Knowledge learning method and knowledge learning device of intelligent agent, electronic equipment and medium

Abstract

The invention discloses a knowledge learning method and device of an intelligent agent, electronic equipment and a computer readable storage medium, and belongs to the technical field of artificial intelligence. The knowledge learning method comprises the steps of receiving dialogue information; performing knowledge classification according to the dialogue information to obtain any one or more of concept type knowledge, scene type knowledge, shape type knowledge and skill type knowledge; respectively storing and updating each type of knowledge according to different types of knowledge; wherein the concept type knowledge is stored in a network ontology language form; the scene type knowledge is stored in an analytic graph form; shape type knowledge is stored in a picture form; skill type knowledge is stored in a text form. In this way, efficient knowledge management and updating can be achieved, and therefore the learning ability and intelligence of the intelligent agent are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly relates to a knowledge learning method, a knowledge learning device, an electronic device, and a computer-readable storage medium for an intelligent agent. Background Art

[0002] An intelligent agent refers to an intelligent system that can interact with the environment and people to achieve specific goals. With the development of technology, intelligent agents have been widely used in many fields such as smart home, intelligent customer service, autonomous driving, and industrial robots. However, in the interaction process, an intelligent agent will receive a large amount of information. How to effectively process and utilize this information to improve the learning ability and intelligence of the intelligent agent has become an urgent problem to be solved. Summary of the Invention

[0003] Embodiments of this application provide a knowledge learning method, a knowledge learning device, an electronic device, and a computer-readable storage medium for an intelligent agent to solve at least one of the above-mentioned technical problems.

[0004] The knowledge learning method for an intelligent agent according to the embodiments of this application includes: Receiving conversation information; Classifying knowledge according to the conversation information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; Storing and updating each type of knowledge separately according to the different types of knowledge; Among them, for the concept type knowledge, it is stored in the form of Web Ontology Language; for the scenario type knowledge, it is stored in the form of an analysis graph; for the shape type knowledge, it is stored in the form of a picture; for the skill type knowledge, it is stored in the form of text.

[0005] In some embodiments, the storing and updating each type of knowledge separately according to the different types of knowledge includes: For the concept type knowledge, obtaining triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; Invoking a link service to respectively determine whether the subject parameter, the predicate parameter, and the object parameter exist in the Web Ontology Language library; If not, updating the Web Ontology Language library according to the concept type knowledge.

[0006] In some embodiments, the storing and updating each type of knowledge separately according to the different types of knowledge includes: For the scenario type knowledge, obtaining triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; Invoke the anchoring module to determine whether the subject parameter exists in the parsing knowledge library; If it exists, determine whether the attribute corresponding to the scenario type knowledge in the parsing knowledge library is an unknown object. If it is an unknown object, update the attribute according to the concept type knowledge; If it does not exist, update the parsing knowledge library according to the scenario type knowledge.

[0007] In some embodiments, the storing and updating of each type of knowledge according to different types of knowledge includes: For the shape type knowledge, calculate the shape similarity between the current shape type knowledge and the existing shape type knowledge in the shape library; Determine whether the shape similarity is less than a first predetermined similarity; When the shape similarity is less than the first predetermined similarity, invoke the matting module to determine the matting result based on the current shape type knowledge; Update the shape library according to the matting result.

[0008] In some embodiments, the storing and updating of each type of knowledge according to different types of knowledge includes: For the skill type knowledge, calculate the skill similarity between the current skill type knowledge and the existing skill type knowledge in the skill library; Determine whether the skill similarity is less than a second predetermined similarity; When the skill similarity is less than the second predetermined similarity, update the skill library according to the skill type knowledge.

[0009] In some embodiments, the parsing knowledge library is used to send the scenario type knowledge to the memory module, and the web ontology language library, the shape library, and the skill library are respectively used to send the concept type knowledge, the shape type knowledge, and the skill type knowledge to the remote database. The knowledge learning method further includes: Actively read the scenario type knowledge from the memory module and actively read the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database through the cognitive module to update the task planning strategy of the cognitive module; and / or Passively receive the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database through the visual perception module and the natural language generation module respectively to update the visual recognition strategy of the visual perception module and the dialogue strategy of the natural language generation module.

[0010] In some embodiments, the knowledge learning method further includes: Receive the parsed graph information generated by the visual perception module; Update the parsed graph library according to the parsed graph information.

[0011] The knowledge learning device of the agent according to the embodiment of the present application includes: An information receiving module, configured to receive dialogue information; A knowledge classification module, configured to classify knowledge according to the dialogue information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; A knowledge update module, configured to respectively perform storage update for each type of knowledge according to different types of knowledge; Among them, for the concept type knowledge, it is stored in the form of web ontology language; for the scenario type knowledge, it is stored in the form of a parsed graph; for the shape type knowledge, it is stored in the form of a picture; for the skill type knowledge, it is stored in the form of text.

[0012] The electronic device according to the embodiment of the present application, the electronic device includes one or more processors and a memory, the memory stores a computer program, and when the computer program is executed by the processor, the knowledge learning method of any one of the above embodiments is implemented.

[0013] The computer-readable storage medium according to the embodiment of the present application, on which a computer program is stored, and when the program is executed by a processor, the knowledge learning method of any one of the above embodiments is implemented.

[0014] The knowledge learning method, knowledge learning device, electronic device, and computer-readable storage medium of the agent according to the embodiment of the present application classify knowledge according to dialogue information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge, and then respectively perform storage update for each type of knowledge according to different types of knowledge. In this way, efficient knowledge management and update can be achieved, thereby improving the learning ability and intelligence of the agent.

[0015] The additional aspects and advantages of the embodiments of the present application will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the embodiments of the present application. Description of the Drawings

[0016] The above and / or additional aspects and advantages of the present application will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, where: Figure 1 is a schematic flowchart of the knowledge learning method of the agent according to some embodiments of the present application; Figure 2It is a schematic framework diagram of the knowledge learning method of the agent in some embodiments of the present application; Figure 3 It is a schematic diagram for knowledge classification of the knowledge learning method of the agent in some embodiments of the present application; Figure 4 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 5 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 6 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 7 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 8 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 9 It is a schematic flowchart of the knowledge learning method of the agent in some embodiments of the present application; Figure 10 It is a schematic module diagram of the knowledge learning device of the agent in some embodiments of the present application; Figure 11 It is a schematic module diagram of the electronic device in some embodiments of the present application; Figure 12 It is a schematic diagram of the connection state between the computer-readable storage medium and the processor in some embodiments of the present application.

[0017] Explanation of reference numerals: Knowledge learning device 100 of the agent, information receiving module 10, knowledge classification module 20, knowledge updating module 30, cognitive module 40, visual perception module 50, natural language generation module 60, electronic device 200, processor 210, memory 220, computer-readable storage medium 300, computer program 310, processor 320. Detailed implementation manners

[0018] The following further describes the embodiments of the present application with reference to the accompanying drawings. The same or similar reference numerals in the drawings represent the same or similar elements or elements with the same or similar functions from beginning to end. In addition, the embodiments of the present application described below with reference to the accompanying drawings are exemplary and are only used to explain the embodiments of the present application, and should not be construed as a limitation of the present application.

[0019] Please refer to Figures 1 to 3 , the knowledge learning method of the agent in the embodiments of the present application includes: 010: Receive conversation information; 020: Classify knowledge according to the conversation information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; 030: Store and update each type of knowledge separately according to the different types of knowledge; Among them, for concept type knowledge, it is stored in the form of Web Ontology Language; for scenario type knowledge, it is stored in the form of an analysis graph; for shape type knowledge, it is stored in the form of a picture; for skill type knowledge, it is stored in the form of text.

[0020] The knowledge learning method of the agent in the embodiment of the present application classifies knowledge according to the conversation information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge, and then stores and updates each type of knowledge separately according to the different types of knowledge. In this way, efficient knowledge management and update can be achieved, thereby improving the learning ability and intelligence of the agent.

[0021] Specifically, an agent refers to an intelligent system that can interact with the environment and people to achieve specific goals. An agent can be a software program, a robot, or other entities with intelligent behaviors.

[0022] Agents can be applied to many fields such as smart homes, intelligent customer service, autonomous driving, and industrial robots. For example, when an agent is applied to a smart home, the agent can automatically adjust household appliances by sensing the environment (such as temperature, light) and user instructions. When an agent is applied to intelligent customer service, the agent can understand user questions through natural language processing and provide accurate answers. When an agent is applied to autonomous driving, the agent can make driving decisions by sensing the road environment (such as traffic signs, pedestrians, other vehicles). When an agent is applied to an industrial robot, the agent can perform complex tasks by sensing the production line status (such as material position, equipment status).

[0023] An agent will receive a large amount of information during the interaction process. How to effectively process and utilize this information to improve the learning ability and intelligence of the agent has become an urgent problem to be solved.

[0024] In the knowledge learning method of the agent in the embodiment of the present application, in 010, receive conversation information.

[0025] Conversation information refers to the information obtained by the agent through interaction with the user or other agents. This information can be natural language conversations, instructions, feedback, etc., reflecting the user's needs, intentions, and the interaction between the agent and other entities.

[0026] Such as Figure 2As shown, the dialogue information may include the speech information input by the auditory perception module (which can be further converted into text information as needed), or the text information input by the text input module, that is, the words spoken / written by others; in addition, the dialogue information may also include the words spoken by the agent itself. In one example, the dialogue information may specifically be dialogue context information. Dialogue context information refers to all information related to the current dialogue during the dialogue process, including but not limited to the current dialogue content, historical dialogue content, etc. Dialogue context information helps the agent learn knowledge, thereby better understanding the user's intention and providing more intelligent services.

[0027] In 020, knowledge classification is performed based on the dialogue information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; In 030, the storage and update of each type of knowledge are performed separately according to the different types of knowledge; Among them, for concept type knowledge, it is stored in the form of Web Ontology Language; for scenario type knowledge, it is stored in the form of an analysis graph; for shape type knowledge, it is stored in the form of a picture; for skill type knowledge, it is stored in the form of text.

[0028] As Figure 2 shown, knowledge classification can be performed based on the dialogue information by a knowledge classification model. The knowledge classification model is, for example, a large model (Large Language Model, LLM). A large model refers to a deep learning model that has undergone large-scale pre-training. By pre-training on a large amount of data, the large model learns the patterns and structures of language and can accurately classify knowledge. The large model can specifically be a Generative Pre-trained Transformer (GPT) model, a Bidirectional Encoder Representations from Transformers (BERT) model, an Enhanced Representation through kNowledge IntEgration (ERNIE) model, a Large Language Model Application (LLaMA) model, etc.

[0029] Please combine Figure 3 , the knowledge classification model can classify the knowledge in the dialogue information into the following four categories: (1) Concept type knowledge Concept type knowledge is also known as Terminology Box (Tbox) type knowledge. Tbox type knowledge refers to knowledge related to concepts and relationships. Tbox type knowledge can be stored in the form of Web Ontology Language (OWL), specifically in an OWL library. Among them, OWL is a language for constructing and representing ontologies, mainly used to describe concepts (Class), properties (Property), and the relationships between them, etc.

[0030] Tbox type knowledge can be used for subsequent agent dialogue generation, visual recognition, task planning, etc. For example, in terms of dialogue generation, Tbox type knowledge provides the semantic basis for concepts and relationships, which can help agents understand users' intentions and generate accurate answers; in terms of visual recognition, through knowledge of concepts and relationships, agents can better identify and classify objects in images; in terms of task planning, Tbox type knowledge provides a conceptual framework for task planning, which can help agents understand the semantics and logic of tasks.

[0031] (2) Scenario type knowledge Scenario type knowledge is also known as Assertion Box (Abox) type knowledge. Abox type knowledge refers to knowledge related to individual instances and scenarios. Abox type knowledge can be stored in the form of a Parse Graph (PG), specifically in a PG library. Among them, PG is a graphical knowledge representation form used to represent the structured information of sentences or scenarios, etc.

[0032] Abox type knowledge can be used for subsequent agent visual recognition and scenario understanding. For example, in terms of visual recognition, Abox type knowledge provides information about individual instances in a specific scenario, which can help agents identify and understand specific objects and their properties in images; in terms of scenario understanding, through Abox type knowledge, agents can better understand specific objects and their mutual relationships in a scenario.

[0033] (3) Shape type knowledge Shape type knowledge refers to knowledge related to shapes. Shape type knowledge can be stored in the form of base64 binary images, specifically in a shape library.

[0034] Shape type knowledge can be used for subsequent agent shape learning and visual recognition. For example, in terms of shape learning, agents can identify and classify objects of different shapes by learning shape type knowledge; in terms of visual recognition, shape type knowledge can help agents identify and match specific shapes in images.

[0035] (4) Skill type knowledge Skill type knowledge refers to knowledge related to task processes and execution steps. Skill type knowledge can be stored in text form, specifically in a skill library.

[0036] Skill type knowledge can be used for subsequent task planning and skill learning of the agent. For example, in terms of task planning, skill type knowledge provides the agent with specific steps and strategies to complete tasks, helping it plan and execute tasks; in terms of skill learning, the agent can master new task execution capabilities by learning skill type knowledge.

[0037] It should be noted that in the implementation mode of this application, knowledge is divided into the above Tbox type knowledge, Abox type knowledge, shape type knowledge, and skill type knowledge. However, in actual applications, the knowledge existing in the dialogue information may only include any one or more of the above types of knowledge. Regardless of which types of knowledge are included, subsequent storage and updates of each type of knowledge are performed separately according to the different types of knowledge. In addition, the above OWL form, PG form, picture form, and text form respectively represent different data structures for corresponding storage of different types of knowledge.

[0038] In this way, by dividing knowledge into four categories: Tbox type knowledge, Abox type knowledge, shape type knowledge, and skill type knowledge, and storing them in different data structures according to different types of knowledge for knowledge update, the agent can manage and utilize knowledge more efficiently. This strategy not only improves the organization and scalability of knowledge but also enhances the agent's capabilities in aspects such as perception, cognition, dialogue generation, and task execution.

[0039] After knowledge classification, the agent's knowledge learning method may further include: respectively determining whether each type of knowledge needs to be learned to decide which part of the knowledge to update. That is to say, when it is determined that a certain type of knowledge needs to be learned, then update that type of knowledge. In this way, the agent can update different types of knowledge according to the different situations of the knowledge existing in the dialogue information, so as to better adapt to environmental changes, understand user needs, and generate more accurate and natural responses.

[0040] Please refer to Figure 2 and Figure 4 , in some implementation modes, storing and updating each type of knowledge separately according to the different types of knowledge (i.e., 030) includes: 031: For concept type knowledge, obtain triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; 032: Call the link service to respectively determine whether the subject parameter, predicate parameter, and object parameter exist in the Web Ontology Language library; 033: If not, update the Web Ontology Language library according to the concept type knowledge.

[0041] Specifically, for the Tbox type knowledge, first use a large model (such as the aforementioned GPT, BERT, ERNIE, etc.) to extract triple information in the form of SPO (subject, predicate, object) from the conversation information. The definition of SPO is as follows: subject: The subject in a sentence, usually the object that performs an action or is being described.

[0042] predicate: The predicate in a sentence, describing the action or state of the subject.

[0043] object: The object in a sentence, usually the recipient of an action or the content being described.

[0044] The following introduces the linking service: The linking service is the Linking service, whose function is to associate text information with concept or property information in the OWL library. The input of the linking service is text information, and the output of the linking service is the concept or property information in the OWL library corresponding to the text information.

[0045] When the linking service works, it can first process the input text information to extract key information; then search for concept or property information in the OWL library that matches the text information, which may involve processes such as semantic search and ontology matching; finally, return the concept or property information that best matches the text information.

[0046] In the implementation manner of this application, by calling the linking service, it can be determined whether the incoming subject parameter, predicate parameter, and object parameter already exist in the OWL library, so as to decide whether to update the Tbox type knowledge. For example, respectively determine whether the subject parameter, predicate parameter, and object parameter already exist in the OWL library. If not, update the OWL library according to the characteristics of the Tbox type knowledge. The characteristics of the Tbox type knowledge may include the aforementioned subject parameter, predicate parameter, and object parameter, as well as the relationship information between these parameters, etc.

[0047] In some embodiments, when the subject parameter and the object parameter do not exist in the OWL library, the concept information in the OWL library can be updated according to the subject parameter and the object parameter. When the predicate parameter does not exist in the OWL library, the attribute information in the OWL library can be updated according to the predicate parameter. When the subject parameter, the predicate parameter, and the object parameter all do not exist in the OWL library, the concept information in the OWL library can be updated according to the subject parameter and the object parameter, the attribute information in the OWL library can be updated according to the predicate parameter, and according to the relationship information between the subject parameter, the predicate parameter, and the object parameter, the relationship information between the subject parameter and the object parameter in the OWL library can be updated, such as subClassOf (indicating that one class is a subclass of another class).

[0048] Please refer to Figure 2 and Figure 5 , in certain embodiments, according to different types of knowledge, each type of knowledge is stored and updated separately (i.e., 030), including: 034: For scenario type knowledge, obtain triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; 035: Call the grounding module to determine whether the subject parameter exists in the parsing library; 036: If it exists, determine whether the attribute corresponding to the scenario type knowledge in the parsing library is an unknown object. If it is an unknown object, update the attribute according to the concept type knowledge; 037: If it does not exist, update the parsing library according to the scenario type knowledge.

[0049] Specifically, the explanation of "obtain triple information" in 031 in the foregoing embodiment also applies to "obtain triple information" in 034 of the embodiment of the present application, and will not be elaborated here.

[0050] The grounding module is introduced below: The grounding module, i.e., the Grounding module, is used to associate text information with specific instances (instances) in the PG library. The input of the grounding module is text information, and the output of the grounding module is the instance identifier (instance_id) information in the PG library corresponding to the text information.

[0051] When the grounding module works, it can first process the input text information to extract key information; then search for specific instances matching the text information in the PG library, which may involve processes such as semantic matching and context analysis; finally, return the instance identifier information that best matches the text information.

[0052] In the embodiment of the present application, by calling the grounding module, it can be determined whether the incoming subject parameter already exists in the PG library, so as to decide whether to update the Abox type knowledge.

[0053] In some embodiments, if the subject parameter already exists in the PG library, indicating that the agent visually sees an object, it is further determined whether the rdf:type (indicating that an instance belongs to a certain class) attribute of this instance of the Abox type knowledge in the PG library is an unknown object. If it is an unknown object, it means that although the agent sees an object, it does not recognize the object, and then the rdf:type attribute is updated according to the Tbox type knowledge learned from the dialogue or known, so that the agent can recognize the object visually. If it is not an unknown object, it means that the agent already recognizes the object and there is no need to update the Abox type knowledge.

[0054] If the subject parameter does not exist in the PG library, it means that the agent does not visually see an object related to this Abox type knowledge, but has learned knowledge related to it through the above dialogue. For example, the agent does not see a stool placed under the table, but someone tells it that there is a stool under the table. At this time, the agent will add a virtual instance to the PG library according to the Abox type knowledge and update the relevant location information to the PG library.

[0055] Please refer to Figure 2 and Figure 6 , in some embodiments, the storage and update of each type of knowledge are performed separately according to different types of knowledge (i.e., 030), including: 038: For shape type knowledge, calculate the shape similarity between the current shape type knowledge and the existing shape type knowledge in the shape library; 039: Determine whether the shape similarity is less than the first predetermined similarity; 040: When the shape similarity is less than the first predetermined similarity, call the matte extraction module to determine the matte extraction result based on the current shape type knowledge; 041: Update the shape library according to the matte extraction result.

[0056] Specifically, for shape type knowledge, by calculating the shape similarity between the current shape type knowledge and the existing shape type knowledge in the shape library, it is decided whether to update the shape library. In this way, the integrity and accuracy of the shape library can be ensured, and at the same time, the duplication of storing similar shapes can be avoided.

[0057] Among them, the shape similarity can be the similarity of shape embedding vectors. First, generate the current shape embedding vector corresponding to the current shape type knowledge through a pre-trained language model, and generate the existing shape embedding vector corresponding to the existing shape type knowledge. Then, calculate the shape embedding vector similarity between the current shape embedding vector and the existing shape embedding vector. In this way, by converting shape knowledge into numerical vectors, it is convenient to calculate similarity.

[0058] The first predetermined similarity can be set artificially in advance. Taking the first predetermined similarity as 0.7 as an example, if the shape similarity is greater than or equal to 0.7, it is considered that the current shape is similar to the existing shape, and there is no need to update the shape library; if the shape similarity is less than 0.7, it means that the current shape is quite different from the existing shape, and the current shape type knowledge needs to be added to the shape library to update the shape library.

[0059] The following introduces the matte extraction module: The function of the matte extraction module is to extract the object information that the user is interested in from the visual picture of the current scene according to the user's instructions or descriptions. The input of the matte extraction module is the visual picture that the agent currently sees, and the output of the matte extraction module is the object information that the user is interested in extracted from the visual picture. For example, there are many things in the room. Someone points to the letter L written on the blackboard and says to the agent "It is a capital letter L", then the matte extraction module will extract the letter L as the result output.

[0060] When the matte extraction module works, it can first use computer vision technology to identify each object in the visual picture of the current scene; then, according to the user's instructions or descriptions, extract the corresponding object information from the visual picture, which may involve processes such as target localization and image segmentation; finally, return the extracted object information as the result.

[0061] In the implementation manner of this application, when it is determined that the shape library needs to be updated, the visual pictures of the relevant frames seen by the agent can be extracted from the current shape type knowledge in combination with the perception time corresponding to the agent recorded in the PG library. Then, the extracted visual pictures are input into the matte extraction module. The matte extraction module will extract the graphic information that the user is interested in from the visual pictures according to the user's pointing or description, and store the extracted graphic information in the shape library in the form of a binary picture. In addition, the dialogue information related to the graphic (such as the user's description of the graphic) can also be stored in the text library in text form.

[0062] Please refer to Figure 2 and Figure 7 , in some implementation manners, the storage and update of each type of knowledge are performed separately according to different types of knowledge (that is, 030), including: 042: For the skill type knowledge, calculate the skill similarity between the current skill type knowledge and the existing skill type knowledge in the skill library; 043: Determine whether the skill similarity is less than a second predetermined similarity; 044: When the skill similarity is less than the second predetermined similarity, update the skill library according to the skill type knowledge.

[0063] Specifically, for the skill type knowledge, by calculating the skill similarity between the current skill type knowledge and the existing skill type knowledge in the skill library, it is determined whether the skill library needs to be updated. In this way, the integrity and accuracy of the skill library can be ensured, and at the same time, the duplication of storing similar skills can be avoided.

[0064] Among them, the skill similarity can be the skill embedding vector similarity. First, generate the current skill embedding vector corresponding to the current skill type knowledge through a pre-trained language model, and generate the existing skill embedding vector corresponding to the existing skill type knowledge. Then, calculate the skill embedding vector similarity between the current skill embedding vector and the existing skill embedding vector. In this way, by converting the skill knowledge into a numerical vector, it is convenient to calculate the similarity.

[0065] The second predetermined similarity can be set artificially in advance. Taking the second predetermined similarity as 0.8 as an example, if the skill similarity is greater than or equal to 0.8, it is considered that the current skill is similar to the existing skill and the skill library does not need to be updated; if the skill similarity is less than 0.8, it means that the current skill is quite different from the existing skill, and the current skill type knowledge needs to be added to the skill library to update the skill library.

[0066] Please refer to Figure 2 、 Figure 8 and Figure 10 In some embodiments, the parsing gallery is used to send the scene type knowledge to the memory module. The web ontology language library, the shape library, and the skill library are respectively used to send the concept type knowledge, the shape type knowledge, and the skill type knowledge to the remote database. The knowledge learning method further includes: 050: Actively read the scene type knowledge from the memory module through the cognitive module 40, and actively read the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database to update the task planning strategy of the cognitive module 40; and / or 060: Passively receive the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database through the visual perception module 50 and the natural language generation module 60 respectively to update the visual recognition strategy of the visual perception module 50 and the dialogue strategy of the natural language generation module 60.

[0067] Specifically, the PG library transfers Abox-type knowledge to the memory module, the OWL library transfers Tbox-type knowledge to the remote database, the shape library transfers shape-type knowledge to the remote database, and the skill library transfers skill-type knowledge to the remote database. In this way, the learned knowledge is stored in the memory module and the remote database, which can not only be used to update the subsequent task planning strategy, visual recognition strategy, and dialogue strategy, but also realize the long-term memory of the agent.

[0068] For example, the cognitive module 40 can actively read Abox-type knowledge from the memory module and actively read Tbox-type knowledge, shape-type knowledge, and skill-type knowledge from the remote database to update the task planning strategy.

[0069] For another example, according to the callback function of the remote database, the visual perception module 50 can passively receive Tbox-type knowledge, shape-type knowledge, and skill-type knowledge from the remote database to real-time update the visual recognition strategy; the Natural Language Generation (NLG) module 60 can passively receive Tbox-type knowledge, shape-type knowledge, and skill-type knowledge from the remote database to real-time update the dialogue strategy.

[0070] After the agent restarts, it can also actively read the stored knowledge from the memory module and the remote database, so that the agent still remembers the knowledge it has learned before, maintaining its intelligence and learning ability.

[0071] Please refer to Figure 2 、 Figure 9 and Figure 10 , in some embodiments, the knowledge learning method further includes: 070: Receive the parsed graph information generated by the visual perception module 50; 080: Update the parsed graph library according to the parsed graph information.

[0072] Specifically, the visual perception module 50 is also the Computer Vision (CV) module. The visual perception module 50 can adopt a visual recognition model. The visual perception module 50 obtains image data through a camera or other visual sensors. The PG information can be the structured data generated after the visual perception module 50 analyzes and understands the image data. The PG information is a kind of perception information, which provides a structured description of specific instances in the scene and can be used to assist the aforementioned knowledge update module 30 to classify knowledge into the aforementioned Abox-type knowledge, Tbox-type knowledge, shape-type knowledge, and skill-type knowledge.

[0073] In the embodiments of the present application, first, the PG information generated by the visual perception module 50 is received, and then the PG library is updated according to the PG information. In this way, the PG library integrates the PG information and the Abox type knowledge, which can combine the perception information and the cognitive knowledge, and with the change of the environment, through dialogue and communication, continuously update and learn the self-cognitive knowledge, thereby improving the perception ability and cognitive ability of the intelligent agent.

[0074] In the embodiments of the present application, after the PG library is updated according to the PG information, the cognitive module 40 can also actively read the PG information from the memory module to further update the task planning strategy of the cognitive module 40.

[0075] Please refer to Figure 2 and Figure 10 For the knowledge learning device 100 of the intelligent agent in the embodiments of the present application, it includes an information receiving module 10, a knowledge classification module 20, and a knowledge update module 30. The information receiving module 10 is used to receive the dialogue information. The knowledge classification module 20 is used to classify the knowledge according to the dialogue information to obtain any one or more of the concept type knowledge, scene type knowledge, shape type knowledge, and skill type knowledge. The knowledge update module 30 is used to store and update each type of knowledge according to different types of knowledge. Among them, for the concept type knowledge, it is stored in the form of the Web Ontology Language; for the scene type knowledge, it is stored in the form of an analysis graph; for the shape type knowledge, it is stored in the form of a picture; for the skill type knowledge, it is stored in the form of text.

[0076] In some embodiments, the knowledge update module 30 is specifically used for: for the concept type knowledge, obtaining triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; calling a link service to respectively determine whether the subject parameter, the predicate parameter, and the object parameter exist in the Web Ontology Language library; if not, updating the Web Ontology Language library according to the concept type knowledge.

[0077] In some embodiments, the knowledge update module 30 is specifically used for: for the scene type knowledge, obtaining triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; calling an anchoring module to determine whether the subject parameter exists in the analysis graph library; if it exists, determining whether the attribute corresponding to the scene type knowledge in the analysis graph library is an unknown object, and if it is an unknown object, updating the attribute according to the concept type knowledge; if it does not exist, updating the analysis graph library according to the scene type knowledge.

[0078] In some embodiments, the knowledge update module 30 is specifically configured to: for the shape type knowledge, calculate the shape similarity between the current shape type knowledge and the existing shape type knowledge in the shape library; determine whether the shape similarity is less than a first predetermined similarity; when the shape similarity is less than the first predetermined similarity, call the matte extraction module to determine the matte extraction result based on the current shape type knowledge; and update the shape library according to the matte extraction result.

[0079] In some embodiments, the knowledge update module 30 is specifically configured to: for the skill type knowledge, calculate the skill similarity between the current skill type knowledge and the existing skill type knowledge in the skill library; determine whether the skill similarity is less than a second predetermined similarity; and when the skill similarity is less than the second predetermined similarity, update the skill library according to the skill type knowledge.

[0080] In some embodiments, the parsing image library is used to send the scene type knowledge to the memory module. The web ontology language library, the shape library, and the skill library are respectively used to send the concept type knowledge, the shape type knowledge, and the skill type knowledge to the remote database. The knowledge learning device 100 further includes a cognitive module 40. The cognitive module 40 is configured to actively read the scene type knowledge from the memory module, and actively read the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database to update the task planning strategy of the cognitive module 40; and / or the knowledge learning device 100 further includes a visual perception module 50 and a natural language generation module 60. The visual perception module 50 and the natural language generation module 60 respectively passively receive the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database to update the visual recognition strategy of the visual perception module 50 and the dialogue strategy of the natural language generation module 60.

[0081] In some embodiments, the information receiving module 10 is further configured to receive the parsed image information generated by the visual perception module 50. The knowledge update module 30 is further configured to update the parsing image library according to the parsed image information.

[0082] It should be noted that the explanations of the knowledge learning method of the intelligent agent in the foregoing embodiments are equally applicable to the knowledge learning device 100 of the intelligent agent in the embodiments of the present application, and will not be elaborated herein. In addition, the knowledge learning device 100 of the intelligent agent in the embodiments of the present application may belong to a part of the intelligent agent or exist independently of the intelligent agent.

[0083] Please refer to Figure 11 , the electronic device 200 in the embodiments of the present application includes one or more processors 210 and a memory 220, and the memory 220 stores a computer program. When the computer program is executed by the processor 210, the knowledge learning method in any of the foregoing embodiments is implemented.

[0084] For example, when a computer program is executed by a processor 210, a knowledge learning method as follows is implemented: 010: Receive conversation information; 020: Classify knowledge according to the conversation information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; 030: Store and update each type of knowledge separately according to the different types of knowledge; Among them, for concept type knowledge, it is stored in the form of Web Ontology Language; for scenario type knowledge, it is stored in the form of an analysis graph; for shape type knowledge, it is stored in the form of a picture; for skill type knowledge, it is stored in the form of text.

[0085] For another example, when a computer program is executed by a processor 210, a knowledge learning method as follows is implemented: 031: For concept type knowledge, obtain triple information, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; 032: Call a linking service to respectively determine whether the subject parameter, the predicate parameter, and the object parameter exist in the Web Ontology Language library; 033: If not, update the Web Ontology Language library according to the concept type knowledge.

[0086] It should be noted that the explanations of the knowledge learning method of the intelligent agent in the foregoing embodiments are equally applicable to the electronic device 200 in the embodiments of the present application, and will not be elaborated herein.

[0087] Please refer to Figure 12 , a computer-readable storage medium 300 in the embodiments of the present application, on which a computer program 310 is stored. When the program is executed by a processor 320, the knowledge learning method in any of the foregoing embodiments is implemented.

[0088] For example, when the program is executed by a processor 320, a knowledge learning method as follows is implemented: 010: Receive conversation information; 020: Classify knowledge according to the conversation information to obtain any one or more of concept type knowledge, scenario type knowledge, shape type knowledge, and skill type knowledge; 030: Store and update each type of knowledge separately according to the different types of knowledge; Among them, for concept type knowledge, it is stored in the form of Web Ontology Language; for scenario type knowledge, it is stored in the form of an analysis graph; for shape type knowledge, it is stored in the form of a picture; for skill type knowledge, it is stored in the form of text.

[0089] For another example, when the program is executed by the processor 320, the following knowledge learning method is implemented: 031: For knowledge of concept types, obtain triple information, which includes a subject parameter, a predicate parameter, and an object parameter; 032: Call the link service to respectively determine whether the subject parameter, the predicate parameter, and the object parameter exist in the Web Ontology Language library; 033: If not, update the Web Ontology Language library according to the knowledge of concept types.

[0090] It should be noted that the explanation of the knowledge learning method of the intelligent agent in the foregoing embodiments also applies to the computer-readable storage medium 300 of the embodiments of the present application, and will not be elaborated here.

[0091] In summary, the knowledge learning method, the knowledge learning device 100, the electronic device 200, and the computer-readable storage medium 300 of the intelligent agent according to the embodiments of the present application classify knowledge according to the dialogue information to obtain any one or more of knowledge of concept types, knowledge of scenario types, knowledge of shape types, and knowledge of skill types, and then respectively store and update each type of knowledge according to the different types of knowledge. In this way, efficient knowledge management and update can be achieved, thereby improving the learning ability and intelligence of the intelligent agent.

[0092] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without conflict, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0093] Any process or method description in the flowchart or described in other ways herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiments of the present application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in an order opposite to that shown or discussed, according to the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0094] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definable sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable storage medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a computer-readable storage medium can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of the computer-readable storage medium include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable storage medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing as necessary, and then stored in a computer memory.

[0095] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0096] Those of ordinary skill in the art can understand that all or part of the steps carried out in the method of the above embodiments can be completed by instructing relevant hardware through a program. The said program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments. In addition, in each of the embodiments of the present application, the functional units can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disc, etc.

[0097] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application. The scope of the present application is defined by the claims and their equivalents.

Claims

1. A knowledge learning method for an intelligent agent, characterized in that: include: Receive conversation information; Classify the knowledge according to the dialogue information to obtain any one or more of concept type knowledge, scene type knowledge, shape type knowledge, and skill type knowledge; According to different types of knowledge, storage and update of each type of knowledge are performed separately; The concept type knowledge is stored in the form of network ontology language; the scene type knowledge is stored in the form of parsing graph; the shape type knowledge is stored in the form of pictures; The skill type knowledge is stored in text form.

2. The knowledge learning method according to claim 1, characterized in that: The storage and updating of each type of knowledge respectively according to different types of knowledge includes: For the concept type knowledge, triple information is obtained, wherein the triple information includes a subject parameter, a predicate parameter, and an object parameter; Calling a link service to respectively determine whether the subject parameter, the predicate parameter, and the object parameter exist in a network ontology language library; If it does not exist, the network ontology language library is updated according to the concept type knowledge.

3. The knowledge learning method according to claim 1, characterized in that: The storage and updating of each type of knowledge respectively according to different types of knowledge includes: For the scene type knowledge, triple information is obtained, where the triple information includes a subject parameter, a predicate parameter, and an object parameter; Calling an anchoring module to determine whether the subject parameter exists in the parsing library; If so, determining whether the attribute corresponding to the scene type knowledge in the parsing library is an unknown object; if so, updating the attribute according to the concept type knowledge; If not, the parsing library is updated according to the scene type knowledge.

4. The knowledge learning method according to claim 1, characterized in that: The storage and updating of each type of knowledge respectively according to different types of knowledge includes: For the shape type knowledge, calculating the shape similarity between the current shape type knowledge and the existing shape type knowledge in the shape library; Determining whether the shape similarity is less than a first predetermined similarity; When the shape similarity is less than the first predetermined similarity, calling a cutout module to determine a cutout result based on current shape type knowledge; The shape library is updated according to the cutout result.

5. The knowledge learning method according to claim 1, characterized in that: The storage and updating of each type of knowledge respectively according to different types of knowledge includes: For the skill type knowledge, calculating the skill similarity between the current skill type knowledge and the existing skill type knowledge in the skill library; Determining whether the skill similarity is less than a second predetermined similarity; When the skill similarity is less than the second predetermined similarity, the skill library is updated according to the skill type knowledge.

6. The knowledge learning method according to any one of claims 1 to 5, characterized in that: The parsing library is used to send the scene type knowledge to the memory module, the network ontology language library, the shape library and the skill library are used to send the concept type knowledge, the shape type knowledge and the skill type knowledge to the remote database respectively, and the knowledge learning method also includes: Actively reading the scene type knowledge from the memory module, and actively reading the concept type knowledge, the shape type knowledge, and the skill type knowledge from the remote database through the cognitive module to update the task planning strategy of the cognitive module; and / or The concept type knowledge, the shape type knowledge and the skill type knowledge are passively received from the remote database through the visual perception module and the natural language generation module respectively to update the visual recognition strategy of the visual perception module and the dialogue strategy of the natural language generation module.

7. The knowledge learning method according to any one of claims 1 to 5, characterized in that: The knowledge learning method also includes: Receiving parsing graph information generated by a visual perception module; The parsing graph library is updated according to the parsing graph information.

8. A knowledge learning device for an intelligent agent, characterized in that: include: An information receiving module, used for receiving dialogue information; A knowledge classification module, used to classify knowledge according to the dialogue information to obtain any one or more of concept type knowledge, scene type knowledge, shape type knowledge, and skill type knowledge; A knowledge updating module is used to update the storage of each type of knowledge according to different types of knowledge; The concept type knowledge is stored in the form of network ontology language; the scene type knowledge is stored in the form of parsing graph; the shape type knowledge is stored in the form of pictures; The skill type knowledge is stored in text form.

9. An electronic device, characterized in that: The electronic device includes one or more processors and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the knowledge learning method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the knowledge learning method described in any one of claims 1 to 7 is implemented.

Images