Customer intelligent recommendation method, system, device and medium based on large language model

Abstract

The application provides a customer intelligent recommendation method, system, device and medium based on a large language model. The method comprises the following steps: constructing an enterprise knowledge graph, defining a business subject and an associated general correlation relationship respectively; constructing a label classification model, outputting the business subject with a classification label through the association mapping of the business subject and the general correlation relationship; pre-setting a first relationship triple, constructing a first prompt instruction for instantiating a query and inputting the large language model to obtain a query result; generating a second relationship triple according to the query result, constructing a second prompt instruction for instantiating a verification and inputting the large language model to obtain a verification result, matching the tail entity after the verification to the business subject of the enterprise knowledge graph, and determining a target recommended customer according to the mapping relationship of the enterprise knowledge graph. The application determines the target customer from the information between the enterprise business industry chain by combining the enterprise knowledge graph with the large language model, improves the robustness and accuracy of customer recommendation, and has strong practicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a customer intelligent recommendation method, system, device and medium based on a large language model. BACKGROUND

[0002] With the development of artificial intelligence and big data technology, business services in the sales field have expanded from customer recommendation on the ToC end to customer recommendation services on the ToB end. However, the customer recommendation services on the ToB end face the problem of multiple sub-industry dimensions. At present, the domestic secondary industry is divided into 97 categories, and the four-level industry classification reaches 1382 categories. Moreover, the customer portraits, customer acquisition factors and business logic among various sub-industries are quite different, the data volume is huge, and the customer recommendation logic is complex.

[0003] The prior art establishes a customer recommendation model through the collected customer behavior data of each industry, analyzes the multi-dimensional similar features of customer enterprises based on the behavior data, classifies the customer industries, and thus completes the purpose of customer recommendation. However, the shortcomings of the prior art are that the customer behavior data cannot represent the correlation information between the upstream and downstream of the industry chain, and the target customer portrait recommended by the generated model has a lot of white noise, and the robustness and accuracy of the recommendation model are low.

[0004] At present, there is an urgent need for a customer intelligent recommendation method based on a large language model to solve the problem of low accuracy of ToB customer recommendation services. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a customer intelligent recommendation method, system, device and medium based on a large language model.

[0006] The first aspect of the present application discloses a customer intelligent recommendation method based on a large language model, comprising:

[0007] S1: constructing an enterprise knowledge graph, and defining a business subject of the enterprise and a generalized related relationship associated with the business subject, respectively;

[0008] S2: constructing a label classification model for classifying the business subject according to the generalized related relationship, and outputting the business subject with a classification label through the association mapping of the business subject and the generalized related relationship;

[0009] S3: pre-setting a first relationship triple including a head entity, a first relationship classification and a tail entity, constructing a first prompt instruction for instantiating a query of the tail entity based on the business subject with a classification label and the generalized related relationship, and inputting the first prompt instruction into the large language model to obtain a query result of a plurality of tail entities;

[0010] S4: generating a plurality of second relation triples including the tail entity, a second relation classification, and the head entity according to the query result, constructing a second prompt instruction for instantiating and verifying the second relation classification, inputting the second prompt instruction into the large language model to obtain a verification result of the second relation classification, and confirming a general related relation between the head entity and the tail entity in the second relation triple through the verification result;

[0011] S5: matching the tail entity after verification to the business subject in the enterprise knowledge graph, and determining a target recommended customer according to a mapping relationship between the business subject and the enterprise in the enterprise knowledge graph.

[0012] In an optional embodiment, the business subject is a main business corresponding to an enterprise, the main business includes products and services, and the general related relation is positioning information of the main business in an industrial chain.

[0013] In an optional embodiment, the head entity in the first relation triple is a business subject input for instantiation, the first relation classification is a general related relation input for instantiation, the tail entity is the business subject output for instantiation, the second relation classification is a plurality of different general related relations between a plurality of tail entities and a single head entity, the second relation classification is used to confirm other business subjects associated with the business subject corresponding to the head entity, and then the target recommended customer of the enterprise associated with the business subject is confirmed.

[0014] In an optional embodiment, constructing a first prompt instruction for instantiating and querying the tail entity based on the business subject with a classification label and the general related relation includes:

[0015] inputting the business subject with a classification into the head entity of the first relation triple, inputting the general related relation as a query condition as the first relation classification of the first relation triple according to an instantiation query requirement, obtaining a first semantic vector representing a query of the business subject matching the query condition through feature extraction, and generating a first prompt instruction for inputting into the large language model for verification according to representation generation of the first semantic vector.

[0016] In an optional embodiment, generating a plurality of second relation triples including the tail entity, a second relation classification, and the head entity according to the query result, and constructing a second prompt instruction for instantiating and verifying the second relation classification include:

[0017] A plurality of tail entities corresponding to the business subject of the head entity are obtained from the query result, a plurality of second relation triples are constructed by matching the business subject corresponding to the plurality of tail entities, the business subject corresponding to the head entity, and a plurality of generalized related relations matched by the two business subjects, a second semantic vector representing the verification of the generalized related relation matching the verification condition is obtained by extracting the features in the text, a second prompt instruction for inputting the large language model for verification is generated according to the second semantic vector representation, so as to obtain the association information of the other business subjects in the industry chain corresponding to the head entity.

[0018] In an optional embodiment, the positioning information is the position of the business subject in the industry chain network constructed by the business subject through a plurality of generalized related relations, and the association information between the main businesses corresponding to different business subjects in the industry chain.

[0019] In an optional embodiment, the large language model is one of a GPT model, an MPT model, a Bloom model, or an ERNIE model.

[0020] The second aspect of the application discloses a customer intelligent recommendation system based on a large language model, which comprises:

[0021] A preset module is configured to construct an enterprise knowledge graph, and define the business subject of the enterprise and the generalized related relation associated with the business subject.

[0022] A classification module is configured to construct a label classification model for classifying the business subject according to the generalized related relation, and output the business subject with a classification label through the association mapping of the business subject and the generalized related relation.

[0023] A query module is configured to preset a first relation triple including a head entity, a first relation classification, and a tail entity, construct a first prompt instruction for instantiating the query of the tail entity based on the business subject with a classification label and the generalized related relation, and input the first prompt instruction into the large language model to obtain a query result of a plurality of tail entities.

[0024] A verification module is configured to generate a plurality of second relation triples including the tail entity, a second relation classification, and a head entity according to the query result, construct a second prompt instruction for instantiating the verification of the second relation classification, input the second prompt instruction into the large language model to obtain a verification result of the second relation classification, and confirm the generalized related relation in which the head entity and the tail entity are matched in the second relation triple through the verification result.

[0025] An output module is configured to match the verified tail entity to the business subject in the enterprise knowledge graph, and determine a target recommended customer according to a mapping relationship between the business subject and the enterprise in the enterprise knowledge graph.

[0026] In an optional embodiment, the query module comprises:

[0027] A first instruction generation module is configured to input the determined classification of the business subject as the head entity of the first relationship triple, input the generalized related relationship as a query condition as a first relationship classification of the first relationship triple according to an instantiation query requirement, obtain a first semantic vector representing that the business subject is queried to match the query condition through text feature extraction, and generate a first prompt instruction for inputting the large language model for verification according to the first semantic vector representation.

[0028] In an optional embodiment, the verification module comprises:

[0029] A second instruction generation module is configured to obtain a plurality of tail entities corresponding to the business subject matched with the head entity from the query result, construct a plurality of second relationship triples from the business subject corresponding to the plurality of tail entities, the business subject corresponding to the head entity, and a plurality of generalized related relationships matched with the two business subjects, obtain a second semantic vector representing that the generalized related relationship is verified to match the verification condition through text feature extraction, and generate a second prompt instruction for inputting the large language model for verification according to the second semantic vector representation, so as to obtain the association information of other business subjects associated with the business subject corresponding to the head entity in the industrial chain.

[0030] The third aspect of the present application discloses a customer intelligent recommendation device based on a large language model, comprising:

[0031] at least one processor, and

[0032] a memory in communication connection with the at least one processor; wherein

[0033] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any one of the customer intelligent recommendation methods based on a large language model disclosed in the first aspect of the present application.

[0034] The fourth aspect of the present application discloses a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are used to make a computer execute the customer intelligent recommendation method based on a large language model disclosed in any one of the first aspect of the present application.

[0035] Compared with the prior art, the present application has the following advantages:

[0036] (1) The present application defines the main business information of the enterprise in the enterprise knowledge graph and the positioning information of the business subject in the industrial chain in advance, and then completes the pre-classification of the input content of the query industrial chain associated information through the constructed label classification model, completes the early data processing process, avoids a large amount of data processing process, and improves the efficiency of querying the recommended customers;

[0037] (2) The present application inputs the enterprise business subject into the head entity of the first relationship triple, and then combines the first relationship classification as the query condition and the tail entity corresponding to the head entity in the associated industrial chain to be queried. The semantic vector prompt instruction input large language model can better express the query problem through pre-setting, and the large language model contains a large amount of corpus and industry knowledge. The related knowledge is extracted from the large language model to construct the upstream and downstream relationship of the industrial chain, and the head entity matching multiple tail entities can be output. Thus, the associated business subject of the enterprise business subject as the query input is preliminarily determined, and then the prompt instruction of the second relationship triple is input into the large language model to infer and verify the second relationship triple. The robustness and accuracy of the query are improved, and the relationship between the upstream and downstream business subjects in the industrial chain is obtained. It is ensured that the input business subject can query all the associated business subjects in the industrial chain, and the problem of white noise in the target customer portrait recommended by the existing generation model is solved. The robustness and accuracy of the recommendation model are low.

[0038] (3) The present application obtains the recommended customer list by combining the mapping relationship between the business subject and the enterprise in the enterprise knowledge graph based on the other related business subjects obtained by the input business subject in the query result, and realizes the matching of the corresponding enterprise by the full-chain related business of the upstream and downstream of the industrial chain. The target customer is accurately obtained, and has strong practicability. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 The flowchart of the customer intelligent recommendation method based on the large language model of the present application;

[0041] Figure 2 The schematic diagram of the customer intelligent recommendation system based on the large language model of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work belong to the scope of protection of the present application.

[0043] In the description of the present application, it should be noted that the orientations or positional relationships indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third", "fourth", etc. are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0044] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two elements. For a person of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0045] Embodiment 1

[0046] Referring to Figure 1 The embodiment of the present application discloses a customer intelligent recommendation method based on a large language model, comprising:

[0047] S1: constructing an enterprise knowledge graph, defining a business subject of the enterprise and a generalized related relationship associated with the business subject, respectively;

[0048] S2: constructing a label classification model for classifying the business subject according to the generalized related relationship, and outputting the business subject with a classification label through the association mapping of the business subject and the generalized related relationship;

[0049] S3: pre-setting a first relationship triple including a head entity, a first relationship classification and a tail entity, constructing a first prompt instruction for instantiating a query of the tail entity based on the business subject with a classification label and the generalized related relationship, and inputting the first prompt instruction into the large language model to obtain a query result of a plurality of tail entities;

[0050] S4: generating a plurality of sets of second relation triples including the tail entity, the second relation classification, and the head entity according to the query result, constructing a second prompt instruction for instantiating and verifying the second relation classification, inputting the second prompt instruction into the large language model to obtain a verification result of the second relation classification, and confirming a general relevant relation between the head entity and the tail entity in the second relation triple through the verification result;

[0051] S5: matching the verified tail entity to the business subject in the enterprise knowledge graph, and determining a target recommended customer according to a mapping relationship between the business subject and the enterprise in the enterprise knowledge graph.

[0052] In an optional embodiment, the business subject is a main business corresponding to the enterprise, the main business includes products and services, and the general relevant relation is positioning information of the main business in an industrial chain.

[0053] It should be noted that the business subject can be a product or service, for example, the product to be queried is a photoresistor, the business subject is a photoresistor, the classification label of the photoresistor is a component, and the other relationship categories included in the general correlation relationship of the photoresistor are raw materials, production equipment, production consumables, components, and packaging materials, etc. The photoresistor as the business subject is input as the head entity of the first relationship triple, and then the first relationship classification can be input as raw materials, production equipment, production consumables, components, and packaging materials, respectively, to construct 5 first relationship triples, and generate query statements such as (what are the raw materials of the photoresistor), and then the query statement is input into the large language model by the prompt instruction, and relevant knowledge is extracted from the large language model based on the characteristics of the large language model containing a large amount of corpus and industry knowledge, and a plurality of tail entities matched with different first relationship classifications of each photoresistor are output, such as the raw materials of the photoresistor include cadmium sulfide, selenium, aluminum sulfide, lead sulfide, and bismuth sulfide. Among them, cadmium sulfide, selenium, aluminum sulfide, lead sulfide, and bismuth sulfide are the business subjects corresponding to the tail entities output by the large language model after matching the first relationship triple, and then a plurality of second relationship triples for verification are constructed through the tail entities output by the query result, and the prompt instruction expressing the verification statement (such as cadmium sulfide is a raw material of the photoresistor) is input into the large language model, and relevant knowledge is extracted from the large language model based on the characteristics of the large language model containing a large amount of corpus and industry knowledge, to verify the accuracy of the query result. After the verification is completed, the output tail entities are combined with the enterprise knowledge graph to confirm that the associated enterprises of the photoresistor include enterprises whose products in the upstream of the industrial chain are cadmium sulfide, and then accurate target recommended customer enterprise information is obtained. Those skilled in the art will readily know that in the actual query process, the head entity and the first and second classification relationships in the first relationship triple and the second relationship triple can be set according to actual conditions, and their role is to ensure that the query result meets the expectations.

[0054] In an optional embodiment, the head entity in the first relationship triple is an instantiated input business subject, the first relationship classification is an instantiated input general correlation relationship, the tail entity is an instantiated output business subject, the second relationship classification is a plurality of tail entities matching a plurality of different general correlation relationships of a single head entity, and the other business subjects associated with the business subject corresponding to the head entity are confirmed through the second relationship classification, and then the target recommended customer enterprises associated with the business subject are confirmed.

[0055] In an optional embodiment, the first prompt instruction for instantiating the query of the tail entity based on the business subject and the general correlation relationship with the classification label comprises:

[0056] The business subject determined to be classified is input as the head entity of the first relationship triple, and the generalized related relationship input as the query condition is input as the first relationship classification of the first relationship triple according to the instantiation query requirement, a first semantic vector representing that the business subject is matched with the query condition is obtained through the feature extraction, and a first prompt instruction for inputting the large language model for checking is generated according to the first semantic vector representation.

[0057] In an optional embodiment, the second prompt instruction for instantiating and checking the second relationship classification is constructed according to the query result, and the second prompt instruction includes:

[0058] A plurality of tail entities matched with the head entity are obtained from the query result, a plurality of second relationship triples are constructed according to the business subjects corresponding to the plurality of tail entities, the business subject corresponding to the head entity, and the plurality of generalized related relationships matched between the two business subjects, a second semantic vector representing that the generalized related relationships are matched with the check condition is obtained through the feature extraction, a second prompt instruction for inputting the large language model for checking is generated according to the second semantic vector representation, and the associated information of the other business subjects in the industrial chain of the business subject corresponding to the head entity is obtained.

[0059] In an optional embodiment, the positioning information is the position of the business subject in the industrial chain network constructed by the plurality of generalized related relationships, and the associated information between the main businesses corresponding to the different business subjects in the industrial chain.

[0060] In an optional embodiment, the large language model is one of a GPT model, an MPT model, a Bloom model, or an ERNIE model.

[0061] The present application predefines the main business information of the enterprise in the enterprise knowledge graph and the positioning information of the generalized related relationship of the business subject in the industrial chain, and then completes the pre-classification of the input content of the query industrial chain associated information through the constructed label classification model, completes the early data processing process, avoids a large amount of data processing process, and improves the efficiency of querying and recommending customers.

[0062] The application can better express the semantic vector prompt instruction input into the large language model by inputting the enterprise business subject as the head entity of the first relationship triple, combining the first relationship classification as the query condition and the tail entity associated in the corresponding industrial chain of the head entity to be queried, and pre-setting the prompt instruction, and can extract relevant knowledge from the large language model containing a large amount of corpus and industry knowledge, construct the upstream and downstream relationship of the industrial chain, output multiple tail entities matched with the head entity, and preliminarily determine the associated business subjects of the enterprise business subject as the query input, and then input the prompt instruction of the second relationship triple into the large language model to reason and verify the second relationship triple, improve the robustness and accuracy of the query, and obtain the relationship description between the upstream and downstream business subjects in the industrial chain, so that the input business subject can query all the associated business subjects in the industrial chain, and solve the problem that the target customer portrait recommended by the existing generation model has a lot of white noise and the robustness and accuracy of the recommendation model are low.

[0063] The application can obtain the recommended customer list by combining the mapping relationship between the business subject and the enterprise in the enterprise knowledge graph based on the other related business subjects obtained by the input business subject in the query result, realize the matching of the corresponding enterprise by the full-chain related business of the upstream and downstream of the industrial chain, and accurately obtain the target customer, which has strong practicability.

[0064] Embodiment 2

[0065] As shown in Figure 2 Embodiment 2 of the application discloses a customer intelligent recommendation system based on a large language model, which comprises:

[0066] A preset module is configured to construct an enterprise knowledge graph, and define a business subject of the enterprise and a generalized related relationship associated with the business subject, respectively.

[0067] A classification module is configured to construct a label classification model for classifying the business subject according to the generalized related relationship, and output the business subject with a classification label through the association mapping between the business subject and the generalized related relationship.

[0068] A query module is configured to pre-set a first relationship triple comprising a head entity, a first relationship classification and a tail entity, construct a first prompt instruction for instantiating the query of the tail entity based on the business subject with a classification label and the generalized related relationship, and input the first prompt instruction into the large language model to obtain a query result of multiple tail entities.

[0069] The checking module is configured to generate a plurality of second relation triples including the tail entity, a second relation classification, and the head entity according to the query result, construct a second prompt instruction for instantiating and checking the second relation classification, input the second prompt instruction into the large language model to obtain a checking result of the second relation classification, and confirm a general related relation between the head entity and the tail entity in the second relation triple through the checking result.

[0070] The output module is configured to match the tail entity after checking with the business subject in the enterprise knowledge graph, and determine a target recommended customer according to a mapping relation between the business subject and the enterprise in the enterprise knowledge graph.

[0071] In an optional embodiment, the query module comprises:

[0072] The first instruction generation module is configured to input the determined classification business subject as the head entity of the first relation triple, input the general related relation as a query condition as a first relation classification of the first relation triple according to an instantiation query requirement, obtain a first semantic vector representing that the business subject is queried to match the query condition through text feature extraction, and generate a first prompt instruction for inputting the large language model for checking according to representation of the first semantic vector.

[0073] In an optional embodiment, the checking module comprises:

[0074] The second instruction generation module is configured to obtain a plurality of tail entities matching the head entity corresponding business subject from the query result, construct a plurality of second relation triples by the plurality of tail entities corresponding business subjects, the head entity corresponding business subject, and a plurality of general related relations between the two business subjects, obtain a second semantic vector representing that the general related relation is checked to match the checking condition through text feature extraction, and generate a second prompt instruction for inputting the large language model for checking according to representation of the second semantic vector, so as to obtain association information of other business subjects associated in the head entity corresponding business subject industrial chain.

[0075] Embodiment 3

[0076] Embodiment 3 of the present application discloses a customer intelligent recommendation device based on a large language model, comprising:

[0077] at least one processor, and

[0078] a memory in communication connection with the at least one processor; wherein

[0079] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of customer intelligent recommendation based on a large language model according to any one of the first aspect of the present application.

[0080] The computer device can be a terminal, and the computer device includes a processor, a memory, a network interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement the method of customer intelligent recommendation based on a large language model. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or can be a key, trackball or touchpad arranged on the shell of the computer device, or can be an external keyboard, touchpad or mouse, etc.

[0081] Embodiment 4

[0082] Embodiment 4 of the present application discloses a computer readable storage medium, the computer readable storage medium stores computer executable instructions, and the computer executable instructions are used to make a computer execute the method of customer intelligent recommendation based on a large language model according to any one of the first aspect of the present application.

[0083] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of the customer intelligent recommendation method based on a large language model. Any reference to memory, storage, databases, or other media in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0084] Alternatively, the above-mentioned modules of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the embodiments of the present application. The aforementioned storage medium includes mobile storage devices, RAM, ROM, magnetic or optical disks, and various media that can store program codes.

[0085] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for customer intelligent recommendation based on a large language model, characterized in that, The method comprises: S1: constructing an enterprise knowledge graph, and defining a business subject of the enterprise and a general correlation relationship associated with the business subject respectively; S2: constructing a label classification model for classifying the business subject according to the general correlation relationship, and outputting the business subject with a classification label through the association mapping of the business subject and the general correlation relationship; S3: pre-setting a first relationship triple including a head entity, a first relationship classification and a tail entity, constructing a first prompt instruction for instantiating to query the tail entity based on the business subject with the classification label and the general correlation relationship, inputting the first prompt instruction into the large language model to obtain query results of the tail entity; S4: generating a plurality of second relationship triples including the tail entity, a second relationship classification and a head entity according to the query results, constructing a second prompt instruction for instantiating to verify the second relationship classification, inputting the second prompt instruction into the large language model to obtain a verification result of the second relationship classification, and confirming the general correlation relationship in which the head entity and the tail entity in the second relationship triple match through the verification result; S5: matching the verified tail entity with the business subject in the enterprise knowledge graph, and determining a target recommended customer according to the mapping relationship between the business subject and the enterprise in the enterprise knowledge graph. The head entity in the first relationship triple is the instantiated input business subject, the first relationship classification is the instantiated input general correlation relationship, the tail entity is the instantiated output business subject, the second relationship classification is a plurality of different general correlation relationships of a plurality of tail entities matching a single head entity, and the other business subjects associated with the business subject corresponding to the head entity are confirmed through the second relationship classification, thereby confirming the enterprise target recommended customer associated with the business subject.

2. The method of claim 1, wherein the method further comprises: The business subject is the main business corresponding to the enterprise, the main business includes products and services, and the general correlation relationship is positioning information of the main business in an industrial chain.

3. The method of claim 1, wherein the method further comprises: The first prompt instruction for instantiating to query the tail entity based on the business subject with the classification label and the general correlation relationship comprises: inputting the determined classified business subject as the head entity of the first relationship triple, inputting the general correlation relationship as the first relationship classification of the first relationship triple as a query condition according to the instantiation query requirement, obtaining a first semantic vector representing the query of the business subject matching the query condition through text feature extraction, and generating a first prompt instruction for inputting the large language model for verification according to the first semantic vector representation.

4. The large language model-based customer intelligent recommendation method according to claim 1, characterized in that, The second prompt instruction for instantiating to verify the second relationship classification according to the query result of generating a plurality of second relationship triples including the tail entity, the second relationship classification and the head entity comprises: The query result is used to obtain a plurality of tail entities corresponding to the business subject of the head entity, and the business subjects corresponding to the tail entities and the head entity and a plurality of generalized related relations matched between the two business subjects are constructed as a plurality of second relation triples. A second semantic vector representing a check condition of the generalized related relation matching check is obtained by extracting features in the text. A second prompt instruction for inputting the large language model for checking is generated according to the second semantic vector representation, so as to obtain the associated information of the other business subjects in the industry chain of the business subject corresponding to the head entity.

5. The method of claim 2, wherein the method further comprises: The positioning information is the position of the business subject in the industry chain network constructed by the plurality of generalized related relations, and the associated information between the main businesses corresponding to the different business subjects downstream of the industry chain.

6. The large language model-based customer intelligent recommendation method according to claim 1, characterized in that, The large language model is one of a GPT model, an MPT model, a Bloom model, or an ERNIE model.

7. A large language model-based customer intelligent recommendation system, characterized in that, The system comprises: A preset module is configured to construct an enterprise knowledge graph, and define the business subject of the enterprise and the generalized related relation associated with the business subject. A classification module is configured to construct a label classification model for classifying the business subject according to the generalized related relation, and output the business subject with a classification label through the association mapping between the business subject and the generalized related relation. A query module is configured to preset a first relation triple including a head entity, a first relation classification, and a tail entity, construct a first prompt instruction for instantiating the query of the tail entity based on the business subject with a classification label and the generalized related relation, and input the first prompt instruction into the large language model to obtain a query result of a plurality of tail entities. A check module is configured to generate a plurality of second relation triples including the tail entity, a second relation classification, and a head entity according to the query result, construct a second prompt instruction for instantiating the check of the second relation classification, input the second prompt instruction into the large language model to obtain a check result of the second relation classification, and confirm the generalized related relation in which the head entity and the tail entity are matched in the second relation triple through the check result. An output module is configured to match the tail entity after the check to the business subject in the enterprise knowledge graph, and determine a target recommended customer according to the mapping relationship between the business subject and the enterprise in the enterprise knowledge graph. The head entity in the first relation triple is an instantiated input business subject, the first relation classification is an instantiated input generalized related relation, the tail entity is an instantiated output business subject, the second relation classification is a plurality of different generalized related relations of a plurality of tail entities matching a single head entity, and the other business subjects associated with the business subject corresponding to the head entity are confirmed through the second relation classification, thereby confirming the enterprise target recommended customer associated with the business subject.

8. The large language model-based customer intelligent recommendation system according to claim 7, characterized in that, The query module comprises: The first instruction generation module is configured to input the determined business subject as the head entity of the first relationship triple, and input the generalized related relationship as the query condition as the first relationship classification of the first relationship triple according to the instantiation query requirement, obtain a first semantic vector representing that the business subject is matched with the query condition through text feature extraction, and generate a first prompt instruction for inputting the large language model for checking according to the first semantic vector representation.

9. The large language model-based customer intelligent recommendation system according to claim 7, characterized in that, The checking module comprises: The second instruction generation module is configured to obtain a plurality of tail entities corresponding to the business subject matched with the head entity from the query result, construct a plurality of second relationship triples by using the business subjects corresponding to the plurality of tail entities, the business subject corresponding to the head entity, and a plurality of generalized related relationships matched with the two business subjects, obtain a second semantic vector representing that the generalized related relationship is matched with the checking condition through text feature extraction, generate a second prompt instruction for inputting the large language model for checking according to the second semantic vector representation, and obtain the association information of the other business subjects associated with the business subject corresponding to the head entity in the industrial chain.

10. A large language model-based customer intelligent recommendation device, characterized in that, Comprise: At least one processor, and The memory is in communication with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the customer intelligent recommendation method based on the large language model according to any one of claims 1 to 6.

11. A computer readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions for causing a computer to execute the customer intelligent recommendation method based on the large language model according to any one of claims 1 to 6.

Images