the concept information through the graph embedding model to form a characterization vector.

3. The method of claim 2, wherein the method comprises:

the graph embedding model is trained by using associated concept information of the semantic net in an unsupervised mode, so that the graph embedding model can use a large amount of semantic net data in the open field.

4. The method of claim 2 or 3, wherein the graph embedding model comprises: any one of a Deepwalk model, a Node2Vec model, a HARP model, and a Walklets model.

5. The method of claim 2, wherein converting the token vectors corresponding to the words into word slot labels and obtaining the intention according to the word slot labels corresponding to all the words comprises:

converting the characterization vector into the word slot tag sequence and the intention sequence through a discriminant model; the distinguishing model obtains the semantics and background knowledge of each word through the graph embedding model so as to model and reason words, and converts the representation vectors corresponding to the words into corresponding word slot labels and corresponding intentions according to the modeling and the reasoning of the distinguishing model.

6. The method of claim 5, wherein the discriminant model comprises: a word slot label classifier and an intent classifier to convert the token vector into a word slot label and an intent, respectively.

7. A semantic web-based query instruction generation apparatus, comprising:

the acquisition module is used for acquiring the input information of the query;

the first conversion module is used for converting each word into a representation vector according to semantic web association of context information corresponding to each word in the input information and associated concept information;

the second conversion module is used for converting the representation vectors corresponding to the words into word slot labels and obtaining intentions according to word slot transitions corresponding to all the words;

and the instruction generating module is used for producing a corresponding instruction according to the intention.

8. A computer device, the device comprising: a memory, and a processor; the memory is to store computer instructions; the processor executes computer instructions to implement the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer instructions which, when executed, perform the method of any one of claims 1 to 6.