User behavior data augmentation method based on rflp-driven large language model

Abstract

The application relates to the technical field of data enhancement and system engineering, in particular to a user behavior data enhancement method based on an RFLP-driven large language model. The method comprises the following steps: determining a core driving problem and an expected application scenario of to-be-generated user behavior data, determining a task target, a data structure requirement, a data characteristic requirement, a constraint condition and a non-functional requirement of a large language model data generation system based on the RFLP driving, and coding the requirement specifications into a plurality of functional modules, forming a data generation function set according to the modules, constructing a logic scheme based on "strategy-component-combination" for the system, and converting the logic scheme into an executable scheme, finally controlling the large language model to execute the executable scheme to generate the to-be-generated user behavior data. Therefore, the problem that a prompt word is difficult to guarantee the quality and consistency of generated data is solved, and scientific methodological support is provided for constructing a high-quality, controllable and expandable synthetic user data generation system.

Description

Technical Field

[0001] This application relates to the fields of data augmentation and systems engineering, and in particular to a user behavior data augmentation method based on a large language model driven by RFLP (Requirements-Functional-Logical-Physical). Background Technology

[0002] User data, as a fundamental element of modern intelligent systems, plays an irreplaceable role in various application scenarios such as user behavior prediction, personalized recommendations, intelligent customer service, and precision marketing. With the rapid development of artificial intelligence and machine learning technologies, the value of high-quality user data is becoming increasingly prominent. The performance of advanced technologies such as deep learning models, recommendation algorithms, and natural language processing systems largely depends on the quality, scale, and diversity of training data. User data not only provides the basic materials for model training but is also a crucial carrier for verifying algorithm effectiveness and optimizing system performance. Its richness and accuracy directly affect the system's level of intelligence and its ability to create business value.

[0003] While user data is of significant value, its practical application faces numerous challenges and limitations. Data scarcity is particularly pronounced in emerging business areas, cold-start scenarios, or specific user groups. In practice, there is often a lack of sufficient historical data to support model training and system optimization, hindering the performance of machine learning models. With increasing privacy concerns, the collection, storage, and use of real user data face more constraints, increasing the complexity of data acquisition and application. Furthermore, data quality and integrity deficiencies are another major challenge. Real user data often suffers from missing values, noise, bias, and other quality issues, and it struggles to cover all possible user behavior patterns and edge cases, affecting the robustness of systems built upon this data. High-quality labeled data is fundamental to supervised learning, but manual labeling is time-consuming and costly, especially in complex tasks requiring domain experts. Traditional data augmentation methods rely primarily on simple data transformations, interpolation, or rule-based generation methods, which struggle to capture complex user behavior patterns and semantic relationships. The generated data often lacks authenticity and diversity, failing to meet the demands of modern artificial intelligence systems for high-quality training data.

[0004] Among related technologies, the emergence of Large Language Models (LLMs) provides a new technical approach to solving the aforementioned problems. LLMs possess powerful text understanding and generation capabilities, enabling them to understand complex instructions, simulate diverse user behavior patterns, and generate high-quality synthetic data, thus providing a technological foundation for generating realistic user data. User data generation can be guided by designing prompts.

[0005] However, effectively designing prompts to guide LLM in generating user data that meets specific needs remains a challenge. Existing prompt designs often lack a systematic approach, rely on experience and trial and error, and struggle to guarantee the quality and consistency of the generated data. Summary of the Invention

[0006] This application provides a user behavior data augmentation method based on RFLP-driven large language models to address the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and making it difficult to guarantee the quality and consistency of generated data. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0007] To achieve the above objectives, the first aspect of this application proposes a user behavior data augmentation method based on an RFLP-driven large language model, comprising the following steps: Identify the core driving issues and expected application scenarios for the user behavior data to be generated; Based on the core driving problem and the expected application scenario, the task objectives, data structure requirements, data characteristic requirements, constraints and non-functional requirements of the RFLP-driven large language model data generation system are determined. The task objective, data structure requirements, data characteristic requirements, constraints, and non-functional requirements are respectively encoded into multiple functional modules, and a functional set of data generation tasks is formed according to the functions of the multiple functional modules. Based on the aforementioned function set, a logical scheme based on "strategy-component-combination" is constructed for the large language model data generation system; Transform the logical scheme into an executable scheme; The large language model is controlled to execute the executable scheme and generate the user behavior data to be generated.

[0008] According to one embodiment of this application, after generating the user behavior data to be generated, the method further includes: The quality of the user behavior data to be generated is verified to obtain the quality verification result. Determine whether the quality verification result meets the preset verification conditions; If the quality verification result does not meet the preset verification conditions, the steps of determining the task objectives, data structure requirements, data characteristic requirements, and constraints of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario are repeated until the quality verification result of the new user behavior data meets the preset verification conditions.

[0009] According to one embodiment of this application, determining the task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario includes: Based on the core driving problem and the expected application scenario, at least one goal to be achieved is identified, and the task objective is determined according to the business requirements corresponding to the at least one goal to be achieved. Based on the core driving problem and the expected application scenario, the necessary data types and attributes of the user behavior data to be generated are determined, and the data structure requirements are determined according to the necessary data types and attributes. Based on the core driving problem and the expected application scenario, the qualitative and quantitative characteristics of the user behavior data to be generated are determined, and the data characteristic requirements are determined according to the qualitative and quantitative characteristics. Based on the core driving problem and the expected application scenario, the operational boundary of the large language model generation process is determined, and the constraint conditions are determined according to the operational boundary.

[0010] According to one embodiment of this application, the non-functional requirements include at least one of the following: accuracy requirements, authenticity requirements, task relevance requirements, simplicity requirements, coherence and fluency requirements, safety and hazard avoidance requirements, bias mitigation requirements, and robustness requirements.

[0011] According to one embodiment of this application, the plurality of functional modules includes multiple types of input analysis functional modules, core generation functional modules, control functional modules, and output functional modules.

[0012] The user behavior data augmentation method based on RFLP-driven large language model proposed in this application identifies the core driving problem and expected application scenario of the user behavior data to be generated. Based on this, the requirements of the RFLP-driven large language model data generation system are determined. These requirements are standardized and encoded into multiple functional modules, forming a data generation function set. Based on this, a logical scheme based on "strategy-component-combination" is constructed for the system and transformed into an executable scheme. Finally, the large language model data is controlled to execute the executable scheme to generate the user behavior data to be generated. This method solves the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and difficulty in guaranteeing the quality and consistency of generated data. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0013] To achieve the above objectives, a second aspect of this application proposes a user behavior data augmentation device based on an RFLP-driven large language model, comprising: The determination module is used to identify the core driving issues and expected application scenarios for the user behavior data to be generated. The requirements specification module is used to determine the task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario. The function parsing module is used to encode the task objective, the data structure requirements, the data characteristic requirements, the constraints, and the non-functional requirements into multiple functional modules, and to form a function set for the data generation task based on the functions of the multiple functional modules. The logic combination module is used to construct a logic scheme based on "strategy-component-combination" for the large language model data generation system based on the function set. The physical processing module is used to convert the logical scheme into an executable scheme; The control module is used to control the large language model to execute the executable scheme and generate the user behavior data to be generated.

[0014] According to one embodiment of this application, the control module is further configured to: The quality of the user behavior data to be generated is verified to obtain the quality verification result. Determine whether the quality verification result meets the preset verification conditions; If the quality verification result does not meet the preset verification conditions, the steps of determining the task objectives, data structure requirements, data characteristic requirements, and constraints of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario are repeated until the quality verification result of the new user behavior data meets the preset verification conditions.

[0015] According to one embodiment of this application, the requirement specification module is specifically used for: Based on the core driving problem and the expected application scenario, at least one goal to be achieved is identified, and the task objective is determined according to the business requirements corresponding to the at least one goal to be achieved. Based on the core driving problem and the expected application scenario, the necessary data types and attributes of the user behavior data to be generated are determined, and the data structure requirements are determined according to the necessary data types and attributes. Based on the core driving problem and the expected application scenario, the qualitative and quantitative characteristics of the user behavior data to be generated are determined, and the data characteristic requirements are determined according to the qualitative and quantitative characteristics. Based on the core driving problem and the expected application scenario, the operational boundary of the large language model generation process is determined, and the constraint conditions are determined according to the operational boundary.

[0016] According to one embodiment of this application, the non-functional requirements include at least one of the following: accuracy requirements, authenticity requirements, task relevance requirements, simplicity requirements, coherence and fluency requirements, safety and hazard avoidance requirements, bias mitigation requirements, and robustness requirements.

[0017] According to one embodiment of this application, the plurality of functional modules includes multiple types of input analysis functional modules, core generation functional modules, control functional modules, and output functional modules.

[0018] According to the user behavior data enhancement device based on RFLP-driven large language model proposed in this application, the core driving problem and expected application scenario of the user behavior data to be generated are determined. Based on this, the requirements of the RFLP-driven large language model data generation system are determined. These requirements are standardized and encoded to obtain multiple functional modules. A data generation function set is formed according to the modules. Based on this, a logical scheme based on "strategy-component-combination" is constructed for the system and transformed into an executable scheme. Finally, the large language model data is controlled to execute the executable scheme to generate the user behavior data to be generated. This solves the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and difficulty in ensuring the quality and consistency of generated data in the prior art. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0019] To achieve the above objectives, a third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the user behavior data augmentation method based on an RFLP-driven large language model as described in the above embodiments.

[0020] To achieve the above objectives, a fourth aspect of this application provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the user behavior data augmentation method based on an RFLP-driven large language model as described in the above embodiments.

[0021] To achieve the above objectives, a fifth aspect of this application provides a computer program product, which, when executed by a processor, implements the user behavior data augmentation method based on an RFLP-driven large language model as described in the above embodiments.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a user behavior data augmentation method based on an RFLP-driven large language model, according to an embodiment of this application. Figure 2 A flowchart generated according to a logic scheme provided in one embodiment of this application; Figure 3 A flowchart of a user behavior data augmentation method based on an RFLP-driven large language model according to a specific embodiment of this application; Figure 4 This is a schematic diagram of a user data generation system for a large language model based on RFLP according to an embodiment of this application; Figure 5 This is a block diagram of a user behavior data augmentation device based on an RFLP-driven large language model provided in an embodiment of this application. Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0025] The following describes, with reference to the accompanying drawings, a user behavior data augmentation method based on an RFLP-driven large language model proposed according to an embodiment of this application. First, the user behavior data augmentation method based on an RFLP-driven large language model proposed according to an embodiment of this application will be described with reference to the accompanying drawings.

[0026] Figure 1 This is a flowchart of a user behavior data augmentation method based on an RFLP-driven large language model according to an embodiment of this application.

[0027] like Figure 1 As shown, this user behavior data augmentation method based on RFLP-driven large language model includes the following steps: In step S101, the core driving issues and expected application scenarios for generating user behavior data are determined.

[0028] User behavior data refers to the set of recordable information about various behaviors and related states generated by users during their interactions with products, systems, platforms, or services.

[0029] Specifically, this application embodiment first requires clarifying the core driving issues and expected application scenarios for generating synthetic user data. The core driving issues refer to the actual obstacles or system bottlenecks existing at the current data application level, including solving data scarcity problems, meeting privacy protection requirements, enhancing model training data, insufficient system robustness, and weak coverage of edge scenarios. The expected application scenario analysis focuses on identifying the specific systems or models that will use the generated data, such as recommendation algorithms, user classification models, and behavior prediction systems. Through in-depth analysis of the specific requirements of downstream systems regarding data format, field types, data volume, and update frequency, it ensures that the generated data is integrated into the existing workflow. Simultaneously, it determines the specific data quality requirements of downstream systems, such as accuracy thresholds, processing latency limits, and concurrent access capabilities. It should be noted that this application embodiment also includes the analysis of existing real data at this stage, including feature analysis of data distribution characteristics, user behavior patterns, time series characteristics, and missing value distribution. It identifies quality problems in existing data, such as data bias, sample imbalance, labeling errors, and timeliness issues. By comparing business needs with existing data, specific data gaps are identified, such as insufficient data for specific user groups, missing behavioral patterns, and inadequate coverage of edge scenarios.

[0030] In step S102, based on the core driving problem and the expected application scenario, the task objectives, data structure requirements, data characteristic requirements, constraints and non-functional requirements of the large language model data generation system driven by RFLP are determined.

[0031] Optionally, in some embodiments, based on the core driving problem and expected application scenario, the task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements of the RFLP-driven large language model data generation system are determined, including: determining at least one goal to be achieved based on the core driving problem and expected application scenario, and determining the task objectives according to the business requirements corresponding to the at least one goal to be achieved; determining the necessary data types and attributes of the user behavior data to be generated based on the core driving problem and expected application scenario, and determining the data structure requirements according to the necessary data types and attributes; determining the qualitative and quantitative characteristics of the user behavior data to be generated based on the core driving problem and expected application scenario, and determining the data characteristic requirements according to the qualitative and quantitative characteristics; and determining the operational boundaries of the large language model generation process based on the core driving problem and expected application scenario, and determining the constraints according to the operational boundaries.

[0032] Among these, business requirements refer to specific, quantifiable, and verifiable requirements proposed from actual business scenarios to achieve specific goals. Necessary data types and attributes refer to the data categories and their specific characteristics that must be included to achieve task objectives and adapt to the expected application scenario. Qualitative characteristics of user behavior data refer to features that cannot be directly quantified numerically but are described by categories, attributes, logical relationships, or properties to characterize the essence of user behavior. Quantitative characteristics of user behavior data refer to user behavior attributes that can be described by specific numerical values, frequencies, dimensions, or quantitative indicators. The operational boundaries of the large language model generation process refer to the insurmountable scope and limitations defined for model generation behavior based on the core driving problem and the expected application scenario.

[0033] Specifically, defining the task objectives involves identifying one or more technical goals to be achieved based on the core driving problem and the expected application scenario. Each goal should directly correspond to a specific business requirement. Examples include simulating user intent and constructing search behavior sequences, thereby providing clear directional objectives for system design. Subsequently, each technical goal needs to be transformed into measurable indicators, such as specific objectives like "generating 1,000 e-commerce user profiles covering different age groups," "creating 500 user interaction sequences containing cold start scenarios," and "generating 100 test cases containing abnormal behavior patterns."

[0034] Determining data structure requirements refers to clarifying the necessary data types and attributes that must be included in the generated user behavior data. Based on the data types and attributes, the content requirements, data structures, and field categories generated by the large language model are defined (including but not limited to user identifiers, item identifiers, interaction types, timestamps, comment text, or query text, etc.). In some embodiments, to support modeling and simulating user behavior sequences, the basic unit of user behavior data is defined as a quaternion sequence. ; in, For user-interactive item sequences, For the corresponding category sequence, This refers to a sequence of interactive actions (such as clicking, purchasing, etc.). The intention sequence is obtained based on the combination of category and action space, satisfying This structure serves as an output specification for user behavior generation, guiding the construction of model suggestions and the design of field parsing rules.

[0035] Determining data characteristic requirements refers to defining the qualitative and quantitative characteristics that the synthetic user data must possess. These characteristics include, but are not limited to: the realism of simulating real user behavior, the diversity (coverage and novelty) of the generated scenarios or user profiles, the controllability of the data volume, and the specific behavioral patterns required for the scenario (such as those observed in real data), statistical distribution characteristics, and internal semantic and behavioral aspects.

[0036] The constraints define the operational boundaries of the large language model generation process (including but not limited to requirements such as compliance with ethical norms and prohibition of including personally identifiable information), and the evaluation indicators provide a quantitative basis for determining the degree of conformity between the generated data and the defined requirements.

[0037] Furthermore, in some embodiments, non-functional requirements include at least one of the following: accuracy requirements, authenticity requirements, task relevance requirements, simplicity requirements, coherence and fluency requirements, safety and hazard avoidance requirements, bias mitigation requirements, and robustness requirements.

[0038] Specifically, the language model user data generation system also needs to meet a series of non-functional requirements during the design of prompt words and model invocation: accuracy / authenticity requirements mean that the output generated by the system should be correct and authentic, especially for informational prompts; task relevance requirements mean that the output must be relevant to the input query and context; conciseness requirements mean that the output should meet the required length limit, without omitting key details and ensuring system performance, and without unnecessary redundancy; coherence and fluency requirements mean that the language used should be natural, grammatically correct, and easy to understand; security and harm avoidance requirements mean that the prompt words should guide the invoked large model to avoid generating harmful, biased, or inappropriate content; bias mitigation requirements mean that the output should be fair and impartial, avoiding the perpetuation of social biases present in the training data; robustness requirements mean that the prompts should still elicit consistent and correct responses when the input wording changes slightly or unexpected edge cases are encountered. It should be noted that non-functional requirements are flexible and open. The above requirements may only involve one or a few in actual scenarios, or they may cover all of them; at the same time, non-functional requirements are not limited to the types listed above, and other related requirements can be incorporated according to specific application scenarios and core objectives.

[0039] In step S103, the task objective, data structure requirements, data characteristic requirements, constraints and non-functional requirements are encoded into multiple functional modules, and a functional set of data generation task is formed according to the functions of the multiple functional modules.

[0040] In this context, standardized coding refers to classifying, defining, identifying, and structuring task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements according to unified and standardized rules and formats. A functional module refers to a structured unit with independent functions, clearly defined boundaries, and specific roles, formed after standardized coding.

[0041] Specifically, based on the user data generation goals and sub-requirements defined in step S102, the basic tasks that the LLM system needs to complete (such as simulating behavior, building preferences, controlling time series, etc.) are identified, and the abstract requirements are broken down into a set of clear functional units. For example, "simulating user search behavior" is broken down into sub-functions such as "generating query terms," ​​"generating click behavior," and "generating comment content." The requirements specification of step S102 is encoded into multiple functional modules. Based on the specific functions of each module, they are organically integrated according to the logical relationship and process sequence of the data generation tasks, thereby constructing a complete and collaborative functional set.

[0042] Optionally, in some embodiments, the multiple functional modules include various types of input analysis functional modules, core generation functional modules, control functional modules, and output functional modules.

[0043] Specifically, the multiple functional modules can be divided into different types: input analysis functional modules, core generation functional modules, control functional modules, and output functional modules, etc.

[0044] Input analysis modules include, but are not limited to, reading and parsing external data, and supporting the preloading of some context, historical behavior or user settings as embedded context in the generated prompts.

[0045] The core generation function modules include, but are not limited to: (1) generating dynamic interactive data and generating user behavior sequences that change over time.

[0046] in For users The sequence of behaviors; For users in time Behavioral events may include operation type, object ID, context information, etc. The timestamp of the event; The length of the behavior sequence.

[0047] (2) Generate static user data and construct user profile-related feature data, including but not limited to demographic attributes and static preference categories. A common representation is to represent the user's static features as a vector:

[0048] in The static feature vector of the user; Let be the value of the i-th feature dimension, such as age, gender, interest category, etc. This represents the total number of dimensions of the features.

[0049] (3) Simulate data that matches user preferences. Based on user characteristics and context, generate behavioral data that matches individual preferences. Ensure that user behavior has semantic consistency in local sequences.

[0050] (4) Simulate user behavior patterns: Construct behavior flow that conforms to real usage paths, support the analysis of behavior transfer and provide training data.

[0051] Control-related functional modules include, but are not limited to: (1) introducing abnormal or marginal behaviors, inserting a certain percentage into the normal behavior path. Abnormal behavior samples are used to train robust models or evaluate anomaly detection capabilities. Abnormal behaviors can originate from specific rules (such as conflict behaviors) or be generated by random perturbations.

[0052] (2) Model real time attributes and add real or simulated timestamp information to the generated interactive sequence. This function enhances the time interpretability of the data and supports time series modeling tasks.

[0053] (3) Check the structural rationality and format legality of the generated behavior: After generating the behavior sequence, perform structural constraint verification on it, including but not limited to: whether it meets the predefined behavior template, field type consistency (e.g., the time field is in timestamp format), and whether the user's intention is maintained between multiple rounds of behavior. Figure 1 Consistency and whether it meets the requirements of the defined data distribution and expectations at each stage.

[0054] Output-related functional modules include, but are not limited to: generating structured output, supporting the generation of formatted text output that meets specified structures (such as JSON (JavaScript Object Notation) / CSV (Comma-Separated Values)), including field mapping, hierarchical nesting, type constraints, etc.

[0055] In step S104, based on the function set, a logical scheme based on "strategy-component-combination" is constructed for the large language model data generation system.

[0056] The logical solution refers to an abstract plan that systematically organizes and designs the operational logic, core elements, and interrelationships of a large language model data generation system based on a set of functions and following a "strategy-component-composition" framework. The executable solution refers to the plan that transforms the logical solution into concrete, implementable operational steps, technical specifications, and detailed implementation rules.

[0057] Specifically, a prompting strategy system is constructed for the large language model data generation system based on the function set defined in step S103. Considering the different performance requirements of various functional tasks regarding input structure, generation complexity, and control precision, the system needs to support multiple types of prompting strategies to ensure the logical structure has generalizability and robustness. Applicable strategies and their invocation logic are defined according to different data generation task objectives. Common strategies include, but are not limited to: (1) Zero-sample cue strategy: Applicable to general tasks where LLMs already have sufficient prior knowledge, directly constructing cue words without the need for examples:

[0058] in, Output generated for the language model; For large language models to respond to prompt words Zero-shot prompt; The task description (Instruction) specifies the target task, such as "Please generate a product review"; This refers to the context, such as input data snippets, questions, background materials, etc.

[0059] (2) Few-shot prompting strategy: Applicable to tasks that require specific format or pattern guidance, demonstrating the expected output generation results through examples:

[0060] in, Few-shot prompts; This is the set of examples, each containing input-output pairs to demonstrate the expected pattern; the remaining symbols have the same meaning as above.

[0061] (3) Chain-of-Thought (CoT): Suitable for complex tasks that require logical reasoning, it guides the model through multiple rounds of deduction to obtain deeper reasoning.

[0062] in, The result generated in the current step, the first Step output; The initial prompt word contains the task and context description; Construct strings for concatenation or progressive hints (such as "Let's reason step by step"); For the first The intermediate output of the step is used to guide the next step of reasoning.

[0063] Strategy selection not only includes the structural definition and calling interface of the three types of strategies, but also needs to be based on the context requirements of the functional task, so that the system can match the most suitable strategy template according to its input characteristics and output target.

[0064] The logical components are divided into three categories based on their responsibilities: core components, strategy-specific components, and function-related components. Core components are common to all strategies and mainly include, but are not limited to, task instruction components, context injection components, and output format control components, used to build the basic structure of prompts and information filling logic. Strategy-specific components serve the invocation needs of specific strategy patterns, such as example injection components required for few-shot tasks and chain triggering and follow-up questioning components required for thought chain reasoning, possessing a certain degree of strategy coupling and activation conditions. Function-related components are geared towards the function set defined in phase S103, providing task-oriented supplementary control during execution data generation, such as timestamp insertion components supporting dynamic behavior sequence construction, or user state maintenance components supporting behavior consistency modeling. The boundaries between components should be clear, but the combination method should be highly decoupled and support configuration based on tasks.

[0065] Construct the composition logic and configuration scheme between "strategy-component-composition". Specifically, this task requires defining the required component structure links for each type of strategy, clarifying the component calling order, splicing method, context passing mechanism, and input / output constraints. The composition logic should support horizontal strategy-component matching, meaning different strategies can choose their own suitable component subsets, and should also support vertical component-level splicing and nesting, enabling complex functions to be implemented through layer-by-layer composition. To ensure the execution stability and structural consistency of prompt words, the system needs to pre-set composition templates (such as zero-sample composition, few-sample composition, thought chain composition, etc.). Each template encapsulates the component sequence required by the corresponding strategy and predefines its parameter structure and slot placeholders. At the same time, the composition scheme needs to consider function-driven multi-task scenarios, integrating multiple component logics in a single prompt word, such as static profile generation + behavior sequence generation + dynamic time control, and implementing function scheduling through instruction annotation, paragraph separation, or multi-round prompt word invocation. Finally, the composition logic will output a set of configurable learning chains.

[0066] A unified data exchange protocol needs to be defined within the logic layer, and the triggering conditions and invocation rules for each combined logic structure need to be established. The data exchange protocol is used to define the information structure passed between components, including but not limited to field naming, nesting level, data type constraints, and format consistency, to ensure that the context between modules can be seamlessly connected during the prompt construction process, especially in complex paths where strategy templates require multiple rounds of information injection, component nesting, or condition activation. To this end, the system needs to establish several core data flow interfaces: the data structure passed from components to the large model includes but is not limited to "task instructions, context information, example fragments, and output constraints", while the results generated by the large model system will be further converted into structured or formatted data for downstream modules to parse and verify.

[0067] Furthermore, it is necessary to establish within the logic layer the judgment conditions and integration paths for whether external mechanisms (including but not limited to RAG (Retrieval-Augmented Generation) retrieval, knowledge invocation, and tool execution) are triggered. Given that the current system is task-oriented, whether to invoke external mechanisms is not determined by the task type, but rather by the logic stage's judgment of input conditions, context configuration, or strategy requirements. For example, if the system identifies a clear open-ended query intent in the prompt words, or if the current component structure annotation requires referencing knowledge-based content, the RAG mechanism can be triggered to incorporate the retrieval module into the assembly.

[0068] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the logical scheme provided in one embodiment of this application. The system logic for generating large language model user data is constructed based on a "strategy-component-combination" approach, according to a set of functionalities. A learning strategy is selected, data generation logic components are designed, and the combination of components is configured to form a logical scheme for generating large language model user data that supports diverse tasks.

[0069] S105 converts the logical scheme into an executable scheme.

[0070] Furthermore, the logical solution needs to be transformed into an executable solution, which mainly includes the following steps: (1) Prompt word construction and instantiation: The system constructs prompt word text that meets the task requirements based on the "strategy-component-composition" structure output from the logic stage. The prompt words should have clear structural blocks, including task instructions, context information, example fragments (if any), output constraint formats, etc., and support dynamic filling using placeholders. The construction of prompt words should conform to the specification constraints of the model used on input length, format, and structure.

[0071] (2) Model Configuration and Operation Control: The system needs to set various generation parameters required for the operation of the large language model. These parameters include, but are not limited to, model selection, generation temperature, maximum output length, and frequency penalty, which directly affect the style, coverage, and controllability of the generated content. At the same time, it is also necessary to select the specific model version to be used at this stage and set whether to call external knowledge mechanisms, such as retrieval-enhanced generation and tool plugins.

[0072] (3) API (Application Programming Interface) Integration and Input Encapsulation: This involves designing the API integration structure for encapsulating model input at the interface layer, based on the language model's call execution mechanism. Building upon the semantic template of the prompt words, the system needs to clearly define the organization of various input fields in the interface request, including prompt content, model parameters, external mechanism identifiers, and task metadata. In conjunction with the deployed model platform (such as a local engine or cloud API), a standardized calling method should be designed, clearly defining field naming, data types, and transmission formats, and establishing an input validation mechanism to ensure that the model can be called uniformly and stably.

[0073] (4) Design an output structure scheme for data reception, processing, and integration, focusing on the use of the large language model's generated results. The system needs to clearly define the target data format of the output results, set field naming conventions, hierarchical organization structure, value type specifications, and mapping relationships to ensure that the generated content has a clear and consistent structure. Based on the characteristics of the current task, design field conversion rules and position alignment methods for model output to target formats (such as JSON, structured tables, database records, etc.), and define necessary format validation logic and fault tolerance strategies. The design of the output structure should combine the format control requirements in the prompt words and connect to the system's external output interface to form a result organization scheme with task adaptability and integrability. The output of this stage (prompt words, parameter configurations, API encapsulation, etc.) serves as the input to the decoder and is decoded by the LLM core module.

[0074] In step S106, the large language model is controlled to execute an executable scheme to generate user behavior data to be generated.

[0075] Specifically, this application deploys the executable learning scheme generated in the physical phase to the target LLM environment. It configures the LLM instance according to the specifications of the physical phase, including model version selection, parameter settings, and other model instance configuration tasks. It tests the API connection with the LLM service to ensure the stability and security of data transmission, completing API integration verification. It verifies the integrity of the runtime environment, including necessary library, tool, and permission configuration environment dependency checks.

[0076] Based on system resources and API limitations, a reasonable batch processing strategy is designed to balance generation efficiency and system stability. A concurrency control mechanism is implemented to manage request queues and limit concurrency, preventing system overload. A real-time progress tracking and status monitoring system is established to promptly detect and handle anomalies. Automatic retry and error recovery strategies are designed to ensure the robustness of the generation process, forming a complete error recovery mechanism. During decoding, the LLM core module receives the input encoded in step S104, generates the raw output through model inference, and then parses it into structured data by the decoder output processing module.

[0077] Therefore, the core driving problem and expected application scenarios of the user behavior data to be generated were identified. Based on this, the requirements for a large language model data generation system driven by RFLP were determined. These requirements were standardized and coded into multiple functional modules, which formed a data generation function set. Based on this, a logical scheme based on "strategy-component-composition" was constructed for the system and transformed into an executable scheme. Finally, the large language model data was controlled to execute the executable scheme to generate the user behavior data to be generated. This solves the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and making it difficult to guarantee the quality and consistency of generated data. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0078] Furthermore, to ensure that the generated target data meets the preset quality standards and application requirements, a data quality assessment is necessary.

[0079] Optionally, in some embodiments, after generating the user behavior data to be generated, the method further includes: Perform quality verification on the generated user behavior data to obtain the quality verification results; Determine whether the quality verification results meet the preset verification conditions; If the quality verification result does not meet the preset verification conditions, the steps of determining the task objectives, data structure requirements, data characteristic requirements and constraints of the RFLP-driven large language model data generation system based on the core driving problem and expected application scenario will be repeated until the quality verification result of the new user behavior data meets the preset verification conditions.

[0080] Specifically, a multi-dimensional evaluation system is implemented: Automated assessment based on evaluation indicators defined in the requirements phase. This includes comparing the statistical distribution characteristics of generated data with real data, including similarity analysis of statistics such as mean, variance, skewness, and kurtosis, to verify statistical features. Verification of whether the generated data conforms to predefined format requirements, field types, and value range constraints is also performed, conducting format compliance checks. The logical consistency within individual user data is checked, such as the matching degree between user profiles and behavioral patterns, to achieve logical consistency verification. The coverage and diversity of the generated data are evaluated. Finally, by comparing patterns with real data, the realism of the generated data is assessed.

[0081] Downstream application performance verification: Verify the practicality of the synthesized data through real-world application scenarios. Train the target model using the generated data, evaluate the performance improvement effect, and complete the model training effect verification. Integrate the generated data into downstream systems to verify compatibility and functional correctness, and conduct system integration testing. Conduct comparative testing in a controlled environment to quantify the business value of the generated data.

[0082] Based on the comprehensive evaluation results above, quality compliance judgment criteria are formulated, including quantitative standards such as statistical feature similarity threshold, format compliance pass rate (e.g., ≥95%), logical consistency score, and downstream application performance improvement indicators. When key quality indicators reach the preset threshold and the evaluation results of two consecutive rounds of data generation are stable, the quality is judged to be up to standard, and the data output process is executed. Final quality confirmation is performed, and the integrity of the generated user behavior data that passes all evaluation criteria is checked to ensure data format standardization, field integrity, and complete consistency with the data structure specifications defined in the requirements phase.

[0083] When the marginal improvement effect of multiple iterations (usually set at 3-5 rounds) falls below a preset threshold (e.g., performance improvement less than 1%), diminishing marginal returns are identified, and the iteration is terminated or the optimization strategy is adjusted. When time, computing resources, or manpower investment reaches the budget limit, resource constraint termination conditions are triggered. When business requirements undergo significant changes or fundamental design flaws are discovered, it is determined that the RFLP process needs to be restarted. If quality does not meet the standards and the termination conditions are not triggered, the process returns to the S102 requirements phase, tracing the identified quality issues back to the requirements level and determining whether there are potential defects, including but not limited to target positioning deviations, incomplete application scenario definitions, improper quality standard settings, missing constraints, and insufficient identification of non-functional requirements. After reviewing the requirements definition, based on the revised requirements, the previous steps are sequentially advanced, and functional decomposition, logical architecture design, physical solution generation, and implementation are re-performed to ensure the consistency and integrity of the entire system design. At each stage, the impact of the revisions on the design of that stage must be reviewed, and corresponding adjustments are made as necessary.

[0084] Therefore, after quality assessment and optimization, the data quality meets the standards, which can improve the model training effect and system reliability, and enhance the application value.

[0085] To enable those skilled in the art to further understand the user behavior data augmentation method based on RFLP-driven large language model of the present application embodiments, the following description is provided in conjunction with specific embodiments.

[0086] like Figure 3 As shown, Figure 3 This is a flowchart of a user behavior data augmentation method based on an RFLP-driven large language model according to a specific embodiment of this application. The method includes the following steps: S301 defines the core issues or overall goals of user data generation, and analyzes and clarifies the expected application scenarios of the synthesized data.

[0087] S302 defines the requirements for a large language model user data generation system based on RFLP.

[0088] S303 refines the system requirements of the previous stage into discrete functions of a large language model user data generation system based on RFLP, forming a functional set of data generation tasks.

[0089] S304. Construct a large language model user data generation system logic based on a "strategy-component-combination" approach according to the function set. Select a learning strategy, design data generation logic components, and configure the combination of components to form a large language model user data generation system logic scheme that supports diverse tasks.

[0090] S305 transforms the logical scheme into the physical implementation of an executable large language model user data generation system. Based on an encoder-decoder framework, it completes the physical process of generating large language model user data, including data input, analysis and preprocessing, prompt word design, model configuration, API integration, and output processing.

[0091] S306, Execution, Evaluation, and Iterative Optimization. Execute the physical implementation scheme of the large language model user data generation system, generate user behavior data, and conduct quality verification.

[0092] S307, determine whether the quality assessment requirements are met. If they are met, proceed to step S308; otherwise, return to step S302. S308 outputs high-quality user behavior data.

[0093] like Figure 4 As shown, Figure 4This is a schematic diagram of a large language model user data generation system based on RFLP driven according to an embodiment of this application. Steps S401 to S405 constitute an encoder, encoding business requirements layer by layer into an executable solution; the LLM and its input / output processing modules in step S406 constitute a decoder, decoding the encoded solution into user data; S406 provides a quality verification and iterative optimization mechanism to ensure the closed-loop operation of the entire system. This architecture design ensures the systematic transformation from requirements to implementation, as well as the continuous optimization of generation quality.

[0094] Therefore, this application adopts a large language model driven by RFLP. RFLP, as a mature model in the field of systems engineering, provides a structured analysis and implementation path for the design of complex systems. Combining the RFLP framework with LLM cue word design enables systematic requirements analysis. Through the requirements phase of RFLP, the specific requirements, quality standards, and application scenarios of the synthesized data can be clearly defined, laying the foundation for subsequent functional design. In the functional phase, the complex user data generation task is decomposed into specific LLM functional requirements, such as user profile generation, behavior sequence simulation, and preference modeling. In the logical phase, appropriate cue strategies (such as zero-shot, few-shot, and thought chain) and component configurations are selected to form an abstract solution architecture. In the physical phase, the logical design is transformed into specific cue word text, model configuration, and API integration to realize an executable data generation system. This combination provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0095] The user behavior data augmentation method based on RFLP-driven large language model proposed in this application identifies the core driving problem and expected application scenario of the user behavior data to be generated. Based on this, the requirements of the RFLP-driven large language model data generation system are determined. These requirements are standardized and encoded into multiple functional modules, forming a data generation function set. Based on this, a logical scheme based on "strategy-component-combination" is constructed for the system and transformed into an executable scheme. Finally, the large language model data is controlled to execute the executable scheme to generate the user behavior data to be generated. This method solves the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and difficulty in guaranteeing the quality and consistency of generated data. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0096] Next, referring to the accompanying drawings, a user behavior data enhancement device based on an RFLP-driven large language model, according to an embodiment of this application, is described.

[0097] Figure 5 This is a block diagram of a user behavior data enhancement device based on an RFLP-driven large language model according to an embodiment of this application.

[0098] like Figure 5 As shown, the user behavior data enhancement device 10 based on the RFLP-driven large language model includes: a determination module 100, a requirement specification module 200, a function parsing module 300, a logic combination module 400, a physical processing module 500, and a control module 600.

[0099] Among them, the determination module 100 is used to determine the core driving issues and expected application scenarios of the user behavior data to be generated; The requirements specification module 200 is used to determine the task objectives, data structure requirements, data characteristic requirements, constraints and non-functional requirements of the RFLP-driven large language model data generation system based on the core driving problem and expected application scenarios. The functional parsing module 300 is used to encode the task objectives, data structure requirements, data characteristic requirements, constraints and non-functional requirements into multiple functional modules, and form a functional set of data generation tasks based on the functions of multiple functional modules. The logic combination module 400 is used to build a logic scheme based on "strategy-component-composition" for the large language model data generation system based on the function set; The physical processing module 500 is used to convert the logical scheme into an executable scheme; Control module 600 is used to control the execution of the executable scheme of the large language model and generate user behavior data to be generated. According to one embodiment of this application, the control module 600 is further configured to: Perform quality verification on the generated user behavior data to obtain the quality verification results; Determine whether the quality verification results meet the preset verification conditions; If the quality verification result does not meet the preset verification conditions, the steps of determining the task objectives, data structure requirements, data characteristic requirements and constraints of the RFLP-driven large language model data generation system based on the core driving problem and expected application scenario will be repeated until the quality verification result of the new user behavior data meets the preset verification conditions.

[0100] According to one embodiment of this application, the requirement specification module 200 is specifically used for: Based on the core driving problem and the expected application scenario, identify at least one goal to be achieved, and determine the task objectives according to the business requirements corresponding to the at least one goal to be achieved. Based on the core driving problem and expected application scenarios, the necessary data types and attributes of the user behavior data to be generated are determined, and the data structure requirements are determined according to the necessary data types and attributes. Based on the core driving problem and the expected application scenario, the qualitative and quantitative characteristics of the user behavior data to be generated are determined, and the data characteristic requirements are determined according to the qualitative and quantitative characteristics. Based on the core driving problem and expected application scenarios, the operational boundaries of the large language model generation process are determined, and constraints are determined based on the operational boundaries.

[0101] According to one embodiment of this application, non-functional requirements include at least one of the following: accuracy requirements, authenticity requirements, task relevance requirements, simplicity requirements, coherence and fluency requirements, safety and hazard avoidance requirements, bias mitigation requirements, and robustness requirements.

[0102] According to one embodiment of this application, the multiple functional modules include various types of input analysis functional modules, core generation functional modules, control functional modules, and output functional modules.

[0103] It should be noted that the foregoing explanation of the user behavior data augmentation method embodiment based on RFLP-driven large language model also applies to the user behavior data augmentation device based on RFLP-driven large language model in this embodiment, and will not be repeated here.

[0104] The user behavior data enhancement device based on an RFLP-driven large language model proposed in this application identifies the core driving problem and expected application scenario of the user behavior data to be generated. Based on this, the requirements for the RFLP-driven large language model data generation system are determined. These requirements are standardized and encoded into multiple functional modules, forming a data generation function set. Based on this, a logical scheme based on "strategy-component-combination" is constructed for the system and transformed into an executable scheme. Finally, the large language model data is controlled to execute the executable scheme to generate the user behavior data to be generated. This solves the problems of existing prompt word design lacking systematicity, relying on experience and trial and error, and difficulty in guaranteeing the quality and consistency of generated data in the prior art. It provides scientific methodological support for building a high-quality, controllable, and scalable synthetic user data generation system.

[0105] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. The electronic device may include: The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0106] When the processor 602 executes the program, it implements the user behavior data augmentation method based on the RFLP-driven large language model provided in the above embodiments.

[0107] Furthermore, electronic devices also include: Communication interface 603 is used for communication between memory 601 and processor 602.

[0108] The memory 601 is used to store computer programs that can run on the processor 602.

[0109] The memory 601 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0110] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0111] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0112] The processor 602 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

[0113] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described user behavior data augmentation method based on an RFLP-driven large language model.

[0114] This application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the user behavior data augmentation method based on an RFLP-driven large language model.

[0115] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0116] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0117] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for augmenting user behavior data based on a large language model driven by RFLP, characterized in that, Includes the following steps: Identify the core driving issues and expected application scenarios for the user behavior data to be generated; Based on the core driving problem and the expected application scenario, the task objectives, data structure requirements, data characteristic requirements, constraints and non-functional requirements of the RFLP-driven large language model data generation system are determined. The task objective, data structure requirements, data characteristic requirements, constraints, and non-functional requirements are encoded into multiple functional modules, and a functional set for the data generation task is formed based on the functions of the multiple functional modules. Based on the aforementioned function set, a logical scheme based on "strategy-component-composition" is constructed for the large language model data generation system; Transform the logical scheme into an executable scheme; Control the large language model to execute the executable scheme and generate the user behavior data to be generated; The process includes, after generating the user behavior data to be generated, the following steps: performing quality verification on the user behavior data to be generated to obtain a quality verification result; determining whether the quality verification result meets a preset verification condition; if the quality verification result does not meet the preset verification condition, then re-executing the steps of determining the task objectives, data structure requirements, data characteristic requirements, and constraints of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario, until the quality verification result of the new user behavior data meets the preset verification condition. The aforementioned logical scheme refers to an abstract scheme that systematically sorts out and designs the operational logic, core elements, and interrelationships of a large language model data generation system based on a functional set and following a strategy-component-combination framework.

2. The method according to claim 1, characterized in that, Based on the core driving problem and the expected application scenario, the task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements of the RFLP-driven large language model data generation system are determined, including: Based on the core driving problem and the expected application scenario, at least one goal to be achieved is identified, and the task objective is determined according to the business requirements corresponding to at least one of the goals to be achieved. Based on the core driving problem and the expected application scenario, the necessary data types and attributes of the user behavior data to be generated are determined, and the data structure requirements are determined according to the necessary data types and attributes. Based on the core driving problem and the expected application scenario, the qualitative and quantitative characteristics of the user behavior data to be generated are determined, and the data characteristic requirements are determined according to the qualitative and quantitative characteristics. Based on the core driving problem and the expected application scenario, the operational boundary of the large language model generation process is determined, and the constraint conditions are determined according to the operational boundary.

3. The method according to claim 2, characterized in that, The non-functional requirements include at least one of the following: accuracy requirements, authenticity requirements, task relevance requirements, simplicity requirements, coherence and fluency requirements, safety and hazard avoidance requirements, bias mitigation requirements, and robustness requirements.

4. The method according to claim 1, characterized in that, The multiple functional modules include various types of input analysis functional modules, core generation functional modules, control functional modules, and output functional modules.

5. A user behavior data augmentation device based on an RFLP-driven large language model, characterized in that, include: The determination module is used to identify the core driving issues and expected application scenarios for the user behavior data to be generated. The requirements specification module is used to determine the task objectives, data structure requirements, data characteristic requirements, constraints, and non-functional requirements of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario. The function parsing module is used to encode the task objective, the data structure requirements, the data characteristic requirements, the constraints, and the non-functional requirements into multiple functional modules, and to form a function set for the data generation task based on the functions of the multiple functional modules. The logic combination module is used to construct a logical scheme based on "strategy-component-composition" for the large language model data generation system based on the function set. The physical processing module is used to convert the logical scheme into an executable scheme; The control module is used to control the large language model to execute the executable scheme and generate the user behavior data to be generated; The control module is further configured to: perform quality verification on the user behavior data to be generated, and obtain a quality verification result; and determine whether the quality verification result meets the preset verification conditions. If the quality verification result does not meet the preset verification conditions, then the steps of determining the task objectives, data structure requirements, data characteristic requirements and constraints of the RFLP-driven large language model data generation system based on the core driving problem and the expected application scenario are re-executed until the quality verification result of the new user behavior data meets the preset verification conditions. The aforementioned logical scheme refers to an abstract scheme that systematically sorts out and designs the operational logic, core elements, and interrelationships of a large language model data generation system based on a functional set and following a strategy-component-combination framework.

6. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the user behavior data augmentation method based on an RFLP-driven large language model as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the user behavior data augmentation method based on RFLP-driven large language models as described in any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the user behavior data augmentation method based on RFLP-driven large language model as described in any one of claims 1-4.

Images