Knowledge structure determination method and device of intelligent agent, equipment, medium and product

Abstract

The embodiment of the invention discloses an agent knowledge structure determination method and device, equipment, a medium and a product. The method comprises the steps that scene data and task data of a target agent in a current operation task scene are acquired; inputting the scene data and the task data into an agent knowledge structure recommendation model to obtain at least one recommendation knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; determining a scoring result corresponding to the recommended knowledge structure by using a preset scoring algorithm; and according to the scoring result, determining a target knowledge structure adopted by the target agent in the current operation task scene from the recommended knowledge structure. According to the technical scheme, the skills of the intelligent agent are accurately improved, and the ability of the intelligent agent to adapt to different industries and scenes is improved.

Description

Technical Field

[0001] This disclosure relates to the fields of artificial intelligence and intelligent agents, and in particular to a method, apparatus, device, medium, and product for determining the knowledge structure of an intelligent agent. Background Technology

[0002] Since the application of intelligent agent technology in large-scale models to empower industry applications, especially with the maturity and popularization of intelligent agent innovative application platforms, more and more intelligent agents have been created and integrated into business, accumulating rich usage records and evaluations. Current intelligent agent optimization techniques mainly rely on manual experience adjustments or extensive optimization based on overall usage / evaluation, lacking structured analysis methods for the core thinking capabilities of intelligent agents (multi-turn conversation organization, task chain logic planning, and task execution step design). Existing technologies mainly have the following limitations: Single evaluation dimension: Traditional methods optimize intelligent agents through macro-level indicators such as overall user satisfaction or click-through rate, failing to pinpoint specific thinking defects such as "insufficient coherence in multi-turn conversations" or "redundant task chain steps"; Vague knowledge structure: Intelligent agent skill texts are mostly unstructured natural language, making it difficult for machines to automatically identify key semantic modules such as "role definition - knowledge background - task instructions - example descriptions"; Low optimization efficiency: Manual analysis of massive interaction logs requires significant manpower and is difficult to form a reusable thinking structure template library. Summary of the Invention

[0003] This disclosure provides a method, apparatus, device, medium, and product for determining the knowledge structure of an intelligent agent, enabling precise improvement of the agent's skills and enhancing the agent's ability to adapt to different industries and scenarios.

[0004] Firstly, a method for determining the knowledge structure of an intelligent agent is provided, including: Acquire scene data and task data of the target intelligent agent in the current running task scenario; The scene data and the task data are input into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library. The scoring result corresponding to the recommended knowledge structure is determined using a preset scoring algorithm; Based on the scoring results, the target knowledge structure adopted by the target agent in the current running task scenario is determined from the recommended knowledge structure.

[0005] Secondly, a device for determining the knowledge structure of an intelligent agent is provided, comprising: The data acquisition module is used to acquire scene data and task data of the target intelligent agent in the current running task scenario; The recommended knowledge structure determination module is used to input the scene data and the task data into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; The scoring result determination module is used to determine the scoring result corresponding to the recommended knowledge structure using a preset scoring algorithm; The knowledge structure determination module is used to determine the target knowledge structure adopted by the target agent in the current running task scenario from the recommended knowledge structure based on the scoring results.

[0006] Thirdly, an electronic device is provided, comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the knowledge structure determination method for an intelligent agent as described in the first aspect above.

[0007] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the knowledge structure determination method for an intelligent agent as described in the first aspect above.

[0008] Fifthly, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the knowledge structure determination method for an intelligent agent as described in the first aspect above.

[0009] This disclosure provides a method, apparatus, device, medium, and product for determining the knowledge structure of an intelligent agent. The method includes: acquiring scene data and task data of a target intelligent agent in a current running task scenario; inputting the scene data and task data into an intelligent agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target intelligent agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; determining a scoring result corresponding to the recommended knowledge structure using a preset scoring algorithm; and determining the target knowledge structure adopted by the target intelligent agent in the current running task scenario from the recommended knowledge structures based on the scoring result. This technical solution obtains scene data and task data of a target intelligent agent in a specific task scenario, inputs them into an intelligent agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target intelligent agent; the recommended knowledge structures are knowledge structures defined in a preset knowledge structure library; a preset scoring algorithm is used to score these recommended knowledge structures; and finally, the target knowledge structure to be adopted by the target intelligent agent in the current task scenario is determined based on the scoring result. This achieves precise improvement of intelligent agent skills and enhances the ability of intelligent agents to adapt to different industries and scenarios.

[0010] It should be understood that the description in this section is not intended to identify key or important features of the embodiments of this disclosure, nor is it intended to limit the scope of the embodiments of this disclosure. Other features of the embodiments of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a method for determining the knowledge structure of an intelligent agent according to Embodiment 1 of this disclosure; Figure 2 This is a schematic diagram of a paragraph merging process provided in Embodiment 1 of this disclosure; Figure 3 This is a schematic diagram of the knowledge structure determination device for an intelligent agent provided in Embodiment 2 of this disclosure; Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this disclosure. Detailed Implementation

[0013] To enable those skilled in the art to better understand the solutions of the embodiments of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the protection scope of the embodiments of this disclosure.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Example 1 Figure 1 This is a flowchart of a method for determining the knowledge structure of an intelligent agent according to Embodiment 1 of this disclosure. This embodiment is applicable to situations involving the determination of the knowledge structure of an intelligent agent. This method can be executed by a device for determining the knowledge structure of an intelligent agent. This device can be implemented in hardware and / or software and can be configured in an electronic device, including but not limited to computers, PCs, electronic devices, and servers, which are devices with data processing capabilities. Figure 1 As shown, the method includes: S110. Obtain the scene data and task data of the target intelligent agent in the current running task scenario.

[0016] In this embodiment, the target intelligent agent can be an intelligent system running in a specific task scenario. The target intelligent agent typically needs to make decisions or perform operations based on its operating environment and task requirements. The current running task scenario can be the specific environment in which the intelligent agent is currently located and the background of the task it is performing.

[0017] Based on the above description, scenario data and task data of the target intelligent agent in the current running task scenario can be obtained. Scenario data can be data related to the environment in which the intelligent agent is located, and task data can be data related to the task currently being performed by the intelligent agent. For example, scenario data may include user industry, user scenario (task background, professional knowledge, precautions), etc., while task data may include task description data, task objective data, task requirement data, task chain information, multi-turn dialogue information, task planning information, usage information, and evaluation data.

[0018] S120. Input the scene data and task data into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in the preset knowledge structure library.

[0019] It is understood that after acquiring scene data and task data, the scene data can be input into the agent knowledge structure recommendation model. This model can then generate at least one recommended knowledge structure for the target agent. The agent knowledge structure recommendation model can be a pre-designed model whose purpose is to recommend suitable knowledge structures for the agent based on the input scene and task data. A knowledge structure can be understood as the knowledge system that the agent relies on when performing tasks.

[0020] Following the above description, the agent knowledge structure recommendation model can output at least one recommended knowledge structure. This recommended knowledge structure can be selected from a pre-defined knowledge structure library and can be considered the most suitable knowledge structure for the current scenario and task. For example, if the agent is an autonomous vehicle operating in a complex urban environment, the recommended knowledge structure might include an advanced traffic signal recognition model and an efficient path planning algorithm. The recommended knowledge structure can provide the agent with the knowledge and decision support needed to perform the current task, helping the agent better adapt to the current operating environment and improve the efficiency and accuracy of task execution.

[0021] S130. Use a preset scoring algorithm to determine the scoring results corresponding to the recommended knowledge structure.

[0022] In this embodiment, the preset scoring algorithm can be a pre-determined algorithm for scoring knowledge structures. For example, the preset scoring algorithm can be determined by weighting and summing the preset human scores, usage frequency, and evaluation data (user satisfaction) of the knowledge structure, and then using the weighted sum as the score for the knowledge structure. Furthermore, the preset scoring algorithm can also be a structure scoring model, which can be a pre-trained model for scoring knowledge structures.

[0023] Specifically, after obtaining the recommended knowledge structure, a preset scoring algorithm can be used to determine the scoring result corresponding to the recommended knowledge structure.

[0024] For example, the scoring metrics and weights are as follows: human scoring Usage Evaluation data Human rating: The rating of type B by bank customer service experts ( Points: It is believed that it can cover the questions of college students regarding materials); Usage: The frequency of use of Type B in the past month (2200 times, accounting for the total usage). (High usage); Rating results: User satisfaction with type B ( Negative feedback only accounted for a small percentage. (far below average); a comprehensive score can be calculated: Assuming usage is converted into "frequency percentage" Calculated in "points", 2200 5000 = 0.44, which can be converted to 2.2 points. The final score is adjusted to... The specific conversion rules can be set according to the business scenario.

[0025] S140. Based on the scoring results, determine the target knowledge structure adopted by the target agent in the current running task scenario from the recommended knowledge structure.

[0026] In this embodiment, after obtaining the scoring result, the scoring result can be compared with a preset scoring threshold, and the knowledge structure corresponding to the scoring result that is higher than the preset scoring threshold can be determined as the target knowledge structure adopted by the target intelligent agent in the current running task scenario.

[0027] For example, knowledge structure types can include type A: "Application Condition Consultation - Q&A - Condition Confirmation" (corresponding to "Application Condition Q&A" knowledge); type B: "Material Requirements Information - Upload Operation Guidance - Error Correction Reminder" (corresponding to "Material Operation Guidance" knowledge); and type C: "Approval Node Inquiry - Progress Feedback - Abnormal Issue Handling" (corresponding to "Approval Progress Follow-up" knowledge). Taking a "bank intelligent customer service agent" as an example, after obtaining the target knowledge structure of the target agent, the obtained target knowledge structure can be verified, such as: 1. A scoring threshold can be set: assuming the bank sets a "comprehensive scoring..." 3.5 points is the passing threshold. Type B (3.84 points) is higher than the threshold, so Type B can be determined as the target knowledge structure of the target intelligent agent. 2. Business adaptability verification: In the actual test of the "credit card application of a recent college graduate" scenario, the knowledge structure of Type B was used to guide the user, and it was found that "the time for answering questions about materials was shortened". The error rate of material upload has been reduced. ", fully adapted to the needs of this scenario, and type B can be determined as the target knowledge structure of the target intelligent agent; 3. Output results: For the scenario of "credit card application for recent college graduates", the optimal knowledge structure is output: type B (material requirements - upload operation guidance - error correction reminder).

[0028] This embodiment provides a method for determining the knowledge structure of an intelligent agent, including: acquiring scene data and task data of a target intelligent agent in the current running task scenario; inputting the scene data and task data into an intelligent agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target intelligent agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; determining the scoring result corresponding to the recommended knowledge structure using a preset scoring algorithm; and determining the target knowledge structure adopted by the target intelligent agent in the current running task scenario from the recommended knowledge structure based on the scoring result. The above technical solution achieves precise improvement of intelligent agent skills, enhancing the agent's ability to adapt to different industries and scenarios.

[0029] As an optional implementation of this embodiment, the method for determining the knowledge structure of an intelligent agent provided in this embodiment further includes: a process for determining an intelligent agent knowledge structure recommendation model, wherein the process for determining the intelligent agent knowledge structure recommendation model includes: 1) Determine the historical training dataset based on the agent's historical operation logs. The historical training dataset includes at least one historical training data, which includes historical scene data and historical task data.

[0030] Specifically, historical operation logs of the intelligent agent can be obtained. These logs are data generated during the agent's operation, recording its behavior, decisions, environmental states, and task execution at different points in time. Based on these logs, a historical training dataset can be determined. This dataset includes historical scenario data and historical task data. Historical scenario data relates to the environmental states during the agent's operation, describing the specific environmental conditions the agent encountered. Historical task data relates to the tasks performed by the agent, describing the specific task requirements the agent needed to complete during its historical operation.

[0031] For example, historical scenario data can include data such as historical user industry, historical user scenario (historical task background, professional knowledge, historical precautions), etc. Historical task data can include historical task description data, historical task objective data, historical task requirement data, historical task chain information, historical multi-turn dialogue information, historical task planning information, historical usage information, and evaluation data. Taking a bank's intelligent customer service agent for the "personal credit card application" scenario as an example, the historical user industry can be the industry in which the intelligent customer service agent operates. The historical task background can include: the user's purpose (e.g., the user wants to apply for a credit card or bank card online), and professional knowledge can be content such as credit card application conditions, application material requirements, and / or credit limit approval rules. The historical task description data can include the operation object (credit card application user); the tasks performed (guiding the user to fill in application information, answering material questions, and following up on the approval progress), etc. The historical task objective data can include: efficiently completing user application guidance, reducing the application failure rate, and task output requirements (generating user application progress reports and error message prompts). The historical task requirement data can include: response time less than a preset duration (e.g., 3 seconds), and answer accuracy greater than a preset value (e.g., ...). ) and the name and version of the main model. Historical task chain information may include: "User inquires about application requirements, guides the filling in of basic information, reminds users to upload materials (ID card, proof of income), answers questions about material review, and informs users of the approval progress." Historical multi-turn dialogue information may include: Example conversation 1 (User: "Can students apply?") Customer Service: "Full-time university students can apply for a campus credit card; a student ID is required." User: "What is the credit limit for the campus card?" Customer service: "Initial credit limit is 500 to 2000 yuan, which can be increased later based on card usage"); Example conversation 2 (User: "How do I get proof of income?") Customer service: "You can provide bank statements for the past 3 months or income certificates stamped by your employer; just take a photo and upload it." Historical task planning information can include "prioritizing answering questions about application requirements, then guiding users through information filling, and finally following up on approval; if users have material issues, they should be handled first." Historical usage information includes: the intelligent customer service has served 5000 users in this scenario in the past month, with the "guiding information filling" step being the most frequently used (3800 times). Evaluation data can include user satisfaction ratings (…). (Points), negative feedback mainly focused on "unclear explanation of material requirements" (accounting for a significant portion of negative feedback). ).

[0032] 2) For any historical training data, determine the knowledge structure label corresponding to the historical training data based on the existing knowledge structure in the preset knowledge structure base.

[0033] It is known that after obtaining historical training data, corresponding knowledge structure labels can be assigned to each historical training data based on the existing knowledge structures in the preset knowledge structure base.

[0034] 3) Use the first preset algorithm to extract scene features from the historical scene data.

[0035] In this embodiment, the first preset algorithm can be a pre-set algorithm for feature extraction, which can be used to extract scene features from the historical scene data.

[0036] For example, the first preset algorithm can be a Recurrent Convolutional Neural Network (RCNN), a text classification model that combines recurrent neural networks and convolutional neural networks. RCNN can capture contextual information of text through a bidirectional long short-term memory network, then extract local features using convolution operations, and finally fuse features through pooling operations, thereby achieving efficient text classification. The feature extraction capabilities of the RCNN model can be used to extract scene features from historical scene data, including global information about the user's industry and the scene.

[0037] 4) Use the second preset algorithm to extract the task features of the historical task data.

[0038] In this embodiment, the second preset algorithm can be a pre-set algorithm for feature extraction, which can be used to extract the task features of the historical task data.

[0039] For example, the second preset algorithm could be the ALBert model, an improved version of the BERT model designed to improve training and inference efficiency by reducing the number of model parameters. Bidirectional Encoder Representations from Transformers (BERT) is a deep learning model for pre-trained language representations. The core idea of ​​BERT is to learn general language features through pre-training on a large amount of unsupervised text data, and then fine-tune it on specific downstream tasks (such as text classification, sentiment analysis, question answering systems, etc.) to achieve efficient natural language processing tasks. Historical task data can be used as input, the ALBert model can learn features and perform deep encoding, outputting task features. These task features can include a comprehensive task feature vector containing user industry, scenario, task, and skill knowledge information.

[0040] 5) The scene features and the task features are concatenated to determine the joint features corresponding to the historical training data.

[0041] Specifically, after obtaining the scene features and task features, the scene features can be projected into a space with the same dimension as the task features. Then, the projected scene features and task features are concatenated. Finally, the concatenated features are flattened to obtain the joint features corresponding to the historical training data.

[0042] 6) Based on the joint features and knowledge structure labels corresponding to each historical training data, a preset loss function is used as the training objective, and a preset parameter tuning algorithm is used to adjust the model parameters of the initial agent knowledge structure recommendation model to determine the agent knowledge structure recommendation model.

[0043] In this embodiment, after obtaining the joint features corresponding to the historical training data, the model parameters of the initial agent knowledge structure recommendation model can be adjusted using the joint features and knowledge structure labels, with a preset loss function as the training objective, and a preset parameter tuning algorithm can be used to determine the agent knowledge structure recommendation model.

[0044] The preset loss function can be a pre-defined loss function, such as the multi-task classification cross-entropy loss function. The preset parameter tuning algorithm can be a pre-defined parameter tuning algorithm, such as the gradient descent algorithm. For example, the multi-task classification cross-entropy loss function is used as the training objective, while simultaneously optimizing the prediction accuracy of three recommendation types. The model parameters are adjusted using the gradient descent algorithm to achieve optimal performance.

[0045] For example, the initial intelligent agent knowledge structure recommendation model may include an input layer, a background layer, a task layer, a fusion layer, and an output layer. The input layer includes user-input industry and historical scene data, as well as historical task data. The background layer uses the TextRCNN model, removing its final classification output layer, and utilizes its feature extraction capabilities to obtain scene features of the user's industry and scene from global context and local pooling. The task layer uses historical task data as input, employs deep encoding with the ALBert model, and outputs comprehensive task features containing user industry, scene, task, and skill knowledge information. The fusion layer first projects the task features output from the background layer into a space consistent with the feature dimensions of the ALBert output through a fully connected layer, then concatenates the two along the feature dimensions, flattens the concatenated joint features, and feeds them into a subsequent fully connected network for processing. The output layer outputs recommendation results for three knowledge structure types: "task chain," "multi-turn conversation," and "task planning."

[0046] For example, the process of building a knowledge structure recommendation model may include: 1. Using "TextRCNN+ALBert" as the base model, taking "user industry (finance-retail banking), scenario (credit card application), and task description (guided application + answering material questions)" as input features; 2. The model learns the "matching relationship between features and knowledge structure types" in historical data (such as "scenarios with many material questions"). Type B has high adaptability. When the input requirement is "credit card application guidance for recent college graduates", the model outputs the optimal recommended structure type: Type B (material requirements are provided). Upload Operation Guide (Error correction reminder) The reason is that recent college graduates have the most questions about uploading "income certificates and academic materials", and type B is the most suitable.

[0047] As an optional implementation method of this embodiment, the process of determining the preset knowledge structure base includes: 1) Determine the historical interaction dataset based on the agent's historical operation logs.

[0048] Specifically, historical interaction datasets can be determined based on the agent's historical operation logs. These datasets can include multi-turn dialogue data, task chain data, and task planning information. Multi-turn dialogue data refers to the records of multiple dialogues between the agent and the user during a single interaction. This data typically includes contextual information about the dialogue and may include multi-turn dialogue text segments. Task chain data refers to the records of a series of sub-tasks executed by the agent to complete a user-requested task during a single interaction. This data reflects the task decomposition and execution process, usually involving the collaborative completion of multiple steps and sub-tasks. Task chain data may include natural language text segments within the task chain. Task planning information refers to the detailed plans and strategies formulated by the agent when executing a series of tasks. This information typically includes detailed task steps, execution order, required resources, expected results, and how to transition from one task to the next. Task planning information is crucial for the agent because it helps the agent manage and execute complex task sequences more effectively. For example, task planning information may include task decomposition, priority ranking, resource allocation, time planning, dependencies, and / or fault tolerance and recovery strategies.

[0049] 2) Use a preset clustering algorithm to perform clustering analysis on each historical interaction data in the historical interaction dataset to determine the clustering results.

[0050] It is known that after obtaining the historical interaction dataset, a pre-defined clustering algorithm can be used to perform cluster analysis on each historical interaction data in the dataset to determine the clustering results. The pre-defined clustering algorithm can be a bottom-up hierarchical clustering (AGglomerative NESting, AGNES) algorithm. This algorithm analyzes each historical interaction data in the dataset, dividing it into different clusters based on data characteristics (such as dialogue content, task type, user behavior, etc.). The clustering result is the output of the cluster analysis, indicating that the historical interaction data has been divided into multiple clusters, each containing similar historical interaction data.

[0051] 3) Determine the preset knowledge structure base based on the clustering results, wherein the preset knowledge structure base includes at least one knowledge structure.

[0052] Specifically, after obtaining the clustering results, a knowledge structure can be determined for each cluster based on the clustering results, thereby constructing a pre-defined knowledge structure library based on the knowledge structures determined for each cluster in the clustering results. The pre-defined knowledge structure library includes at least one knowledge structure.

[0053] For example, using clustering algorithms, structured text such as "Application Condition Consultation Paragraph," "Material Upload Question Paragraph," and "Application Process Execution Paragraph" are clustered. It is found that core skills knowledge can be divided into three categories: "Application Condition Answers," "Material Operation Guidance," and "Approval Progress Follow-up." Furthermore, three knowledge structure types can be defined: Type A: "Application Condition Consultation..." Q&A Condition Confirmation (corresponding to "Application Condition Q&A" knowledge); Type B: "Material Requirements Instructions" Upload Operation Guide Error Correction Alert (corresponding to "Material Handling Guidance" knowledge); Type C: "Approval Node Inquiry" Progress feedback "Handling of Abnormal Issues" (corresponding to "Approval Progress Follow-up" knowledge).

[0054] It should be explained that the scoring results corresponding to each knowledge structure in the preset knowledge structure base can also be determined. Specifically, after obtaining the agent's historical operation logs, typical data can be collected from the historical operation logs. From the collected typical data, a portion of the best cases can be extracted. Based on the industry, scenario, and task, business experts can be organized to conduct in-depth analysis and dimensional decomposition of the agent's specific performance in three core dimensions: multi-turn conversation, task chain planning, and task planning. Corresponding scoring standards can be formulated: Multi-turn conversation dimension: includes sub-dimensions such as dialogue coherence, accuracy of intent recognition, context maintenance ability, and naturalness of topic transition; Task chain planning dimension: covers sub-dimensions such as step logic, path efficiency, redundancy handling, and anomaly response mechanism; Task planning dimension: includes sub-dimensions such as goal clarity, step rationality, resource allocation optimization, and result predictability.

[0055] Based on the aforementioned sub-dimensions, scores can be assigned to each of the three aspects—multi-turn conversations, task chain planning, and task planning—according to preset scoring criteria and weight allocation schemes, and considering industry scenarios and business conditions. The scored data undergoes systematic review and correction. The specific process includes: randomly selecting a subset of labeled samples to verify their consistency with the defined scoring criteria, recording any labeled items with significant deviations; stratified sampling based on different industries, scenarios, and task types to ensure review coverage of all key dimensions and sub-dimensions; for data with significant differences between labeled data and preset judgment criteria, corrections can be made based on business logic and best practices for multi-turn conversations / task chains; calibration feedback and retraining: a calibration report is generated based on the review results, and relearning and label adjustments are performed based on feedback, which can then serve as a benchmark dataset for model training and evaluation.

[0056] Specifically, assuming the final personnel score is 'score', the agent usage count is 'count', and the highest customer rating is 'cscore', the score fusion parameter scParam = (count...) sum(all counts)) cscore, the combined score mscore=score scParam, the final score for each item is .

[0057] As an optional implementation of this embodiment, the step of determining the historical interaction dataset based on the agent's historical operation logs includes: 1) Determine the initial historical interaction data based on the agent's historical operation logs.

[0058] Specifically, the initial historical interaction data can be determined based on the agent's historical operation logs. The initial historical interaction data can be the raw, unprocessed interaction data.

[0059] 2) Use a text vector generation model to vectorize each initial historical interaction data to determine the text vector corresponding to each initial historical interaction data.

[0060] Specifically, after obtaining the initial historical interaction data, a text vector generation model can be used to vectorize each initial historical interaction data point to determine the corresponding text vector. This text vector generation model can be a model for vectorizing natural language segments. For example, a text vector generation model could be a BGE (BAAI General Embedding) model, a general text embedding model that can efficiently convert text into high-quality semantic vector representations. The BGE model can be used to transform natural language segments in multi-turn conversations and task chains of intelligent agents into high-dimensional semantic vectors.

[0061] 3) Calculate the similarity between each text vector using a similarity determination algorithm to determine the first similarity matrix.

[0062] In this embodiment, the similarity determination algorithm can be an algorithm for calculating similarity. For example, the similarity determination algorithm can be a cosine similarity algorithm, which is a method for measuring the degree of similarity between two vectors in a direction. It measures their similarity by calculating the cosine value of the angle between the two vectors. The value range of cosine similarity is within... Values ​​between 1 and 1 indicate that the two vectors are more similar; values ​​closer to 1 indicate greater similarity. A value of 1 indicates that two vectors are less similar, while a value of 0 indicates that two vectors are orthogonal (i.e., completely unrelated). A similarity determination algorithm can calculate the similarity between each text vector, and thus determine the first similarity matrix.

[0063] 4) Merge each initial historical interaction data according to the first similarity matrix to determine the historical interaction dataset.

[0064] It is known that after obtaining the first similarity matrix, the initial historical interaction data can be merged based on the first similarity matrix to determine the historical interaction dataset. For example, the BGE (BAAI General Embedding) model can be used to transform natural language segments in multi-turn conversations and task chains of intelligent agents into high-dimensional semantic vectors; the cosine similarity of adjacent segment vectors can be calculated; the similarity determination introduces multi-feature fusion (including semantic vector similarity, segment spacing, segment length ratio, etc.), and adopts an adaptive threshold mechanism based on data distribution (initial suggested threshold: 0.75). If the threshold is exceeded (0.85), paragraph merging is performed; semantic vectors are regenerated for the merged semantically coherent paragraphs, and large paragraph units with clear semantic boundaries are output, thus forming high-quality structured text (i.e., historical interaction dataset) for multi-turn conversations and task chains.

[0065] Figure 2 This embodiment provides a schematic diagram of a paragraph merging process, such as... Figure 2 As shown, the text to be analyzed (initial historical interaction data) can be vectorized, converting each paragraph into a vector form to facilitate similarity calculation. Here, each paragraph is converted into a text vector. By calculating the similarity between two consecutive paragraph vectors, it is determined whether they are semantically similar. Similarity calculation may use methods such as cosine similarity. If two consecutive paragraphs are semantically similar, they are merged into one paragraph, and a new paragraph is started from this; otherwise, a new paragraph is started directly. The above steps are repeated until all paragraphs have been processed.

[0066] As an optional implementation of this embodiment, the step of using a preset clustering algorithm to perform clustering analysis on each historical interaction data in the historical interaction dataset to determine the clustering results includes: 1) The historical interaction dataset is vectorized using the text vector generation model to determine the text vector corresponding to each historical interaction data.

[0067] Specifically, after obtaining the historical interaction dataset, a text vector generation model can be used to vectorize the historical interaction dataset, thereby determining the text vector corresponding to each historical interaction data.

[0068] 2) Use a preset clustering algorithm to perform clustering analysis on each text vector to determine the clustering results.

[0069] It is known that after obtaining the text vectors corresponding to each historical interaction data, a preset clustering algorithm can be used to perform clustering analysis on each text vector, thereby obtaining the clustering results.

[0070] As an optional implementation of this embodiment, the step of using a preset clustering algorithm to perform clustering analysis on each text vector to determine the clustering result includes: 1) Calculate the similarity between each text vector using a similarity determination algorithm to determine the second similarity matrix.

[0071] In this embodiment, the similarity determination algorithm can be a cosine similarity determination algorithm. The similarity determination algorithm can be used to calculate the similarity between each text vector, thereby determining a second similarity matrix. Each element in the second similarity matrix can represent the similarity between two text vectors.

[0072] 2) Cluster each text vector based on the second similarity matrix to determine an initial cluster set; the initial cluster set includes at least one initial cluster.

[0073] Specifically, after obtaining the second similarity matrix, each text vector can be clustered according to the similarity in the second similarity matrix, and the text vectors with high similarity can be merged into an initial cluster, thus obtaining the initial cluster set.

[0074] 3) Calculate the inter-class distance between each initial cluster using the inter-class distance determination algorithm to determine the inter-class distance matrix.

[0075] In this embodiment, the inter-class distance determination algorithm can be an algorithm used to calculate the distance between clusters, such as: SingleLinkage, CompleteLinkage, and AverageLinkage. SingleLinkage calculates the distance between the two closest points between two clusters, CompleteLinkage calculates the distance between the two farthest points between two clusters, and AverageLinkage determines the distance between two clusters by calculating the average distance between all point pairs in the two clusters.

[0076] Specifically, the inter-class distance determination algorithm can be used to calculate the inter-class matrix between each initial cluster, and then the inter-class distance matrix can be constructed.

[0077] 4) Determine the target initial cluster based on the inter-class distance matrix, and merge the target initial clusters to update the initial cluster set.

[0078] Following the above description, after obtaining the inter-class distance matrix, the two clusters with the closest inter-class distance can be used as the target initial clusters, the target initial clusters can be merged into one cluster, and the initial cluster set can be updated.

[0079] 5) Update the inter-class distance matrix based on the updated initial cluster set.

[0080] Specifically, after obtaining the updated initial cluster set, the inter-class distance determination algorithm can be used again to calculate the inter-class distance between each cluster in the updated initial cluster set, and then the inter-class distance matrix can be updated.

[0081] 6) If the number of clusters in the updated initial cluster set does not reach the target number of clusters, then return to the step of determining the target initial clusters based on the inter-class distance matrix until the number of clusters in the updated initial cluster set reaches the target number of clusters; the target number of clusters is determined based on preset evaluation rules and preset evaluation indicators.

[0082] In this embodiment, the target number of clusters can be determined based on preset evaluation rules and preset evaluation indicators. The preset evaluation rules can be the elbow rule, and the preset evaluation indicator can be the silhouette coefficient. The optimal number of clusters can be determined using the silhouette coefficient and the elbow method. For example, the elbow method can be used first to initially evaluate the optimal number of clusters, which calculates the sum of squared deviations (within) corresponding to different numbers of clusters (K). The ClusterSumofSquares (WSS) is plotted, and a curve of WSS versus K is generated, selecting the K value corresponding to the inflection point of the curve as a candidate. To further improve reliability, the SilhouetteScore and Calinski coefficients are introduced. Internal evaluation metrics such as the Harabasz index are used in conjunction with the elbow rule results to make comprehensive decisions, ensuring that the number of clusters in the clustering results matches the actual business scenario.

[0083] It should be explained that after obtaining the updated initial cluster set, it can be determined whether the number of clusters in the updated initial cluster set reaches the target number of clusters. If the number of clusters in the updated initial cluster set does not reach the target number of clusters, steps 4) and 5) can be returned to execute until the number of clusters in the updated initial cluster set reaches the target number of clusters.

[0084] 7) Determine the updated initial cluster set as the clustering result.

[0085] Specifically, if the number of clusters in the updated initial cluster set reaches the target number of clusters, the updated initial cluster set can be determined as the clustering result.

[0086] In this embodiment, the knowledge structure of the determined target intelligent agent can also be verified. The verification process may include: 1) Inputting target scenario information: receiving industry attributes and specific task information of the target application scenario; 2) Generating candidate structures: calling the knowledge structure recommendation model and outputting possible candidate knowledge structure types in the scenario; 3) Scoring and filtering: using the structure scoring model to comprehensively score all candidate structures and filtering out all candidate knowledge structures with scores higher than a set threshold (initial suggested value is 0.8); 4) Output and suggestions: outputting all candidate knowledge structures exceeding the threshold as the optimized "optimal" candidate set, and providing corresponding deployment and implementation suggestions for each candidate knowledge structure for the business to make a final selection based on specific needs. Simultaneously, the recommendation effect can be continuously monitored, and model updates will be triggered when the following situations occur: the actual usage rate of the recommended structure decreases beyond the threshold; new industry scenarios lead to an expansion of structure types; a major model version update leads to changes in structure adaptability.

[0087] The construction process of the structural scoring model includes: Input layer: receiving user input of industry attributes, scene context, task description, skill type, and skill knowledge structure information; Background layer: using the TextRCNN model, removing its final classification output layer, and using its feature extraction capabilities to obtain global information about the user's industry and scene from the global context and local pooling, providing background reference for subsequent scoring; Task layer: using all information as input, utilizing the ALBert model for deep encoding, and outputting a comprehensive task feature vector containing user industry, scene, task, and skill knowledge information; Fusion layer: first, projecting the feature vector output by the background layer to a space with the same feature dimension as the ALBert output through a fully connected layer, then concatenating the two along the feature dimension, and then flattening the concatenated joint features before feeding them into the subsequent fully connected network for processing; Output layer: outputting a continuous numerical value as a comprehensive score of the input knowledge structure; Model training: using the mean squared error loss function, with the mean squared error between the model's predicted score and the real human-annotated score as the optimization objective, and adjusting the model parameters through the gradient descent algorithm. Model evaluation and business validation: After obtaining the structural scoring model, the final evaluation effect of the model needs to be confirmed according to the actual scenario requirements to ensure its usability in practical applications.

[0088] The technical solution provided in this application constructs a structured knowledge template library by mining multi-turn conversations, task chains, and task planning information from intelligent agent operation data. It then utilizes recommendation and scoring models to recommend optimal knowledge structures for specific scenarios, thereby achieving precise optimization of the intelligent agent's thinking ability. Compared with existing technologies, the main improvements of this invention include: 1. Introducing a dynamic paragraph segmentation method based on semantic similarity: By using BGE model vectorization and adaptive thresholding mechanisms, the problem of fuzzy unstructured text parsing in existing technologies is solved, achieving automated semantic segmentation of skill texts. 2. Employing a hierarchical clustering knowledge structure mining algorithm: Combining the elbow rule and multi-index decision-making, a reusable thinking template library (preset knowledge structure library) is automatically generated, overcoming the limitation of existing technologies lacking industry-adaptive templates. 3. Designing a dual-model collaborative optimization framework: By forming a closed loop through a knowledge structure recommendation model and a scoring algorithm, the entire process of "recommendation-evaluation-optimization" is automated, improving optimization efficiency. 4. Proposing a multi-dimensional fusion scoring formula: Integrating manual scoring, usage frequency, and evaluation data, the scoring results of different industries are standardized, solving the problem of single evaluation dimensions in existing technologies.

[0089] The technical effects that the technical solution of this application can achieve include: 1) Breaking through the limitations of existing technologies that only focus on the overall performance of intelligent agents, this technology focuses on the evaluation and enhancement of specific intelligent thinking abilities such as multi-turn conversations and task chain planning, thus achieving precise improvement of intelligent agent skills.

[0090] 2) The introduction of annotation mechanisms and industry scenario task features enables the model to learn more professional and practical business knowledge, thereby improving the effectiveness and relevance of intelligent agent intelligent thinking enhancement.

[0091] 3) By building a knowledge reserve of intelligent agents, we have achieved in-depth mining and utilization of historical interaction log data, providing rich knowledge support for the continuous optimization of intelligent agents and improving the ability of intelligent agents to adapt to different industries and scenarios.

[0092] Example 2 Figure 3 This is a schematic diagram of the structure of a knowledge structure determination device for an intelligent agent provided in Embodiment 2 of this disclosure; as shown Figure 3 As shown, the device includes: a data acquisition module 210, a recommendation knowledge structure determination module 220, a scoring result determination module 230, and a knowledge structure determination module 240.

[0093] Among them, the data acquisition module 210 is used to acquire scene data and task data of the target intelligent agent in the current running task scenario; The recommended knowledge structure determination module 220 is used to input the scene data and the task data into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; The scoring result determination module 230 is used to determine the scoring result corresponding to the recommended knowledge structure using a preset scoring algorithm; The knowledge structure determination module 240 is used to determine, based on the scoring results, the target knowledge structure adopted by the target agent in the current running task scenario from the recommended knowledge structures. Embodiment 2 of this disclosure provides a knowledge structure determination device for an intelligent agent, which enables precise improvement of the intelligent agent's skills and enhances the intelligent agent's ability to adapt to different industries and scenarios.

[0094] Furthermore, the device also includes: an agent knowledge structure recommendation model determination module, the agent knowledge structure recommendation model determination module being used for: The historical training dataset is determined based on the historical operation logs of the intelligent agent. The historical training dataset includes at least one historical training data, which includes historical scene data and historical task data. For any historical training data, the knowledge structure label corresponding to the historical training data is determined based on the existing knowledge structure in the preset knowledge structure base. The first preset algorithm is used to extract scene features from the historical scene data; The task features of the historical task data are extracted using a second preset algorithm; The scene features and the task features are concatenated to determine the joint features corresponding to the historical training data; Based on the joint features and knowledge structure labels corresponding to each historical training data, a preset loss function is used as the training objective, and a preset parameter tuning algorithm is used to adjust the model parameters of the initial agent knowledge structure recommendation model in order to determine the agent knowledge structure recommendation model.

[0095] Furthermore, the device also includes: a preset knowledge structure base determination module, the preset knowledge structure base determination module comprising: The historical interaction dataset determination submodule is used to determine the historical interaction dataset based on the agent's historical operation logs; The clustering analysis submodule is used to perform clustering analysis on each historical interaction data in the historical interaction dataset using a preset clustering algorithm to determine the clustering results; A determination submodule is used to determine the preset knowledge structure library based on the clustering results, wherein the preset knowledge structure library includes at least one knowledge structure.

[0096] Furthermore, the historical interaction dataset determination submodule also includes: Determine the initial historical interaction data based on the agent's historical operation logs; A text vector generation model is used to vectorize each initial historical interaction data to determine the text vector corresponding to each initial historical interaction data. A similarity determination algorithm is used to calculate the similarity between each text vector in order to determine the first similarity matrix; The initial historical interaction data are merged based on the first similarity matrix to determine the historical interaction dataset.

[0097] Furthermore, the cluster analysis submodule also includes: The vectorization processing unit is used to perform vectorization processing on the historical interaction dataset using the text vector generation model to determine the text vector corresponding to each historical interaction data. The clustering result determination unit is used to perform clustering analysis on each text vector using a preset clustering algorithm to determine the clustering result.

[0098] Furthermore, the clustering results determining the unit are also used for: A similarity determination algorithm is used to calculate the similarity between each text vector in order to determine the second similarity matrix; The text vectors are clustered based on the second similarity matrix to determine an initial cluster set; the initial cluster set includes at least one initial cluster. The inter-class distance matrix is ​​determined by calculating the inter-class distance between each initial cluster using an inter-class distance determination algorithm. The target initial cluster is determined based on the inter-class distance matrix, and the target initial clusters are merged to update the initial cluster set; Update the inter-class distance matrix based on the updated initial cluster set; If the number of clusters in the updated initial cluster set does not reach the target number of clusters, then return to the step of determining the target initial clusters based on the inter-class distance matrix, until the number of clusters in the updated initial cluster set reaches the target number of clusters; the target number of clusters is determined based on preset evaluation rules and preset evaluation indicators; The updated initial cluster set is determined as the clustering result.

[0099] The knowledge structure determination device for intelligent agents provided in this disclosure can execute the knowledge structure determination method for intelligent agents provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of executing the method.

[0100] Example 3 Figure 4A schematic diagram of the structure of an electronic device 10 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the embodiments of the present disclosure described and / or claimed herein.

[0101] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0102] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0103] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microprocessor, etc. Processor 11 performs the various methods and processes described above, such as methods for determining the knowledge structure of an intelligent agent.

[0104] In some embodiments, the agent knowledge structure determination method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the agent knowledge structure determination method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the agent knowledge structure determination method by any other suitable means (e.g., by means of firmware).

[0105] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0106] Computer programs for implementing the methods of embodiments of this disclosure may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0107] In the context of embodiments of this disclosure, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0108] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0109] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0110] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0111] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the embodiments of this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of the embodiments of this disclosure can be achieved, and this document does not impose any limitations.

[0112] The specific embodiments described above do not constitute a limitation on the scope of protection of the embodiments disclosed herein. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the embodiments disclosed herein should be included within the scope of protection of the embodiments disclosed herein.

[0113] This disclosure also provides a computer program product, including a computer program and / or instructions, which, when executed by a processor, implements the knowledge structure determination method for an intelligent agent as provided in any embodiment of this application.

[0114] In implementing a computer program product, computer program code for performing the operations of the embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0115] Note that the above are merely preferred embodiments and the technical principles applied in this disclosure. Those skilled in the art will understand that this disclosure is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the protection scope of this disclosure. Therefore, although the embodiments of this disclosure have been described in detail above, this disclosure is not limited to the above embodiments. More other equivalent embodiments may be included without departing from the concept of this disclosure, and the scope of this disclosure is determined by the scope of the appended claims.

Claims

1. A method for determining the knowledge structure of an intelligent agent, characterized in that, include: Acquire scene data and task data of the target intelligent agent in the current running task scenario; The scene data and the task data are input into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library. The scoring result corresponding to the recommended knowledge structure is determined using a preset scoring algorithm; Based on the scoring results, the target knowledge structure adopted by the target agent in the current running task scenario is determined from the recommended knowledge structure.

2. The method according to claim 1, characterized in that, The process of determining the agent knowledge structure recommendation model includes: The historical training dataset is determined based on the historical operation logs of the intelligent agent. The historical training dataset includes at least one historical training data, which includes historical scene data and historical task data. For any historical training data, the knowledge structure label corresponding to the historical training data is determined based on the existing knowledge structure in the preset knowledge structure base. The first preset algorithm is used to extract scene features from the historical scene data; The task features of the historical task data are extracted using a second preset algorithm; The scene features and the task features are concatenated to determine the joint features corresponding to the historical training data; Based on the joint features and knowledge structure labels corresponding to each historical training data, a preset loss function is used as the training objective, and a preset parameter tuning algorithm is used to adjust the model parameters of the initial agent knowledge structure recommendation model in order to determine the agent knowledge structure recommendation model.

3. The method according to claim 1, characterized in that, The process of determining the preset knowledge structure base includes: Historical interaction datasets are determined based on the agent's historical operation logs; A preset clustering algorithm is used to perform clustering analysis on each historical interaction data in the historical interaction dataset to determine the clustering results; The preset knowledge structure base is determined based on the clustering results, and the preset knowledge structure base includes at least one knowledge structure.

4. The method according to claim 3, characterized in that, The historical interaction dataset determined based on the agent's historical operation logs includes: Determine the initial historical interaction data based on the agent's historical operation logs; A text vector generation model is used to vectorize each initial historical interaction data to determine the text vector corresponding to each initial historical interaction data. A similarity determination algorithm is used to calculate the similarity between each text vector in order to determine the first similarity matrix; The initial historical interaction data are merged based on the first similarity matrix to determine the historical interaction dataset.

5. The method according to claim 4, characterized in that, The step of using a preset clustering algorithm to perform clustering analysis on each historical interaction data in the historical interaction dataset to determine the clustering results includes: The historical interaction dataset is vectorized using the text vector generation model to determine the text vector corresponding to each historical interaction data. The clustering results are determined by performing clustering analysis on each text vector using a preset clustering algorithm.

6. The method according to claim 5, characterized in that, The step of using a preset clustering algorithm to perform clustering analysis on each text vector to determine the clustering result includes: A similarity determination algorithm is used to calculate the similarity between each text vector in order to determine the second similarity matrix; Each text vector is clustered based on a second similarity matrix to determine an initial cluster set; the initial cluster set includes at least one initial cluster. The inter-class distance matrix is ​​determined by calculating the inter-class distance between each initial cluster using an inter-class distance determination algorithm. The target initial cluster is determined based on the inter-class distance matrix, and the target initial clusters are merged to update the initial cluster set; Update the inter-class distance matrix based on the updated initial cluster set; If the number of clusters in the updated initial cluster set does not reach the target number of clusters, then return to the step of determining the target initial clusters based on the inter-class distance matrix, until the number of clusters in the updated initial cluster set reaches the target number of clusters; the target number of clusters is determined based on preset evaluation rules and preset evaluation indicators; The updated initial cluster set is determined as the clustering result.

7. A device for determining the knowledge structure of an intelligent agent, characterized in that, include: The data acquisition module is used to acquire scene data and task data of the target intelligent agent in the current running task scenario; The recommended knowledge structure determination module is used to input the scene data and the task data into the agent knowledge structure recommendation model to obtain at least one recommended knowledge structure corresponding to the target agent; the recommended knowledge structure is a knowledge structure defined in a preset knowledge structure library; The scoring result determination module is used to determine the scoring result corresponding to the recommended knowledge structure using a preset scoring algorithm; The knowledge structure determination module is used to determine the target knowledge structure adopted by the target agent in the current running task scenario from the recommended knowledge structure based on the scoring results.

8. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the knowledge structure determination method for an intelligent agent as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the knowledge structure determination method for an agent as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for determining the knowledge structure of an intelligent agent as described in any one of claims 1-6.

Images