LLM feedback mechanism-based atomic power recommendation method and device

Abstract

The invention discloses an LLM feedback mechanism-based atomic power recommendation method and device, and the method comprises the steps: obtaining a to-be-matched capability input by a target object, and carrying out the first vectorization processing of the to-be-matched capability, and obtaining a to-be-matched vector; based on the to-be-matched vectors, combining historical to-be-matched capabilities with a preset atomic energy vector library, and obtaining candidate atomic energy through similarity matching; according to the to-be-matched capability and the candidate atomic capability, outputting a recommendation result through a preset large language model; in response to a selection instruction of the target object, determining a final recommendation result; and checking according to the final recommendation result to obtain an adoption result of the recommendation result, and adjusting model parameters of the large language model based on the adoption result. According to the method, efficient and accurate atomic power recommendation can be realized, model self-optimization and technical sustainable development can be realized based on an LLM feedback mechanism, and the method can be widely applied to the technical field of data processing.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an atomic capability recommendation method and device based on an LLM feedback mechanism. Background Art

[0002] The design concept of atomic capabilities is to promote rapid innovation, flexible combination and efficient operation. However, in the actual application process, it is often necessary to have a certain understanding of each capability, which requires a high business threshold. In addition, in actual telecommunications services, when users select the atomic capabilities they need to use according to the requirements document, the accuracy is difficult to achieve the expected level. The selection of atomic capabilities is affected to a certain extent by human subjective judgment. In addition, directly providing recommended results is too simple and lacks recommendation reasons, resulting in poor user experience. Summary of the invention

[0003] The present invention proposes an atomic capability recommendation method and device based on the LLM feedback mechanism, aiming to solve one of the technical problems in the related art at least to a certain extent. The embodiment of the present invention can realize atomic capability recommendation efficiently and accurately.

[0004] On the one hand, an embodiment of the present invention provides an atomic capability recommendation method based on an LLM feedback mechanism, including:

[0005] Acquire the to-be-matched capability input by the target object, perform a first vectorization process on the to-be-matched capability, and obtain a to-be-matched vector;

[0006] Based on the vector to be matched, the historical capabilities to be matched and the preset atomic capability vector library, candidate atomic capabilities are obtained through similarity matching;

[0007] According to the capabilities to be matched and the candidate atomic capabilities, the recommendation results are output through the preset large language model;

[0008] The recommendation results include the capability name and recommendation reason corresponding to each candidate atomic capability;

[0009] In response to the selection instruction of the target object, determining a final recommendation result;

[0010] The adoption result of the recommendation result is obtained according to the final recommendation result verification, and the model parameters of the large language model are adjusted based on the adoption result; wherein the adjusted large language model is applied to the subsequent atomic capability recommendation.

[0011] Optionally, performing a first vectorization process on the to-be-matched capability to obtain a to-be-matched vector includes the following steps:

[0012] The preset text vector model is used to perform text vectorization operation on the to-be-matched capability to obtain the to-be-matched vector.

[0013] Optionally, the method further comprises the following steps:

[0014] Get the atomic capabilities list;

[0015] Based on the atomic capability list, a preset text vector model is used to perform text vectorization operations with a single atomic capability as the dimension to obtain the atomic capability text vector corresponding to each single atomic capability;

[0016] All atomic capability text vectors are stored in a preset vector database and sorted to obtain an atomic capability vector library.

[0017] Optionally, based on the vector to be matched, combined with the historical capabilities to be matched and a preset atomic capability vector library, candidate atomic capabilities are obtained through similarity matching, including the following steps:

[0018] Perform a split operation on the vector to be matched to obtain multiple sub-vectors;

[0019] Based on the sub-vector, the text similarity between the vector to be matched and each atomic capability text vector in the atomic capability vector library is obtained by a preset similarity algorithm; the preset similarity algorithm includes vector distance calculation and asymmetric calculation;

[0020] Acquire a first number of atomic capability text vectors from the atomic capability vector library as vector search results according to the order of text similarity from high to low;

[0021] Obtain all historical capabilities to be matched and their corresponding historical recommended atomic capabilities;

[0022] Based on the sub-vector, the matching similarity between the vector to be matched and each historical capability to be matched is obtained through a preset similarity algorithm;

[0023] Acquire historical recommended atomic capabilities corresponding to the second number of historical capabilities to be matched according to the order of matching similarity from high to low as vector matching results;

[0024] Based on the vector matching results, the text similarity corresponding to each atomic capability text vector in the vector retrieval results is adjusted by inverted indexing;

[0025] A third number of atomic capability text vectors are obtained from the vector retrieval results in descending order based on the adjusted text similarity as candidate atomic capabilities.

[0026] Optionally, outputting a recommendation result through a preset large language model according to the capability to be matched and the candidate atomic capability includes the following steps:

[0027] The structured text is constructed based on the capabilities to be matched and the candidate atomic capabilities in combination with the preset text template;

[0028] Based on structured text, the preset large language model is used to output recommendation results.

[0029] Optionally, the selection instruction includes a first selection instruction and a second selection instruction; and in response to the selection instruction of the target object, determining the final recommendation result includes the following steps:

[0030] In response to the first selection instruction of the target object, determining a final recommendation result among the recommendation results;

[0031] Alternatively, in response to the second selection instruction of the target object, the final recommendation result is determined by searching in the atomic capability vector library.

[0032] Optionally, obtaining an adoption result of the recommendation result according to the final recommendation result verification, and adjusting the model parameters of the large language model based on the adoption result, includes the following steps:

[0033] Verify and compare the final recommendation result with the recommended result to determine whether the final recommendation result exists in the candidate atomic capability;

[0034] Among them, when the final recommendation result exists in the candidate atomic capability, the adoption result also includes whether the recommendation reason corresponding to the final recommendation result is adopted;

[0035] Adjust the model parameters of the large language model based on the adoption results combined with the adoption results corresponding to the recommendation records of the historical capabilities to be matched within the pre-selected range;

[0036] Among them, the adjustment range of model parameters is negatively correlated with the degree of adoption of the adoption results.

[0037] On the other hand, an embodiment of the present invention provides an atomic capability recommendation device based on the LLM feedback mechanism, including:

[0038] The first module is used to obtain the to-be-matched capability input by the target object, and perform a first vectorization process on the to-be-matched capability to obtain a to-be-matched vector;

[0039] The second module is used to obtain candidate atomic capabilities through similarity matching based on the vector to be matched, the historical capabilities to be matched and the preset atomic capability vector library;

[0040] The third module is used to output the recommendation results through a preset large language model according to the capabilities to be matched and the candidate atomic capabilities;

[0041] The recommendation results include the capability name and recommendation reason corresponding to each candidate atomic capability;

[0042] The fourth module is used to determine the final recommendation result in response to the selection instruction of the target object;

[0043] The fifth module is used to verify the adoption result of the recommendation result according to the final recommendation result, and adjust the model parameters of the large language model based on the adoption result; wherein the adjusted large language model is applied to the subsequent atomic capability recommendation.

[0044] Optionally, the device further comprises:

[0045] Module 6, for obtaining an atomic capability list;

[0046] The seventh module is used to perform text vectorization operation based on the atomic capability list and a single atomic capability as a dimension using a preset text vector model to obtain an atomic capability text vector corresponding to each single atomic capability;

[0047] The eighth module is used to store all atomic capability text vectors into a preset vector database and organize them to obtain an atomic capability vector library.

[0048] On the other hand, an embodiment of the present invention provides an electronic device, including: a processor and a memory; the memory is used to store programs; the processor executes the program to implement the above-mentioned atomic capability recommendation method based on the LLM feedback mechanism.

[0049] On the other hand, an embodiment of the present invention provides a computer storage medium, in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to implement the above-mentioned atomic capability recommendation method based on the LLM feedback mechanism.

[0050] The embodiment of the present invention obtains the to-be-matched capability input by the target object, performs a first vectorization process on the to-be-matched capability, and obtains the to-be-matched vector; based on the to-be-matched vector, the candidate atomic capability is obtained by similarity matching in combination with the historical to-be-matched capability and the preset atomic capability vector library; according to the to-be-matched capability and the candidate atomic capability, the recommendation result is output through the preset large language model; wherein the recommendation result includes the capability name and recommendation reason corresponding to each candidate atomic capability; in response to the selection instruction of the target object, the final recommendation result is determined; according to the final recommendation result verification, the adoption result of the recommendation result is obtained, and the model parameters of the large language model are adjusted based on the adoption result; wherein the adjusted large language model is applied to the subsequent atomic capability recommendation. The embodiment of the present invention includes at least the following beneficial effects: the embodiment of the present invention significantly improves the intelligence level of the system and the user experience through innovative means such as efficient and accurate to-be-matched capability processing, intelligent candidate atomic capability recommendation, and personalized recommendation reason generation, and realizes efficient and accurate atomic capability recommendation; and based on the LLM feedback mechanism, it can realize model self-optimization and sustainable development of technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.

[0052] Figure 1 A schematic diagram of an implementation environment for atomic capability recommendation provided by an embodiment of the present invention;

[0053] Figure 2 A flowchart of an atomic capability recommendation method based on an LLM feedback mechanism provided by an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of an extended process of an atomic capability recommendation method based on an LLM feedback mechanism provided in an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of an example of the expansion process of step S200 provided in an embodiment of the present invention;

[0056] Figure 5 A schematic diagram of an example of the expansion process of step S300 provided in an embodiment of the present invention;

[0057] Figure 6 A schematic diagram of an example of the expansion process of step S400 provided in an embodiment of the present invention;

[0058] Figure 7 A schematic diagram of an example of the architecture principle of an atomic capability recommendation process provided in an embodiment of the present invention;

[0059] Figure 8 A schematic diagram of a specific application process example of the atomic capability recommendation process provided in an embodiment of the present invention;

[0060] Fig. 9 A schematic diagram of the structure of an atomic capability recommendation device based on an LLM feedback mechanism provided by an embodiment of the present invention;

[0061] Fig.10 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0063] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100", "second / S200", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0064] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0065] To facilitate the understanding of the technical solution of the present invention, the technical feature terms that may appear in the embodiments of the present invention are first explained:

[0066] LLM (Large Language Model): refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, dialogue, etc., and are an important path to artificial intelligence.

[0067] AA (Atomic Ability): A technical concept that decomposes a specific technology or service into the smallest, independently deployable and manageable units; these units are usually highly standardized and modular and can be used alone or combined to provide more complex functions.

[0068] It is understandable that the atomic capability recommendation method based on the LLM feedback mechanism provided in the embodiment of the present invention can be applied to any computer device with data processing computing capabilities, and this computer device can be various types of terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, a tablet computer, a laptop computer, a desktop computer, etc., but is not limited to this.

[0069] like Figure 1 FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention. Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected to a network wirelessly or wired to complete data transmission and exchange.

[0070] Server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0071] In addition, the server 101 can also be a node server in the blockchain network. Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm.

[0072] The terminal 102 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal 102 and the server 101 may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiment of the present invention.

[0073] Based on the example Figure 1 In the implementation environment shown, an embodiment of the present invention provides an atomic capability recommendation method based on the LLM feedback mechanism. The following is explained using the example of the atomic capability recommendation method based on the LLM feedback mechanism being applied to the terminal 102. It can be understood that the atomic capability recommendation method based on the LLM feedback mechanism can also be applied to the server 101.

[0074] Reference Figure 2 , Figure 2 The flowchart of the atomic capability recommendation method based on the LLM feedback mechanism applied to the terminal provided by the embodiment of the present invention, the execution subject of the atomic capability recommendation method based on the LLM feedback mechanism can be any of the aforementioned computer devices (including servers or terminals). Figure 2 , the method comprises the following steps:

[0075] S100, obtaining the to-be-matched capability input by the target object, performing a first vectorization process on the to-be-matched capability, and obtaining a to-be-matched vector;

[0076] It should be noted that, in some embodiments, performing a first vectorization process on the capability to be matched to obtain a vector to be matched may include the following steps: performing a text vectorization operation on the capability to be matched using a preset text vector model to obtain a vector to be matched.

[0077] For example, in some specific implementations, for the capability data to be matched X n , you can use the GTE text vector-Qwen1.5 model to perform text vectorization operations and obtain vectorized matching capabilities

[0078] S200, based on the to-be-matched vector, combined with the historical to-be-matched capabilities and the preset atomic capability vector library, obtain candidate atomic capabilities through similarity matching;

[0079] Among them, in some embodiments, such as Figure 3 As shown, the method may also include the following steps: T100, obtaining an atomic capability list; T200, based on the atomic capability list, using a preset text vector model to perform text vectorization operations with a single atomic capability as a dimension, to obtain an atomic capability text vector corresponding to each single atomic capability; T300, storing all atomic capability text vectors into a preset vector database, and organizing them to obtain an atomic capability vector library.

[0080] For example, in some specific implementations, for the atomic capability inventory data Y n , can be a single atomic capacity Y i Use the GTE text vector-Qwen1.5 model to perform text vectorization operations for the dimension and obtain the atomic capability of vectorization Then, the atomic capability text vector Stored in the faiss vector database.

[0081] It should be noted that in some embodiments, Figure 4As shown, step S200 may include the following steps: S201, splitting the vector to be matched to obtain multiple sub-vectors; S202, based on the sub-vectors, obtaining the text similarity between the vector to be matched and each atomic capability text vector in the atomic capability vector library through a preset similarity algorithm; the preset similarity algorithm includes vector distance calculation and asymmetric calculation; S203, obtaining a first number of atomic capability text vectors from the atomic capability vector library as vector retrieval results according to the order of text similarity from high to low; S204, obtaining all historical capabilities to be matched and their corresponding historical recommended atomic capabilities; S205, obtaining the text similarity between the vector to be matched and each atomic capability text vector in the atomic capability vector library; S206, obtaining the text similarity between the vector to be matched and each atomic capability text vector in the atomic capability vector library; S207, obtaining the text similarity between the vector to be matched and the atomic capability text vector in the atomic capability vector library; S208, obtaining the text similarity between the vector to be 205. Based on the sub-vector, obtain the matching similarity between the vector to be matched and each historical capability to be matched through a preset similarity algorithm; S206. Obtain the historical recommended atomic capabilities corresponding to the second number of historical capabilities to be matched as the vector matching result in the order of matching similarity from high to low; S207. Based on the vector matching result, adjust the text similarity corresponding to each atomic capability text vector in the vector retrieval result by means of an inverted index; S208. Obtain the third number of atomic capability text vectors from the vector retrieval result as candidate atomic capabilities in the order of the adjusted text similarity from high to low.

[0082] For example, in some specific implementations, step S200 may be implemented through the following process:

[0083] Step S2-1: Perform vector search on the atomic capability library according to the capability to be matched:

[0084] Step S2-1-1: Split the vector X to be matched into M sub-vectors, and calculate the m The distance from the kth cluster center in the atomic capability library, where K cluster centers are randomly selected from the atomic capability vector library in the initialization phase, that is, the integrated distance matrix is ​​obtained:

[0085]

[0086] Where X m represents the mth subvector of the vector to be matched, Represents the mth sub-vector and the corresponding kth cluster center vector in the atomic capability library.

[0087] Step S2-1-2: The distance calculation formula between the to-be-matched vector X and the atomic capability vector Y is as follows:

[0088]

[0089] in, It is composed of the cluster center vectors of M sub-vectors. It represents the mth sub-vector in the i-th vector to be matched and the corresponding kth cluster center vector in the atomic capability library, and A is the distance matrix.

[0090] Step S2-1-3: Use asymmetric calculation to calculate vector distance, i.e. text similarity. The calculation formula is as follows:

[0091]

[0092] Step S2-2: Introduce historical recommendation results to fine-tune the vector retrieval results of S2-1:

[0093] Step S2-2-1: Obtain the historical capability to be matched X′ and its corresponding recommended atomic capability Y′;

[0094] Step S2-2-2: Calculate the similarity between the capability to be matched X and the historical capability to be matched X′:

[0095]

[0096] Step S2-2-3: storing the atomic capabilities H corresponding to the first t historical capabilities to be matched corresponding to the capability to be matched X;

[0097] Step S2-3: Inverted index to speed up vector query L:

[0098] Step S2-3-1: Cluster the atomic capability vectors into k′ clusters using k-means, and each cluster is represented by a cluster center vector;

[0099] Step S2-3-2: Find the w cluster centers closest to the vector to be matched X;

[0100] Step S2-3-3: For each cluster center, calculate the residual vector between the vector to be matched and the cluster center;

[0101] Step S2-3-4: traverse the postings chain of each cluster center, and calculate the distance D between the vector to be matched and the candidate atomic capability vector on the postings chain according to steps S2-2-1 to S2-2-3;

[0102] Step S2-3-5: Fine-tune the distance D of each candidate atomic capability vector according to the atomic capability H to obtain D′:

[0103] D′=σd h

[0104] where d h Indicates the appearance of the candidate atomic vector in H, σ is the adjustment factor, ranging from 0 to 1;

[0105] Step S2-4: Obtain the candidate atomic capabilities Y of the top k similarities (minimum distance) D′ k .

[0106] S300: Outputting recommendation results through a preset large language model according to the capabilities to be matched and the candidate atomic capabilities.

[0107] The recommendation results include the capability name and recommendation reason corresponding to each candidate atomic capability;

[0108] It should be noted that in some embodiments, Figure 5 As shown, step S300 may include the following steps: S301, constructing a structured text based on the capability to be matched and the candidate atomic capability in combination with a preset text template; S302, outputting a recommendation result based on the structured text using a preset large language model.

[0109] For example, in some specific implementations, a prompt template may be constructed (a prompt template is a structured text used to guide a large language model to generate a specific type of response), which may be implemented as follows:

[0110] Step S3-1-1: Based on the capability to be matched X and the candidate atomic capability Y k Define the prompt template:

[0111] Prompt = "You are a professional capability recommendation expert. Based on capability description X, select candidate capability Y. k Select the first t closest capabilities and give the reasons for recommendation. The recommendation result is output in the following form {ability name: recommendation reason}”;

[0112] Step S3-1-2: Use LLM to recommend atomic capabilities and output reasons for recommendations. The model is defined as follows:

[0113]

[0114] in is the LLM model, where v is the word embedding, c is the parameter of the LLM model, (p x:Y ) is a large model input in word units, including the capability to be matched x and the candidate atomic capability Y.

[0115] S400, in response to a selection instruction of a target object, determining a final recommendation result;

[0116] It should be noted that the selection instruction includes a first selection instruction and a second selection instruction; in some embodiments, such as Figure 6 As shown, step S400 may include the following steps: S401, in response to a first selection instruction of the target object, determining a final recommendation result in the recommendation results; or, S402, in response to a second selection instruction of the target object, searching in the atomic capability vector library to determine the final recommendation result.

[0117] S500: Verify the final recommendation result to obtain an adoption result of the recommendation result, and adjust the model parameters of the large language model based on the adoption result.

[0118] Among them, the adjusted large language model is used for subsequent atomic capability recommendations;

[0119] It should be noted that, in some embodiments, step S500 may include the following steps: verifying and comparing the final recommendation result with the recommendation result to determine whether the final recommendation result exists in the candidate atomic capability; wherein, when the final recommendation result exists in the candidate atomic capability, the adoption result also includes whether the recommendation reason corresponding to the final recommendation result is adopted; adjusting the model parameters of the large language model based on the adoption result combined with the adoption result corresponding to the recommendation record of the historical capabilities to be matched within the pre-selected range; wherein the adjustment range of the model parameters is negatively correlated with the adoption degree of the adoption result.

[0120] For example, in some specific implementations, the model can be fine-tuned based on the feedback of the user adoption results, which can be specifically implemented as follows:

[0121] First, accumulate actual recommendation records, clean the data, and then provide feedback for model fine-tuning:

[0122] Step S4-1: Recording user selections through a third-party interface Yrec_a :h rec_c}, and send it back to the capability feedback module;

[0123] Step S4-2: Cleaning the user selection data;

[0124] Step S4-2-1: Verify whether the user's final selection is in the original recommendation list. If so, adjust the final selection to the first priority; if not, clear the candidate content and replace it with the user's final selection;

[0125] Step S4-2-2: Verify whether the user's final option recommendation reason is adopted. If so, keep the recommendation reason; if not, clear the recommendation reason;

[0126] Step S4-2-3: Perform data storage operation on all cleaned data;

[0127] Step S4-3: Monitor and count the stored data. If the number is greater than L mnt (You can customize the setting, such as 1000), distribute the data for cyclic iteration model effect.

[0128] In some preferred implementations, the principle of model fine-tuning can be implemented as follows:

[0129] Step S5-1: Obtain the actual recommendation result related data set {X′|(h Yrec_a :h rec_c ),h his_a}, where h Yrec_a is the recommended capability based on the actual application of matching capability, h rec_c Yes Yrec_a The corresponding reason for recommendation, h his_a is the initial recommended capacity;

[0130] Step S5-2: Based on the data set {X′|(h Yrec_a :h rec_c ),h his_a}Construct a fine-tuning sample set with the following structure:

[0131]

[0132] Step S5-3: Fine-tune the LLM model using the lora method. Specifically, the core of the lora (i.e. LoRA, Low-Rank Adaptation) model adjustment is to optimize model performance by adjusting a series of parameters while reducing computing and memory requirements. LoRA is a method that uses low-rank adaptation to adapt downstream tasks by training a small number of new parameters without fine-tuning the parameters of the entire large model, thereby reducing computing and memory requirements while maintaining model performance.

[0133] In order to explain the principle of the technical solution of the present invention in detail, the overall process of the present invention is described below in combination with some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and cannot be regarded as a limitation of the present invention.

[0134] In view of the relevant shortcomings of the prior art, the present invention proposes an atomic capability recommendation method based on the LLM feedback mechanism. By applying the language large model basic capabilities and capability feedback mechanism, the capability items in the solution are retrieved and recommended in the atomic capability library, and recommendation opinions are provided for reference to solve the problems of low matching rate of atomic capability recommendation and lack of visual matching decision-making basis.

[0135] First of all, it should be noted that in some specific application scenarios, the atomic capability recommendation process of the embodiment of the present invention can be implemented as follows: Figure 7 The functional modules shown are implemented as follows:

[0136] 1. Preprocessing module:

[0137] The input of this module is structured text (including at least capability name, capability description and other auxiliary text content), and text vectorization operation is performed on the related features of each capability for subsequent matching recommendations. The structured text can be atomic vector base library data and descriptions related to the capability to be matched.

[0138] 2. Atomic capability module:

[0139] The main purpose of this module is to provide vectorized storage and retrieval functions for atomic capabilities. The storage structure is designed for multiple dimensions of atomic capabilities, and the capability is represented by a unique ID; the retrieval capability supports semantic-level retrieval based on content, and also supports the function of adjusting the retrieval results based on historical recommended atomic capability experience.

[0140] 3.LLM recommended modules:

[0141] This module combines the pre-retrieved atomic capability candidate set with the prompt template to implement LLM recommendation capabilities, and uses LLM's text generation capabilities to give recommendations from the perspective of the application of recommended atomic capabilities. In addition, it receives the content of the capability feedback module, performs fine-tuning, and synchronously adjusts the atomic capability recommendation order and recommendation suggestions.

[0142] 4. Capability feedback module:

[0143] This module accumulates the actual selection data details of users through the user feedback mechanism. After data cleaning, it is fed back to the atomic capability module and LLM recommendation module for cyclic fine-tuning and optimization of the overall recommendation data.

[0144] like Figure 8 As shown, taking the above atomic capability recommendation process as an example of a specific application scenario, the main process of the technical solution of the present invention can be implemented as follows:

[0145] Step S1: Based on the atomic capability list, construct an atomic capability vector library; vectorize the capabilities to be matched:

[0146] Step S1-1: For the atomic capability list data Y n , with a single atomic capacity Y i Use the GTE text vector-Qwen1.5 model to perform text vectorization operations for the dimension and obtain the atomic capability of vectorization

[0147] Step S1-2: For the capability data X to be matched n , use the GTE text vector-Qwen1.5 model to perform text vectorization operations and obtain the vectorized matching capability

[0148] Step S1-3: Convert atomic capability text vector Stored in the faiss vector database.

[0149] Step S2: Retrieve candidate atomic capabilities from the atomic capability vector library based on the capabilities to be matched and historical recommendation results:

[0150] Step S2-1: Perform vector search on the atomic capability library according to the capability to be matched:

[0151] Step S2-1-1: Split the vector X to be matched into M sub-vectors, and calculate the m The distance from the kth cluster center in the atomic capability library, where K cluster centers are randomly selected from the atomic capability vector library in the initialization phase, that is, the integrated distance matrix is ​​obtained:

[0152]

[0153] Where X m represents the mth subvector of the vector to be matched, Represents the mth sub-vector and the corresponding kth cluster center vector in the atomic capability library.

[0154] Step S2-1-2: The distance calculation formula between the to-be-matched vector X and the atomic capability vector Y is as follows:

[0155]

[0156] in, It is composed of the cluster center vectors of M sub-vectors. It represents the mth sub-vector in the i-th vector to be matched and the corresponding kth cluster center vector in the atomic capability library, and A is the distance matrix.

[0157] Step S2-1-3: Use asymmetric calculation to calculate vector distance, i.e. text similarity. The calculation formula is as follows:

[0158]

[0159] Step S2-2: Introduce historical recommendation results to fine-tune the vector retrieval results of S2-1:

[0160] Step S2-2-1: Obtain the historical capability to be matched X′ and its corresponding recommended atomic capability Y′;

[0161] Step S2-2-2: Calculate the similarity between the capability to be matched X and the historical capability to be matched X′:

[0162]

[0163] Step S2-2-3: storing the atomic capabilities H corresponding to the first t historical capabilities to be matched corresponding to the capability to be matched X;

[0164] Step S2-3: Inverted index to speed up vector query L:

[0165] Step S2-3-1: Cluster the atomic capability vectors into k′ clusters using k-means, and each cluster is represented by a cluster center vector;

[0166] Step S2-3-2: Find the w cluster centers closest to the vector to be matched X;

[0167] Step S2-3-3: For each cluster center, calculate the residual vector between the vector to be matched and the cluster center;

[0168] Step S2-3-4: traverse the postings chain of each cluster center, and calculate the distance D between the vector to be matched and the candidate atomic capability vector on the postings chain according to steps S2-2-1 to S2-2-3;

[0169] Step S2-3-5: Fine-tune the distance D of each candidate atomic capability vector according to the atomic capability H to obtain D′:

[0170] D′=σd h ;

[0171] where d h Indicates the appearance of the candidate atomic vector in H, σ is the adjustment factor, ranging from 0 to 1;

[0172] Step S2-4: Obtain the candidate atomic capabilities Y of the top k similarities (minimum distance) D′ k ;

[0173] Step S3: The candidate atomic capability Y output from step S2 k And prompt templates to recommend atomic capabilities and output reasons for recommendations; fine-tune the model based on S4 feedback results:

[0174] Step S3-1: Build the prompt template:

[0175] Step S3-1-1: Based on the capability to be matched X and the candidate atomic capability Y k Define the prompt template:

[0176] Prompt = "You are a professional capability recommendation expert. Based on capability description X, select candidate capability Y. k Select the first t closest capabilities and give the reasons for recommendation. The recommendation result is output in the following form {ability name: recommendation reason}”;

[0177] Step S3-1-2: Use LLM to recommend atomic capabilities and output reasons for recommendations. The model is defined as follows:

[0178]

[0179] in is the fine-tuned LLM model, where v is the word embedding, c is the fine-tuned parameter, (p x:Y ) is a large model input in word units, including the capability to be matched x and the candidate atomic capability Y;

[0180] Step S3-2: Fine-tune the model based on the feedback results of S4:

[0181] Step S3-2-1: Obtain the actual recommendation result related data set {X′|(h Yrec_a :h rec_c ),h his_a}, where h Yrec_a is the recommended capability based on the actual application of matching capability, h rec_c Yes Yrec_a The corresponding reason for recommendation, h his_a is the initial recommended capacity;

[0182] Step S3-2-2: Based on the data set {X′|(h Yrec_a :h rec_c ),h his_a}Construct a fine-tuning sample set with the following structure:

[0183]

[0184] Step S3-2-3: Fine-tune the model of step S3-1-2 using the LoRa method;

[0185] Step S3-3: According to Obtain atomic capability recommendations and reasons for recommendation Yres_a :h res_c};

[0186] Step S4: Accumulate actual recommendation records, clean the data and then feed it back to steps S2 and S3 for model fine-tuning:

[0187] Step S4-1: Recording user selections through a third-party interface Yrec_a :h rec_c}, and send it back to the capability feedback module;

[0188] Step S4-2: Cleaning the user selection data;

[0189] Step S4-2-1: Verify whether the user's final selection is in the original recommendation list. If so, adjust the final selection to the first priority; if not, clear the candidate content and replace it with the user's final selection;

[0190] Step S4-2-2: Verify whether the user's final option recommendation reason is adopted. If so, keep the recommendation reason; if not, clear the recommendation reason;

[0191] Step S4-2-3: Perform data storage operation on all cleaned data;

[0192] Step S4-3: Monitor and count the stored data. If the number is greater than L mnt , distribute the data to step S2 and step S3, loop and iterate the model effect.

[0193] Furthermore, in some specific application scenarios, the present invention is used for the application of its own capabilities in the deconstruction link of the telecommunication solution. The following is a detailed example of the recommended steps of atomic capabilities, which cannot be regarded as a limitation on the technical solution of the present invention. The specific implementation process can be:

[0194] Step S1: the capability to be matched is "unified converged communication scheduling system";

[0195] Step S2: Retrieve candidate atomic capabilities from the atomic capability vector library based on the capabilities to be matched and historical recommendation results;

[0196] Step S2-1: The candidate capabilities matched from the atomic capability vector library based on the "unified converged communication scheduling system" are as shown in Table 1 below:

[0197] Table 1

[0198]

[0199]

[0200] Step S2-2: Matching historical data from historical to-be-matched vectors based on the "unified converged communication scheduling system", i.e., {"unified geographic information service", ...};

[0201] Step S2-3: Store the historical atomic capability data corresponding to the historical vectors to be matched as shown in Table 2:

[0202] Table 2

[0203]

[0204] Step S2-4: Merge the candidate atomic capabilities of step S2-2 and step S2-3;

[0205] Step S3: Recommend atomic capabilities and output reasons for the candidate atomic capabilities and prompt templates output in step S2; fine-tune the model according to the feedback results of S4;

[0206] Step S3-1: construct a prompt template;

[0207] Step S3-1-1: Based on the capability to be matched X and the candidate atomic capability Y k Define the prompt template:

[0208] Prompt = "You are a professional capability recommendation expert. According to the capability description "Unified Converged Communication Dispatch System", select the top 5 closest capabilities from the candidate capabilities "Converged Communication Command and Dispatch Platform Service" (the five-level capability labels are General, Visual Networking, Audio and Video Application, Audio and Video Collection and Upload, Audio), "Converged Communication Dispatch System" (the five-level capability labels are General Emergency, Emergency Industry, Emergency Command and Rescue and Supervision and Law Enforcement, Command and Rescue, Converged Communication), "...", and give the reasons for the recommendation. The recommendation results are output in the following format {Capability Name: Recommendation Reason}";

[0209] Step S3-1-2: Use LLM to recommend atomic capabilities and output reasons for recommendations. An example is as follows:

[0210] {Capability name: Converged communication command and dispatch platform service, recommendation reason: the capability text has high similarity, the capability description does not specify the scenario, and general scenario capabilities are recommended first};

[0211] Step S3-2: fine-tune the model according to the feedback result of S4;

[0212] Step S3-2-1: Fine-tune the sample set. The example is as follows:

[0213] {

[0214] “instruction”: “Select relevant capabilities and give reasons”,

[0215] "input": "You are a professional capability recommendation expert. According to the capability description "Unified Converged Communication Dispatch System", select the top 5 closest capabilities from the candidate capabilities "Converged Communication Command and Dispatch Platform Service" (the five-level capability labels are General, Visual Networking, Audio and Video Application, Audio and Video Collection and Upload, Audio), "Converged Communication Dispatch System" (the five-level capability labels are General Emergency, Emergency Industry, Emergency Command and Rescue and Supervision and Law Enforcement, Command and Rescue, Converged Communication), "...", and give reasons for the recommendation. The recommendation results are output in the following format {Capability Name: Recommendation Reason}",

[0216] “output”: “{Capability name: Converged Communication Command and Dispatch Platform Service, Recommendation reason: Capability text similarity is high, capability description does not specify the scenario, general scenario capabilities are recommended first}”

[0217] };

[0218] Step S3-2-2: Please refer to the previous embodiment for the fine-tuning process, which will not be described here in detail;

[0219] Step S4: accumulate actual recommendation records, perform data cleaning and then feed back to step S2 and step S3 for model fine-tuning. The data cleaning process is described in detail in the above embodiment and will not be repeated here.

[0220] In summary, compared to the application of traditional text similarity models to deal with the atomic capability recommendation problem, the present invention adopts a method of combining large model technology with atomic capability recommendation to provide accurate recommendation services in the professional field of atomic capabilities; and, compared with traditional atomic capability recommendations, the present invention only recommends based on current matching capabilities, integrates historical recommendation records, and improves the accuracy of atomic capability recommendations by increasing matching dimensions; at the same time, the present invention provides recommendation reasons for recommended atomic capabilities by applying large model technology to improve recommendation perception. Compared with the prior art, the present invention includes at least the following beneficial effects:

[0221] Efficient and accurate processing of capabilities to be matched: By first vectorizing the capabilities to be matched, complex input information is converted into a vector form that is easy to calculate and match, greatly improving processing efficiency and matching accuracy. This vectorization processing method can capture the key features of the capabilities to be matched and provide a solid foundation for subsequent similarity matching.

[0222] Intelligent candidate atomic capability recommendation: Based on the vector to be matched, combined with the historical capabilities to be matched and the preset atomic capability vector library, the candidate atomic capability closest to the capability to be matched can be quickly screened out through the similarity matching algorithm. This process not only improves the intelligence level of the recommendation, but also dynamically adjusts the recommendation strategy according to the changes in historical data and user needs to ensure the accuracy and effectiveness of the recommendation.

[0223] Personalized recommendation reason generation: Using the preset large language model, personalized recommendation reasons can be automatically generated based on the capabilities to be matched and the candidate atomic capabilities. These recommendation reasons can not only help users better understand the characteristics and advantages of the candidate atomic capabilities, but also provide more targeted suggestions and guidance based on user preferences and needs.

[0224] Improve user experience and satisfaction: Through the above intelligent recommendation and personalized reason generation, the present invention can provide users with a more convenient, efficient and personalized service experience. When faced with complex capabilities to be matched, users can quickly find the most suitable atomic capabilities, thereby improving work efficiency and satisfaction.

[0225] Realize self-optimization of the model: Determine the final recommendation result according to the selection instructions of the target object, and adjust the model parameters of the large language model based on the adoption results. This process realizes the self-learning and optimization of the model, so that the model can continuously adapt to the changes in the needs of the target object and improve the accuracy and satisfaction of the recommendation.

[0226] In summary, the embodiments of the present invention significantly improve the intelligence level of the system and user experience through innovative means such as efficient and accurate processing of capabilities to be matched, intelligent recommendation of candidate atomic capabilities, and generation of personalized recommendation reasons, and achieve efficient and accurate atomic capability recommendation; and based on the LLM feedback mechanism, it can achieve model self-optimization and sustainable development of technology.

[0227] On the other hand, Fig. 9 As shown, the embodiment of the present invention provides an atomic capability recommendation device 900 based on the LLM feedback mechanism, which may include:

[0228] The first module 901 is used to obtain the to-be-matched capability input by the target object, and perform a first vectorization process on the to-be-matched capability to obtain a to-be-matched vector;

[0229] The second module 902 is used to obtain candidate atomic capabilities through similarity matching based on the vector to be matched, the historical capabilities to be matched and the preset atomic capability vector library;

[0230] The third module 903 is used to output a recommendation result through a preset large language model according to the capability to be matched and the candidate atomic capability;

[0231] The recommendation results include the capability name and recommendation reason corresponding to each candidate atomic capability;

[0232] The fourth module 904 is used to determine the final recommendation result in response to the selection instruction of the target object;

[0233] The fifth module 905 is used to verify the adoption result of the recommendation result according to the final recommendation result, and adjust the model parameters of the large language model based on the adoption result; wherein the adjusted large language model is applied to the subsequent atomic capability recommendation.

[0234] Optionally, the device may further include:

[0235] Module 6, for obtaining an atomic capability list;

[0236] The seventh module is used to perform text vectorization operation based on the atomic capability list and a single atomic capability as a dimension using a preset text vector model to obtain an atomic capability text vector corresponding to each single atomic capability;

[0237] The eighth module is used to store all atomic capability text vectors into a preset vector database and organize them to obtain an atomic capability vector library.

[0238] The contents of the method embodiments of the present invention are all applicable to the device embodiments. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0239] On the other hand, an embodiment of the present invention further provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above sensitive information method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0240] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0241] like Fig.10 As shown, Fig.10 A specific example of the hardware structure of an electronic device 1000 according to an embodiment is shown. The electronic device 1000 includes:

[0242] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0243] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the network node population optimization method of the embodiment of the present invention;

[0244] Input / output interface 1003, used to implement information input and output;

[0245] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0246] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0247] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0248] The electronic device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0249] The contents of the method embodiments of the present invention are all applicable to the electronic device embodiments. The functions specifically implemented by the electronic device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0250] Another aspect of an embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the above method.

[0251] It should be noted that the computer-readable medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read to Only Memory, CD to ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0252] The contents of the method embodiments of the present invention are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0253] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0254] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0255] It should be noted that, although several modules of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0256] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD to ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present invention.

[0257] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0258] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0259] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the solution of the present invention, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0260] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution device, apparatus or device (such as a computer-based device, a device including a processor, or other device that can fetch instructions from an instruction execution device, apparatus or device and execute the instructions), or in conjunction with such instruction execution device, apparatus or device. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution device, apparatus or device, or in conjunction with such instruction execution device, apparatus or device.

[0261] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0262] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0263] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0264] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0265] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. An atomic capability recommendation method based on LLM feedback mechanism, characterized in that: The following steps are involved: Acquire the capability to be matched input by the target object, and perform a first vectorization process on the capability to be matched to obtain a vector to be matched; Based on the vector to be matched, combined with historical capabilities to be matched and a preset atomic capability vector library, candidate atomic capabilities are obtained through similarity matching; Outputting a recommendation result through a preset large language model according to the capability to be matched and the candidate atomic capability; The recommendation result includes the capability name and recommendation reason corresponding to each candidate atomic capability; In response to the selection instruction of the target object, determining a final recommendation result; An adoption result of the recommendation result is obtained by verifying the final recommendation result, and a model parameter of the large language model is adjusted based on the adoption result; wherein the adjusted large language model is applied to subsequent atomic capability recommendations.

2. The atomic capability recommendation method based on the LLM feedback mechanism according to claim 1 is characterized in that: The first vectorization processing is performed on the capability to be matched to obtain a vector to be matched, comprising the following steps: A preset text vector model is used to perform a text vectorization operation on the capability to be matched to obtain the vector to be matched.

3. The atomic capability recommendation method based on the LLM feedback mechanism according to claim 1 is characterized in that: The method further comprises the following steps: Get the atomic capabilities list; Based on the atomic capability list, a text vectorization operation is performed using a preset text vector model with a single atomic capability as a dimension to obtain an atomic capability text vector corresponding to each single atomic capability; All the atomic capability text vectors are stored in a preset vector database, and the atomic capability vector library is obtained by sorting them.

4. The atomic capability recommendation method based on the LLM feedback mechanism according to claim 1 is characterized in that: The method of obtaining candidate atomic capabilities by similarity matching based on the vector to be matched in combination with historical capabilities to be matched and a preset atomic capability vector library comprises the following steps: Performing a split operation on the vector to be matched to obtain multiple sub-vectors; Based on the sub-vector, obtaining the text similarity between the vector to be matched and each atomic capability text vector in the atomic capability vector library by a preset similarity algorithm; the preset similarity algorithm includes vector distance calculation and asymmetric calculation; Acquire a first number of the atomic capability text vectors from the atomic capability vector library as vector search results according to the order of the text similarity from high to low; Obtaining all the historical capabilities to be matched and their corresponding historical recommended atomic capabilities; Based on the sub-vector, obtaining the matching similarity between the vector to be matched and each of the historical capabilities to be matched by using the preset similarity algorithm; Acquire, according to the order of the matching similarities from high to low, the historical recommended atomic capabilities corresponding to a second number of the historical capabilities to be matched as vector matching results; Based on the vector matching result, adjusting the text similarity corresponding to each of the atomic capability text vectors in the vector search result by means of an inverted index; A third number of the atomic capability text vectors are obtained from the vector search results based on the adjusted text similarities in descending order as the candidate atomic capabilities.

5. The atomic capability recommendation method based on LLM feedback mechanism according to claim 1 is characterized in that: Outputting the recommendation result through a preset large language model according to the capability to be matched and the candidate atomic capability includes the following steps: Constructing a structured text according to the capability to be matched and the candidate atomic capability in combination with a preset text template; Based on the structured text, the recommendation result is outputted using the preset large language model.

6. The atomic capability recommendation method based on LLM feedback mechanism according to claim 1 is characterized in that: The selection instruction includes a first selection instruction and a second selection instruction; and determining the final recommendation result in response to the selection instruction of the target object includes the following steps: In response to the first selection instruction of the target object, determining the final recommendation result among the recommendation results; Alternatively, in response to the second selection instruction of the target object, the final recommendation result is determined by searching in the atomic capability vector library.

7. The atomic capability recommendation method based on LLM feedback mechanism according to claim 1 is characterized in that: The step of obtaining an adoption result of the recommendation result by verifying the final recommendation result, and adjusting the model parameters of the large language model based on the adoption result, comprises the following steps: Verify and compare the final recommendation result with the recommendation result to determine whether the final recommendation result exists in the candidate atomic capability; Wherein, when the final recommendation result exists in the candidate atomic capability, the adoption result also includes whether the recommendation reason corresponding to the final recommendation result is adopted; Adjusting the model parameters of the large language model based on the adoption result combined with the adoption result corresponding to the recommendation record of the historical to-be-matched capability within a preselected range; Among them, the adjustment range of the model parameters is negatively correlated with the adoption degree of the adoption results.

8. An atomic capability recommendation device based on LLM feedback mechanism, characterized in that: include: The first module is used to obtain the to-be-matched capability input by the target object, and perform a first vectorization process on the to-be-matched capability to obtain a to-be-matched vector; The second module is used to obtain candidate atomic capabilities through similarity matching based on the vector to be matched, in combination with historical capabilities to be matched and a preset atomic capability vector library; The third module is used to output a recommendation result through a preset large language model according to the capability to be matched and the candidate atomic capability; The recommendation result includes the capability name and recommendation reason corresponding to each candidate atomic capability; A fourth module is used to determine a final recommendation result in response to the selection instruction of the target object; The fifth module is used to verify the adoption result of the recommendation result according to the final recommendation result, and adjust the model parameters of the large language model based on the adoption result; wherein the adjusted large language model is applied to subsequent atomic capability recommendations.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images