Method and device for knowledge retrieval and intelligent questioning based on intelligent marketing agent

Abstract

The application relates to a knowledge retrieval and intelligent question method and device based on an intelligent marketing intelligent agent, and relates to the technical field of computers.The method comprises the following steps: processing to-be-processed data of a business domain to obtain entity instances, entity relationship instances and entity attribute instances in the to-be-processed data; clustering all the entity instances through a machine learning model to obtain clustering results; generating a hierarchical structure representing hierarchical relationships between all categories based on the association relationships between the categories in the clustering results through the machine learning model; and generating first knowledge belonging to the business domain based on the hierarchical structure, the clustering results, the entity relationship instances and the entity attribute instances through the machine learning model, wherein the first knowledge comprises category definitions, attribute definitions, relationship definitions and constraint rules, the first knowledge is used for supporting the intelligent marketing intelligent agent to realize knowledge retrieval and intelligent questioning, the semantic understanding capability of the machine learning model is used for completing automatic construction of knowledge, and the effect of knowledge construction is improved.

Description

Technical Field

[0001] This article relates to the field of computer technology, specifically to a knowledge retrieval and intelligent questioning method and device based on intelligent marketing agents. Background Technology

[0002] Intelligent data querying refers to data querying and analysis tools based on artificial intelligence technology. It combines natural language processing and data query engines, enabling users to query, process, and analyze data through natural language interaction. For example, an intelligent agent can invoke a large language model to understand natural language and combine this with knowledge from the corresponding business domain (i.e., the business area) to achieve knowledge retrieval and reasoning decisions, obtaining query or analysis results and returning them to the user. The business domain can be understood as the scope of a business's operations. The core step in achieving this process includes knowledge construction, namely, defining what categories exist in the business domain, what attributes each category possesses, and what relationships exist between categories. Therefore, how to construct this knowledge is a current technical challenge. Summary of the Invention

[0003] This content section is provided to briefly introduce the ideas, which will be described in detail in the examples section later. This content section is not intended to identify key or essential features of the claimed content, nor is it intended to limit the scope of the claimed content.

[0004] Firstly, a knowledge retrieval and intelligent questioning method based on intelligent marketing agents is provided, including: Acquire data to be processed, which belongs to the business domain; The data to be processed is processed to obtain entity instances, entity relationship instances, and entity attribute instances from the data to be processed. All the entity instances are clustered using a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions. The machine learning model generates a hierarchical structure based on the association between categories in the clustering results. The hierarchical structure is used to represent the hierarchical relationship between all the categories. The machine learning model generates first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and the business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

[0005] Secondly, a knowledge retrieval and intelligent questioning device based on an intelligent marketing agent is provided, including: The first acquisition module is used to acquire data to be processed, which belongs to the business domain. The processing module is used to process the data to be processed in order to obtain entity instances, entity relationship instances and entity attribute instances in the data to be processed. The clustering module is used to cluster all the entity instances using a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions. The first generation module is used to generate a hierarchical structure based on the association between categories in the clustering results using the machine learning model. The hierarchical structure is used to represent the hierarchical relationship between all the categories. The second generation module is used to generate first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances using the machine learning model. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and the business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

[0006] Thirdly, an electronic device is provided, comprising: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device to implement the steps of the method described in the first aspect.

[0007] The above method processes the data to be processed to obtain entity instances, entity relationship instances, and entity attribute instances. A machine learning model is then used to cluster all entity instances, yielding clustering results. Entity instances belonging to the same cluster satisfy preset conditions. Based on the relationships between clusters in the clustering results, a hierarchical structure is generated using the machine learning model. This hierarchical structure represents the hierarchical relationships between all categories. Finally, based on the hierarchical structure, clustering results, entity relationship instances, and entity attribute instances, a first knowledge is generated using the machine learning model. This first knowledge includes category definitions, attribute definitions, relationship definitions, and constraint rules within the business domain. At the schema layer representing the knowledge graph, the knowledge graph is obtained by processing business data from the schema layer and the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain. In this way, through the semantic understanding capabilities of the machine learning model, the extraction and reasoning of entity instances, entity relationship instances, and entity attribute instances, category reasoning and induction, hierarchical relationship reasoning between categories, and constraint rule reasoning are realized to complete the automated construction of knowledge, thereby improving the efficiency of knowledge construction. In addition, the constructed intelligent marketing agent can perform knowledge retrieval based on the first knowledge and the knowledge graph as the data foundation, and then realize the corresponding intelligent questioning based on the knowledge retrieval results. Attached Figure Description

[0008] The above and other features, advantages, and aspects of this document will become more apparent when viewed in conjunction with the accompanying drawings and the following examples. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale. In the drawings: Figure 1 These are schematic diagrams illustrating application environments based on various scenarios; Figure 2 This is a flowchart illustrating knowledge retrieval and intelligent questioning methods based on intelligent marketing agents, presented according to certain scenarios. Figure 3 This is a block diagram of a knowledge retrieval and intelligent questioning device based on intelligent marketing agents, shown according to some scenarios; Figure 4 These are schematic diagrams of the structure of electronic devices shown in various scenarios. Detailed Implementation

[0009] The following description will be given in more detail with reference to the accompanying drawings. While certain scenarios are shown in the drawings, it should be understood that this document can be implemented in various forms and should not be construed as limited to the scenarios described herein. Rather, these scenarios are provided to provide a more thorough and complete understanding of this document. It should be understood that the accompanying drawings and the scenarios depicted are for illustrative purposes only and are not intended to limit the scope of this document.

[0010] It should be understood that the steps described in the method may be performed in different orders and / or in parallel. Furthermore, the method may include additional steps and / or omit the steps shown. The scope of this document is not limited in this respect.

[0011] The term "comprising" and its variations can be open-ended, meaning "including but not limited to". The term "based on" can mean "at least partially based on". The term "one case" means "at least one case"; the term "another case" means "at least one additional case"; the term "some cases" means "at least some cases". Definitions of other terms will be given in the following description.

[0012] It should be noted that the concepts of "first" and "second" are only used to distinguish different devices, modules or units, and are not used to limit the order of the functions performed by these devices, modules or units or their interdependencies.

[0013] It should be noted that the modifiers “one” and “multiple” can be illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as “one or more”.

[0014] The names of messages or information exchanged between multiple devices are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0015] It is understandable that the data involved in the technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of relevant regulations.

[0016] Intelligent agents are a core concept in the field of artificial intelligence technology. They mainly refer to software and hardware entities that can autonomously perceive their environment, make independent decisions, and execute actions to achieve preset goals. Intelligent marketing agents can refer to the application of intelligent agents in marketing scenarios to achieve intelligent data collection within those scenarios.

[0017] Machine learning models can refer to mathematical / algorithm programs that can automatically learn patterns from data without requiring manual rule writing.

[0018] Large language models can be a type of machine learning model, also known as large models. They are artificial intelligence systems based on deep learning, and their core capabilities are understanding, reasoning about, and generating natural language.

[0019] Figure 1 These are schematic diagrams illustrating application environments based on various scenarios. (Refer to...) Figure 1Client 101 can provide an interface for tenant 102 to interact with. This interface can at least support tenant 102 in selecting data to be processed and triggering the automated construction of first knowledge within the business domain to which the data belongs. The constructed first knowledge can support the intelligent marketing agent in realizing intelligent data collection within the business domain. Client 101 can be, for example, a mobile phone or a laptop.

[0020] Figure 2 This is a flowchart illustrating a knowledge retrieval and intelligent questioning method based on an intelligent marketing agent, illustrated under certain scenarios. This method is executed by the aforementioned client. (Refer to...) Figure 2 The knowledge retrieval and intelligent questioning method based on intelligent marketing agents may include steps S210, S220, S230, S240 and S250.

[0021] Step S210: Obtain the data to be processed, which belongs to the business domain.

[0022] For enterprise-level tenants, taking a marketing scenario as an example, a tenant typically serves different business systems (e.g., customer relationship management systems and supply chain systems), and these different business systems serve different business domains. The data to be processed can be business data from a specific business domain, and this data can be structured, unstructured, or semi-structured; this document does not impose any limitations on this. As an example, the data to be processed can also be unimodal or multimodal data, with modalities such as text, image, and audio modalities. Furthermore, the data to be processed can include product manuals and customer relationship management data, etc.

[0023] Step S220: Process the data to be processed to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed.

[0024] Entity relationship instances can include relationship type instances, relationship direction instances, and relationship attribute instances. Relationship type instances describe the relationships between entity instances, relationship direction instances describe which entity instance points to another entity instance, and relationship attribute instances describe the attributes possessed by the relationship type instance. For example, if the relationship type instance is cooperation, the attributes of cooperation could be the start time of cooperation and the duration of cooperation, etc. Furthermore, the methods for obtaining entity instances, entity relationship instances, and entity attribute instances can be found below.

[0025] As an example, machine learning models can be leveraged to understand natural language to extract and mine entity instances, entity relationship instances, and entity attribute instances from the data. As another example, rule-based methods can be used to obtain these instances from the data. Compared to the method using machine learning models, rule-based methods may have higher accuracy in specific business domains, while machine learning model-based methods have stronger generalization capabilities. Therefore, different methods for obtaining entity instances, entity relationship instances, and entity attribute instances can be tailored to the characteristics of different business domains. The machine learning models mentioned in this article can be, for example, large language models; explanations and descriptions of large language models can be found above.

[0026] Step S230: Cluster all the entity instances using a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions.

[0027] Clustering here can be achieved by leveraging the understanding capabilities of a large language model to group semantically similar entity instances into the same category. Therefore, the aforementioned preconditions can include that the semantic similarity between entity instances belonging to the same category is greater than a predefined semantic similarity. For example, grouping "face cream A" and "serum B" into a skincare product category can be inferred by the large language model based on its understanding of the relevant descriptions of "face cream A" and "serum B".

[0028] The clustering results are used to describe multiple categories, each of which includes entity instances with a semantic similarity greater than a preset semantic similarity.

[0029] Step S240: Based on the association between categories in the clustering results, the machine learning model generates a hierarchical structure, which is used to represent the hierarchical relationship between all the categories.

[0030] The relationships between categories can be used to represent the semantic inclusion relationships between categories. These semantic inclusion relationships can be obtained through semantic understanding and reasoning on all categories using a large language model. Hierarchical relationships, also known as parent-child relationships, can be exemplified by the following: "Skincare Products" can be a subcategory of "Makeup Products," and "Makeup Products" can be a subcategory of "Consumer Goods."

[0031] Step S250: Based on the hierarchical structure, clustering results, entity relationship instances, and entity attribute instances, the machine learning model generates first knowledge. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

[0032] The first layer of knowledge can be understood as the schema layer of a knowledge graph. It defines the categories within the business domain, the attributes of each category, the relationships between categories, category constraints, attribute constraints, and relationship constraints. The category definitions represent the meaning of each category (i.e., which categories exist in the business domain), the attribute definitions represent the attributes possessed by each category (i.e., which attributes the category has), and the relationship definitions represent the relationships between categories (i.e., what relationships exist between categories). The constraint rules can be used to represent the limitations (i.e., constraints) on at least one of the categories, attributes, and relationships. For example, they can restrict the hierarchical structure of categories, the scope of application and bidirectional semantics of relationships, and the type, value range, and enumeration range of attributes. The knowledge graph can be obtained by processing business data based on the schema layer. This knowledge graph can include entity instances, entity relationship instances, and entity attribute instances.

[0033] As can be seen from the above, intelligent marketing agents can understand business domain-oriented questions described in natural language. With the assistance of first knowledge, intelligent marketing agents can improve the accuracy of knowledge retrieval (entity retrieval, relation retrieval, or attribute retrieval), thereby improving the accuracy of intelligent question results that depend on retrieval results.

[0034] Furthermore, by leveraging the semantic understanding capabilities of machine learning models, the extraction and reasoning of entity instances, entity relationship instances, and entity attribute instances, as well as category reasoning and induction, hierarchical relationship reasoning between categories, and constraint rule reasoning, can be achieved to automate knowledge construction and improve the efficiency of knowledge construction. In addition, the constructed intelligent marketing agent can perform knowledge retrieval using primary knowledge and knowledge graphs as data foundations, thereby realizing corresponding intelligent questioning.

[0035] In some cases, the steps described above for processing the data to be processed to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed may include: processing the data to be processed and first context information through the machine learning model to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed, wherein the first context information includes a preset template, and the preset template includes at least one of a preset category definition, a preset attribute definition, a preset relationship definition, and a preset constraint rule, wherein the preset category definition, the preset attribute definition, the preset relationship definition, and the preset constraint rule belong to the business domain.

[0036] The preset templates for different business domains can be different. For explanations and descriptions of preset category definitions, preset attribute definitions, preset relationship definitions, and preset constraint rules, please refer to the above-mentioned category definitions, attribute definitions, relationship definitions, and constraint rules.

[0037] Using the above method, based on the preset template, the large model can avoid building the first knowledge in the business domain from scratch. That is, based on the preset template, the preset category definitions, preset attribute definitions, preset relationship definitions, and preset constraint rules in the preset template can be optimized (for example, by inducing new types or updating existing types through the large language model). This will result in the category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain, which can further improve the efficiency of construction and improve the accuracy of the first knowledge.

[0038] In some cases, the first context information may also include a terminology table in the business domain. The terminology table may include terms that are frequently used in the corresponding business domain. These terms can provide additional semantic information for the large language model, assisting the large language model in extracting entity instances, entity relationship instances, entity attribute instances, and clustering, etc.

[0039] In some cases, the first context information may also include a glossary and preset templates in the business domain, thereby improving the completeness of the first context information.

[0040] In some cases, the steps described above, which involve processing the data to be processed and the first context information using the machine learning model to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed, may include: inputting the data to be processed and the first context information into the machine learning model; guiding the machine learning model to extract entity instances and related information of the entity instances from the data to be processed based on the first context information using a first prompt word, wherein the first context information includes a preset template in the business domain and / or a terminology in the business domain; guiding the machine learning model to extract entity relationship instances and related information of the entity relationship instances from the data to be processed based on the first context information and the second context information using a second prompt word, wherein the second context information includes the entity instances and related information of the entity instances; and guiding the machine learning model to infer the entity attribute instances and related information of the entity attribute instances from the data to be processed based on the first context information, the second context information, and the third context information using a third prompt word, wherein the third context information includes entity relationship instances and related information of the entity relationship instances.

[0041] The relevant information of an entity instance may include the name of the entity instance extracted from the data to be processed and a description of the category to which the entity instance belongs. The relevant information of an entity relationship instance may include a description of the relationship type, relationship direction, and relationship attributes. The relevant information of an entity attribute instance may include a description of the data type, value range, and business meaning extracted from the data to be processed.

[0042] By using the above methods, various contextual information is provided as reference information for the machine learning model in the extraction and inference process, thereby assisting the machine learning model in extracting entity instances, entity relationship instances, and entity attribute instances.

[0043] In some cases, when a machine learning model infers clustering results, it can perform semantic analysis based on the fifth contextual information to achieve clustering. This fourth contextual information can include the first, second, third, and fourth contextual information. Explanations of the first, second, and third contextual information can be found above. The fourth contextual information can include the entity attribute instance and related information about that entity attribute instance.

[0044] By using the methods described above, various contextual information is provided as reference information when the machine learning model clusters, thereby improving the accuracy of the clustering results obtained through inference.

[0045] In some cases, the above-mentioned knowledge retrieval and intelligent questioning method based on intelligent marketing agents may further include: acquiring first knowledge to be fused and second knowledge to be fused, wherein the first knowledge to be fused and the second knowledge to be fused belong to first knowledge in different business domains; generating second knowledge based on the first knowledge to be fused and the second knowledge to be fused through the machine learning model, wherein the second knowledge belongs to the different business domains, and the second knowledge is used to support the intelligent marketing agent in realizing intelligent questioning in the different business domains.

[0046] As can be seen from the above, for enterprise-level tenants, a tenant typically faces different business systems, and different business systems face different business domains.

[0047] Similar to the first knowledge layer, the second knowledge layer is also a schema layer of the knowledge graph. However, unlike the first knowledge layer, which focuses on a single business domain, the second knowledge layer is oriented towards multiple business domains. It integrates first knowledge from different business domains, defining the categories, attributes, relationships, constraints, attribute constraints, and relationship constraints within each category. Similarly, the knowledge graph here can also include entity instances, entity relationship instances, and entity attribute instances. These instances are extracted from business data (e.g., data to be processed) across different business domains.

[0048] Both first-level and second-level knowledge can provide a unified API (Application Programming Interface) for upper-layer applications (such as intelligent marketing agents) to call.

[0049] By integrating first knowledge from different business domains in the above manner, a unified knowledge view can be generated for the same tenant but different business systems. This shields the heterogeneity of underlying data (i.e., first knowledge) in different business domains from the upper-layer applications, thereby enabling cross-domain intelligent questioning and solving the limitation of data silos. In addition, integrating first knowledge from different business domains can improve the reusability of concepts and avoid redundant modeling.

[0050] In some cases, the step of generating second knowledge based on the first and second knowledge to be fused using the machine learning model may include: determining a first concept pair based on the first and second knowledge to be fused using the machine learning model, wherein the first concept pair includes a first concept and a second concept, one of the first and second concepts belongs to the first knowledge to be fused, and the other of the first and second concepts belongs to the second knowledge to be fused, and the semantics of the first and second concepts are the same, and the concepts are used to describe categories, attributes under the categories, relationships under the categories, and constraint rules under the categories; aligning the first and second concepts at a preset granularity to obtain an alignment result; determining the conflict between the first and second knowledge to be fused at the preset granularity based on the alignment result using the machine learning model; performing conflict resolution processing on the first and / or second knowledge to be fused using the machine learning model, wherein the conflict resolution processing is used to eliminate the conflict; and generating second knowledge based on the processing result of the conflict resolution processing.

[0051] First, let's explain the concepts. In the first knowledge, a category definition, the category it maps to, the relation definitions, attribute definitions, and constraint rules under that category constitute a concept. That is, a concept represents the category definition, relation definition, attribute definition, and constraint rules. The semantics of the first and second concepts are identical; here, semantics can refer to the same business meaning. As an example, we can first determine the second concept pair based on the vector representations of each concept. If the vector representations of the two concepts in the second concept pair (i.e., the third and fourth concepts below) are greater than a preset vector similarity, then we can determine whether the semantics of the two concepts in the second concept pair are identical from the semantic dimension. Here, semantics can refer to the aforementioned business meaning. For details on determining vector representations and the determination of the first concept pair, please refer to the following content.

[0052] The aforementioned conflict can refer to differences in the way concepts are expressed at a preset granularity (i.e., descriptions or definitions, such as descriptions of the hierarchy, the syntax of fields corresponding to attributes, or constraint rules). The preset granularity can be multiple, including at least one of hierarchy granularity, attribute granularity, and constraint rule granularity. Since alignment only achieves a bidirectional mapping of concepts at the preset granularity, the descriptions or definitions, category hierarchies, and constraint rules at each preset granularity remain independent and may conflict. Therefore, based on alignment, it is possible to further determine whether there is a conflict between the first and second knowledge to be fused at the preset granularity based on their respective descriptions or definitions at the preset granularity.

[0053] For example, taking hierarchical granularity as an example, in the first knowledge to be fused, category A is at the first level, and in the second knowledge to be fused, category B is at the second level. The first and second levels are different, and category A and category B are the same category (the result of alignment). Moreover, the concepts corresponding to category A and category B can form a first concept pair. In this case, it can be determined that there is a conflict between the first and second knowledge to be fused at the hierarchical granularity. Taking attribute granularity as another example, the channel enumeration value of attribute a1 of category A is a numeric code, and the channel enumeration value of attribute b1 of category B is an English letter. In the encoding, attributes a1 and b1 are the same attribute (the result of alignment), but the types of their attribute values ​​are different. In this case, it can be determined that the first and second knowledge to be fused conflict at the attribute granularity. Taking the constraint rule granularity as an example, the value range of attribute a2 of category A has no lower limit, while the value range of attribute b1 of category B must be greater than a certain value. Attributes a2 and b2 are the same attribute (the result of alignment), but the value ranges of their attribute values ​​are different. In this case, it can be determined that the first and second knowledge to be fused conflict at the constraint rule granularity.

[0054] Once a conflict is identified, the reasoning capabilities of machine learning models can be used to analyze the causes of the conflict and deduce corresponding conflict resolution strategies. Then, conflict resolution processing can be performed based on the conflict resolution strategies to resolve the conflict.

[0055] As an example, conflict resolution strategies may include at least one of the following: The first conflict resolution strategy is used to indicate the selection of a general category definition, attribute definition, or relationship definition. Here, "general" means that both can be used if there is a conflict. The second conflict resolution strategy, the third conflict resolution strategy is used to select a general constraint rule, such as a unified range of values; The third conflict resolution strategy, and the fourth conflict resolution strategy, are used to create a parent concept, with the first concept and the second concept serving as child concepts of the parent concept.

[0056] The conflict resolution strategy is used to resolve the conflict between the first and / or second knowledge to be merged, thereby obtaining the first and second knowledge to be merged without conflict. On this basis, the cross-domain knowledge that has resolved the conflict is integrated to obtain the pattern layer (i.e., the second knowledge) of the cross-domain knowledge graph.

[0057] In this way, machine learning models automatically identify conflicts between primary knowledge across different domains, and use these models to deduce conflict resolution strategies, thereby generating a cross-domain pattern layer. This enables intelligent marketing agents to perform cross-domain knowledge retrieval or reasoning, thus overcoming the limitations of data silos.

[0058] In some cases, the step of determining a first concept pair based on the first and second knowledge to be fused using a machine learning model may include: encoding concepts in the third knowledge to be fused in a unified semantic space using the machine learning model to obtain vector representations of the concepts, wherein the third knowledge to be fused is either the first or the second knowledge to be fused; determining a second concept pair based on the vector representations of each concept, wherein the second concept pair includes a third concept and a fourth concept, one of the third and fourth concepts belongs to the first knowledge to be fused, and the other of the third and fourth concepts belongs to the second knowledge to be fused, wherein the similarity between the first and second vector representations is greater than a preset vector similarity, the first vector representation is used to represent the third concept, and the second vector representation is used to represent the fourth concept; verifying the second concept pair semantically using the machine learning model, wherein the verification is used to determine whether the semantics of the third and fourth concepts are the same; and determining the second concept pair that passes the verification as the first concept pair.

[0059] In this paper, concepts can be encoded based on information related to them in multiple dimensions, thus obtaining a vector representation of the corresponding concept. This multi-dimensional information can include encoding category definitions, attribute definitions, and relation definitions under the concept, thereby obtaining a vector representation of the corresponding concept. Representing the concept's vector representation based on multi-dimensional information related to the concept can improve the accuracy of the encoding.

[0060] After obtaining the vector representations of each concept in the first and second knowledge sets to be fused, the vector similarity between each pair of concepts can be calculated and compared with a preset vector similarity. Based on the comparison results, it can be determined whether the pair of concepts can form a second concept pair. It is understood that one concept in each pair belongs to the first knowledge set to be fused, and the other concept in the pair belongs to the second knowledge set to be fused. Furthermore, the preset vector similarity can be set according to the actual situation.

[0061] Understandably, if the third and fourth concepts in a second concept pair are semantically identical, then the second concept pair can be considered to have passed the verification; conversely, if the third and fourth concepts in a second concept pair are semantically different, then the second concept pair can be considered to have failed the verification.

[0062] As an example, normalization and / or standardization processes can be used to generate vector representations of concepts so that the vector representations are in a unified semantic space.

[0063] The above method first encodes in a unified semantic space, thereby standardizing the semantic space and ensuring that concepts from different business domains can be similarly calculated and compared in the same metric space, which helps improve the accuracy of the first concept pair. In addition, under the premise of similar vector representations, the second concept pair with similar vector representations is further verified at the semantic level, which can further improve the accuracy of the first concept pair.

[0064] In some cases, the above-mentioned knowledge retrieval and intelligent questioning method based on intelligent marketing agents may further include the following steps: in response to changes in the data source, an update task is performed through the large language model, the update task being used to update the first knowledge based on the changes to generate third knowledge, the data source being used to represent the data used to generate the first knowledge.

[0065] Third knowledge can serve as new first knowledge, and can also update the aforementioned second knowledge based on the new first knowledge. As an example, changes to the data source can include the addition of data and modifications to data, etc. As an example, the data source can include at least one of the following: The data to be processed; The preset template; The terminology list.

[0066] Using the above method, when the data source changes, the primary knowledge is automatically updated through the large language model, so that the primary knowledge can automatically evolve based on the changes in the data source.

[0067] In some cases, when changes include the addition of new data, incremental updates can be used to update the first knowledge to generate the third knowledge, thus avoiding a full reconstruction.

[0068] In some cases, the above-mentioned knowledge retrieval and intelligent questioning method based on intelligent marketing agents may further include the following steps: in response to the generation operation of fourth knowledge, recording the generation information of the fourth knowledge, the generation information being used to manage the lifecycle of the fourth knowledge and / or evaluate the quality of the fourth knowledge, the fourth knowledge including at least one of the first knowledge, the second knowledge, and the third knowledge.

[0069] As an example, the generated information may include the version number of the fourth knowledge, which can be used for lifecycle management, such as managing the lifecycle of the fourth knowledge through version rollback.

[0070] As an example, the generated information may also include the version number mentioned above, the data on which the fourth knowledge is based, the generation time, and the generation task associated with the fourth knowledge. Based on this generated information, problem investigation can be carried out, thereby enabling the evaluation of the quality of the fourth knowledge. The generation task can refer to the task used to generate the fourth knowledge, which can perform the steps mentioned above, such as instance extraction and clustering, used to generate the fourth knowledge.

[0071] By recording the generation information of the fourth knowledge through the above methods, it is possible to ensure that the entire knowledge modeling process can be traced, thereby improving the transparency and credibility of knowledge construction and providing a data foundation for lifecycle management and quality assessment.

[0072] In some cases, as can be seen from the above, the data to be processed includes at least one type of data, which is a structured type, an unstructured type, or a semi-structured type. The knowledge retrieval and intelligent data collection method based on intelligent marketing agents may further include: performing preprocessing on the data corresponding to the type according to the data preprocessing method corresponding to the type, where the data belongs to the data to be processed; in this case, the step of processing the data to be processed may include: processing the preprocessed result.

[0073] As an example, the above data preprocessing can be performed before the data to be processed is processed.

[0074] When dealing with structured data (such as data tables in a business domain database), the corresponding data preprocessing method can be to scan the table names, field names, field types, and other metadata in the data tables. This metadata can be used as input to a large language model for the construction or generation of first knowledge. The metadata extracted from structured data can provide an initial schema framework, that is, the metadata extracted from structured data can be used as initial category definitions, attribute definitions, and relation definitions, etc.

[0075] For unstructured data types (such as plain text, images, or audio content without a fixed format), data preprocessing can involve converting non-textual data into text, then breaking it down into semantically coherent text fragments, and using these fragments as input to a large language model. This text can provide a wealth of instance information, allowing for the extraction of entity instances, entity relationship instances, and entity attribute instances. Furthermore, it supports the large language model in inducing new categories or subcategories and uncovering new attributes.

[0076] For semi-structured data types (such as JSON files), data preprocessing can involve using tools to extract field definitions and hierarchical structures from the file. This information can then be used as input for a large language model. In particular, semi-structured data can simultaneously provide the framework of an initial schema and various instance information, thereby supporting the construction of first-order knowledge.

[0077] The above approach uses different preprocessing methods for different data types, thus enabling hierarchical data preprocessing to parse different types of data. Different preprocessing methods are adapted to the original characteristics of different types of data, making the parsing more accurate and efficient. In addition, the above preprocessing can provide standardized information for the construction of first knowledge, so that the upper-level modeling module (such as a large language model) does not need to distinguish whether the underlying data is a database or a voice file, simplifying the processing logic of the modeling module.

[0078] In some cases, the above-mentioned knowledge retrieval and intelligent questioning method based on intelligent marketing agents may also include the following steps: in response to the confirmation operation, performing the activation operation of the fifth knowledge, wherein the fifth knowledge includes at least one of the first knowledge, the second knowledge and the third knowledge.

[0079] This confirmation operation is triggered by an expert in the business domain to indicate that the fifth knowledge has passed human verification. Only after the fifth knowledge has been validated can it support post-processing, which can include the aforementioned intelligent question counting and cross-business domain knowledge fusion.

[0080] Especially for business scenarios with high accuracy requirements for knowledge, a verification process can be added before the knowledge is post-processed. This can improve the accuracy of post-processing when such instructions are applied in such business scenarios.

[0081] In some cases, if the aforementioned fifth piece of knowledge fails to pass human verification, experts can provide the machine learning model with suggestions for modifying the fifth piece of knowledge, so that the machine learning model can improve the quality of subsequent knowledge construction based on the modification suggestions.

[0082] Based on the same concept, a knowledge retrieval and intelligent data query device based on an intelligent marketing agent is provided. Figure 3 This is a block diagram illustrating a knowledge retrieval and intelligent question-and-answer device based on an intelligent marketing agent, shown in some scenarios. (Refer to...) Figure 3 The knowledge retrieval and intelligent question-and-answer device 300 based on intelligent marketing agents includes: The first acquisition module 301 is used to acquire data to be processed, which belongs to the business domain. Processing module 302 is used to process the data to be processed to obtain entity instances, entity relationship instances and entity attribute instances in the data to be processed. Clustering module 303 is used to cluster all the entity instances through a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions. The first generation module 304 is used to generate a hierarchical structure based on the association between categories in the clustering results using the machine learning model. The hierarchical structure is used to represent the hierarchical relationship between all the categories. The second generation module 305 is used to generate first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances through the machine learning model. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and the business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

[0083] In some cases, the processing module 302 is used to: The machine learning model processes the data to be processed and the first context information to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed. The first context information includes a preset template, which includes at least one of a preset category definition, a preset attribute definition, a preset relationship definition, and a preset constraint rule. The preset category definition, the preset attribute definition, the preset relationship definition, and the preset constraint rule belong to the business domain.

[0084] In some cases, the knowledge retrieval and intelligent questioning device 300 based on intelligent marketing agents further includes: The second acquisition module is used to acquire the first knowledge to be fused and the second knowledge to be fused, wherein the first knowledge to be fused and the second knowledge to be fused belong to the first knowledge of different business domains; The third generation module is used to generate second knowledge based on the first knowledge to be fused and the second knowledge to be fused through the machine learning model. The second knowledge belongs to the different business domains and is used to support the intelligent marketing agent to realize intelligent data collection in the different business domains.

[0085] In some cases, the third generation module includes: The first determining submodule is used to determine a first concept pair based on the first knowledge to be fused and the second knowledge to be fused using the machine learning model. The first concept pair includes a first concept and a second concept. One of the first concept and the second concept belongs to the first knowledge to be fused, and the other of the first concept and the second concept belongs to the second knowledge to be fused. The semantics of the first concept and the second concept are the same. The concept is used to describe the category, the attributes under the category, the relationships under the category, and the constraint rules under the category. The alignment submodule is used to align the first concept and the second concept at a preset granularity to obtain an alignment result; The second determining submodule is used to determine the conflict between the first knowledge to be fused and the second knowledge to be fused at the preset granularity based on the alignment result using the machine learning model; The conflict resolution submodule is used to perform conflict resolution processing on the first knowledge to be fused and / or the second knowledge to be fused through the machine learning model, wherein the conflict resolution processing is used to eliminate the conflict. A generation submodule is used to generate second knowledge based on the processing results of the conflict resolution process.

[0086] In some cases, the first determining submodule is used to: The machine learning model encodes the concepts in the third knowledge to be fused in a unified semantic space to obtain vector representations of the concepts. The third knowledge to be fused is either the first knowledge to be fused or the second knowledge to be fused. Based on the vector representations of each concept, a second concept pair is determined, which includes a third concept and a fourth concept. One of the third concept and the fourth concept belongs to the first knowledge to be fused, and the other of the third concept and the fourth concept belongs to the second knowledge to be fused. The similarity between the first vector representation and the second vector representation is greater than a preset vector similarity. The first vector representation is used to characterize the third concept, and the second vector representation is used to characterize the fourth concept. The machine learning model is used to verify the second concept pair at the semantic level, and the verification is used to determine whether the third concept and the fourth concept are semantically the same. The second concept pair that passes the verification is determined as the first concept pair.

[0087] In some cases, the knowledge retrieval and intelligent questioning device 300 based on intelligent marketing agents further includes: An update module is used to perform an update task through the machine learning model in response to changes in the data source. The update task is used to update the first knowledge based on the changes to generate third knowledge. The data source is used to represent the data used to generate the first knowledge.

[0088] In some cases, the knowledge retrieval and intelligent questioning device 300 based on intelligent marketing agents further includes: A recording module is configured to record the generation information of the fourth knowledge in response to the generation operation of the fourth knowledge. The generation information is used to manage the lifecycle of the fourth knowledge and / or evaluate the quality of the fourth knowledge. The fourth knowledge includes at least one of the first knowledge, the second knowledge, and the third knowledge.

[0089] In some cases, the data to be processed includes at least one type of data, which is a structured type, an unstructured type, or a semi-structured type. The knowledge retrieval and intelligent data collection device 300 based on intelligent marketing agents further includes: The preprocessing module is used to perform preprocessing on the data corresponding to the type according to the data preprocessing method corresponding to the type, wherein the data belongs to the data to be processed; The processing module 302 is used to process the preprocessed results.

[0090] Among them, regarding Figure 3 The implementation principles of each module in the knowledge retrieval and intelligent questioning device 300 based on intelligent marketing agents can be referred to the above content, and the knowledge retrieval and intelligent questioning device 300 based on intelligent marketing agents has the same technical effect as the knowledge retrieval and intelligent questioning method based on intelligent marketing agents described above.

[0091] Based on the same concept, a computer-readable medium is provided, on which a computer program is stored, wherein when executed by a processing device, the computer program implements the steps of the above-described method, and the computer-readable medium has the technical effects achievable by implementing the aforementioned related methods; the technical effects can be referred to the above content. Based on the same concept, a computer program product is provided, comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described method, and the computer program product has the technical effects achievable by implementing the aforementioned related methods; the technical effects can be referred to the above description. Based on the same concept, an electronic device is provided, comprising: A storage device on which computer programs are stored; A processing device is used to execute the computer program in the storage device to implement the steps of the above method, and the electronic device has the technical effects that can be achieved by implementing the above-mentioned related methods, and the technical effects can be referred to the above content.

[0092] The following is for reference. Figure 4 It shows an electronic device suitable for implementing the above method (e.g. Figure 1 The diagram below shows the structure of the client (400). Terminal devices can include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Personal Computers), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs (Televisions), desktop computers, etc. Figure 4 The electronic device shown is merely an example and should not be construed as limiting its functionality or scope of use.

[0093] like Figure 4 As shown, electronic device 400 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from storage device 408 into random access memory (RAM) 403. The random access memory 403 also stores various programs and data required for the operation of electronic device 400. The processing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0094] Typically, the following devices can be connected to the input / output interface 405: input devices 406 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 407 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 408 including, for example, magnetic tape, hard disk, etc.; and communication devices 409. Communication device 409 allows electronic device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4An electronic device 400 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0095] In particular, depending on certain circumstances, the processes described in the above-referenced flowchart can be implemented as computer software programs. For example, a computer program product is provided, comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. This computer program can be downloaded and installed from a network via communication device 409, or installed from storage device 408, or installed from read-only memory 402. When the computer program is executed by processing device 401, it performs the functions defined in the above-described methods.

[0096] It should be noted that the aforementioned computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM, or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In one case, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In another case, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0097] In some scenarios, clients can communicate using any currently known or future-developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include Local Area Networks (LANs), Wide Area Networks (WANs), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad-hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0098] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0099] The aforementioned computer-readable medium carries one or more programs. When the aforementioned one or more programs are executed by the electronic device, the electronic device causes the following: to acquire data to be processed, the data to be processed belonging to a business domain; to process the data to be processed to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed; to cluster all the entity instances using a machine learning model to obtain clustering results, wherein entity instances belonging to the same category in the clustering results satisfy preset conditions; to generate a hierarchical structure based on the association relationships between categories in the clustering results using the machine learning model, the hierarchical structure being used to represent the hierarchical relationships between all the categories; and to generate first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances using the machine learning model, the first knowledge belonging to the business domain, the first knowledge including category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain, the first knowledge being used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent data collection in the business domain.

[0100] Computer program code for performing the above operations can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages, as well as conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0101] The flowcharts and block diagrams in the accompanying figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products under various scenarios. Each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that, in some alternative cases, the functions indicated in the blocks may occur in a different order than those indicated in the figures. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0102] The modules mentioned above can be implemented in software or hardware. In some cases, the name of a module does not necessarily limit the module itself; for example, the first acquisition module can also be described as a "module for acquiring data to be processed".

[0103] The functions described above can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field-Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application-Specific Standard Parts (ASSPs), Systems on Chips (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0104] In this context, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0105] The above description is merely illustrative and explains the technical principles employed. Those skilled in the art should understand that the scope of this document is not limited to the specific combinations of the above-described technical features, but should also cover any combination of the above-described technical features or their equivalents without departing from the above concept.

[0106] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. Multitasking and parallel processing may be advantageous in certain contexts. Similarly, while some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of this paper. Certain features described in the context of a single example can also be implemented in combination in a single example. Conversely, various features described in the context of a single example can also be implemented individually or in any suitable sub-combination in multiple examples.

[0107] Although this document has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims. Regarding the aforementioned apparatus, the specific manner in which the various modules perform their operations has already been described in detail in the section concerning the method, and will not be elaborated upon here.

Claims

1. A knowledge retrieval and intelligent questioning method based on intelligent marketing agents, comprising: Acquire data to be processed, which belongs to the business domain; The data to be processed is processed to obtain entity instances, entity relationship instances, and entity attribute instances from the data to be processed. All the entity instances are clustered using a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions. The machine learning model generates a hierarchical structure based on the association between categories in the clustering results. The hierarchical structure is used to represent the hierarchical relationship between all the categories. The machine learning model generates first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and the business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

2. The method according to claim 1, wherein processing the data to be processed to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed includes: The machine learning model processes the data to be processed and the first context information to obtain entity instances, entity relationship instances, and entity attribute instances in the data to be processed. The first context information includes a preset template, which includes at least one of a preset category definition, a preset attribute definition, a preset relationship definition, and a preset constraint rule. The preset category definition, the preset attribute definition, the preset relationship definition, and the preset constraint rule belong to the business domain.

3. The method according to claim 1, further comprising: Acquire a first knowledge to be merged and a second knowledge to be merged, wherein the first knowledge to be merged and the second knowledge to be merged belong to first knowledge in different business domains; The machine learning model generates second knowledge based on the first and second knowledge to be fused. The second knowledge belongs to the different business domains and is used to support the intelligent marketing agent in realizing intelligent questioning in the different business domains.

4. The method according to claim 3, wherein generating second knowledge based on the first knowledge to be fused and the second knowledge to be fused using the machine learning model includes: The machine learning model determines a first concept pair based on the first knowledge to be fused and the second knowledge to be fused. The first concept pair includes a first concept and a second concept. One of the first concept and the second concept belongs to the first knowledge to be fused, and the other of the first concept and the second concept belongs to the second knowledge to be fused. The semantics of the first concept and the second concept are the same. The concept is used to describe the category, the attributes under the category, the relationships under the category, and the constraint rules under the category. Align the first concept and the second concept at a preset granularity to obtain an alignment result; Based on the alignment results, the machine learning model determines the conflict between the first knowledge to be fused and the second knowledge to be fused at the preset granularity. The machine learning model is used to perform conflict resolution processing on the first knowledge to be fused and / or the second knowledge to be fused, and the conflict resolution processing is used to eliminate the conflict. Based on the processing results of the conflict resolution process, second knowledge is generated.

5. The method according to claim 4, wherein determining the first concept pair based on the first knowledge to be fused and the second knowledge to be fused using the machine learning model includes: The machine learning model encodes the concepts in the third knowledge to be fused in a unified semantic space to obtain vector representations of the concepts. The third knowledge to be fused is either the first knowledge to be fused or the second knowledge to be fused. Based on the vector representations of each concept, a second concept pair is determined, which includes a third concept and a fourth concept. One of the third concept and the fourth concept belongs to the first knowledge to be fused, and the other of the third concept and the fourth concept belongs to the second knowledge to be fused. The similarity between the first vector representation and the second vector representation is greater than a preset vector similarity. The first vector representation is used to characterize the third concept, and the second vector representation is used to characterize the fourth concept. The machine learning model is used to verify the second concept pair at the semantic level, and the verification is used to determine whether the third concept and the fourth concept are semantically the same. The second concept pair that passes the verification is determined as the first concept pair.

6. The method according to claim 1, further comprising: In response to changes in the data source, an update task is performed through the machine learning model. The update task is used to update the first knowledge based on the changes to generate third knowledge. The data source is used to represent the data used to generate the first knowledge.

7. The method according to claim 3 or 6, further comprising: In response to the generation operation of the fourth knowledge, the generation information of the fourth knowledge is recorded. The generation information is used to manage the lifecycle of the fourth knowledge and / or evaluate the quality of the fourth knowledge. The fourth knowledge includes at least one of the first knowledge, the second knowledge, and the third knowledge.

8. The method according to claim 1, wherein the data to be processed includes at least one type of data, the type being a structured type, an unstructured type, or a semi-structured type, and the method further includes: Preprocessing is performed on the data corresponding to the type according to the data preprocessing method corresponding to the type, where the data belongs to the data to be processed; The processing of the data to be processed includes: The preprocessed results are then processed.

9. A knowledge retrieval and intelligent questioning device based on an intelligent marketing agent, comprising: The first acquisition module is used to acquire data to be processed, which belongs to the business domain. The processing module is used to process the data to be processed in order to obtain entity instances, entity relationship instances and entity attribute instances in the data to be processed. The clustering module is used to cluster all the entity instances using a machine learning model to obtain clustering results. In the clustering results, entity instances belonging to the same category meet preset conditions. The first generation module is used to generate a hierarchical structure based on the association between categories in the clustering results using the machine learning model. The hierarchical structure is used to represent the hierarchical relationship between all the categories. The second generation module is used to generate first knowledge based on the hierarchical structure, the clustering results, the entity relationship instances, and the entity attribute instances using the machine learning model. The first knowledge belongs to the business domain and includes category definitions, attribute definitions, relationship definitions, and constraint rules in the business domain. The first knowledge is used to represent the pattern layer of the knowledge graph. The knowledge graph is obtained by processing the pattern layer and the business data of the business domain. The first knowledge and the knowledge graph are used to support the intelligent marketing agent in realizing knowledge retrieval and intelligent questioning in the business domain.

10. An electronic device, comprising: A storage device on which computer programs are stored; A processing apparatus for executing the computer program in the storage device to implement the method of any one of claims 1-8.

Images