A parameterized fusion analysis method and device for multi-source heterogeneous data
By using a parametric fusion analysis method, the problems of repetitive development and insufficient real-time performance in the fusion of multi-source heterogeneous data are solved. This method enables data to be configurable, scalable, and traceable, reduces access and change costs, improves data reuse rate, and ensures data security and regulatory compliance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN INST OF INFORMATION TECH
- Filing Date
- 2026-05-12
- Publication Date
- 2026-07-31
AI Technical Summary
Existing technologies for multi-source heterogeneous data fusion and analysis suffer from several drawbacks, including repetitive point-to-point customization and integration, insufficient real-time performance of traditional ETL batch processing, fragmented result formats, strong coupling between business rules and data retrieval, and risks of context length explosion and sensitive data leakage when combined with large models.
A parametric fusion analysis method for multi-source heterogeneous data is adopted. By configuring and acquiring data sources, datasets and transformation strategy chains, data acquisition is performed and uniformly mapped into key-value pairs. Parametric processing and transformation orchestration are then carried out, and finally, a fused data view is output and traced.
It reduces access and modification costs, improves dataset reuse, and ensures data security and regulatory compliance through layered transformation and data retrieval.
Smart Images
Figure CN122489643A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of big data processing and artificial intelligence technology, specifically relating to a parametric fusion analysis method and apparatus for multi-source heterogeneous data. Background Technology
[0002] In the context of digital transformation, enterprise operational data is widely distributed across OLTP business databases, OLAP analysis databases, log and retrieval clusters, third-party SaaS APIs, and self-built algorithm services, exhibiting characteristics of multi-source, heterogeneous, real-time, and batch processing. Management and business teams require unified indicator definitions, reusable analytical assets, and auditable data lineage to support scenarios such as regulatory reporting, operational dashboards, management control panels, and intelligent Q&A.
[0003] In existing technologies, multi-source data fusion and analysis typically employ the following methods: developing independent adaptation code for each data source, using ETL tools for batch extraction, transformation, and loading, or hard-coding different interfaces and concatenating the results. However, existing solutions generally suffer from the following problems: (1) Point-to-point customized integration: Each new data source requires independent development of adaptation code, resulting in serious duplication of work, fragmented versions, and difficulty in reusing data retrieval logic.
[0004] (2) Limitations of traditional ETL batch processing: It mainly extracts data at fixed intervals, and lacks real-time and interactive analysis capabilities; changes in business rules require rerunning the entire ETL task, resulting in slow response.
[0005] (3) Fragmented result format: Relational databases return row and column structures, search engines return column names plus row arrays, and HTTP services return JSON objects or arrays. Various response formats lack a unified mapping, and upper-layer applications need to write a lot of adaptation code.
[0006] (4) Strong coupling between business rules and data retrieval: Data cleaning, dictionary translation, field conversion and other logics are scattered in various places, making it difficult to audit, reuse and version management.
[0007] (5) Risks of combining with large models: Directly using the original library tables or interface results as input to a large language model (LLM) can easily lead to context length explosion, model illusion, and leakage of sensitive data.
[0008] Therefore, there is an urgent need for a configurable, scalable, and traceable method for fusing and analyzing multi-source heterogeneous data to solve the above problems. Summary of the Invention
[0009] The technical problem to be solved by the present invention is to provide a configurable, scalable, and traceable multi-source heterogeneous data fusion and analysis method and apparatus to address the shortcomings of the prior art.
[0010] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A parametric fusion analysis method for multi-source heterogeneous data includes the following steps: Step 1, Configuration Acquisition: Acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence. Step 2, Heterogeneous Execution and Normalization: Select the corresponding execution path according to the data source type identifier, perform data acquisition operations and obtain raw response data; according to the mapping rules corresponding to the data source type identifier, map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Step 3: Parameterized execution of the dataset. Receive request parameters, validate the request parameters, and if the validation passes, replace the placeholders in the dynamic template with the request parameters to generate executable content. Execute the executable content to obtain the original response data described in Step B. Step 4, Transformation and Arrangement: Obtain the record list generated in Step B, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Step 5, merge output and traceability: output a merged data view containing the final record list, and save traceability information. The traceability information includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
[0011] As a further preferred embodiment of the parametric fusion analysis method for multi-source heterogeneous data of the present invention, the data source type identifier includes one or more of the following: SQL for relational databases, Elasticsearch for retrieval analysis engines, HTTP for general HTTP services, and LLM for large language model inference services.
[0012] As a further preferred embodiment of the parametric fusion analysis method for multi-source heterogeneous data of the present invention, in step 2, the original response data is uniformly mapped into a record list in key-value pair form according to the mapping rules corresponding to the data source type identifier, specifically including the following steps: Step 2.1: When the data source type is identified as SQL, the column names in the row and column structure returned by the relational database are mapped to keys, and the corresponding row values are mapped to values to generate a set of records; Step 2.2: When the data source type is identified as es, the column name array returned by the retrieval analysis engine is aligned with the row data array by column to generate a set of records; Step 2.3: When the data source type is identified as http or LLM, if the returned JSON data is an object, it is mapped to a single record; if the returned JSON data is an array, each array element is mapped to a record.
[0013] As a further preferred embodiment of the parameterized fusion analysis method for multi-source heterogeneous data of the present invention, in step 3, the request parameters are verified, specifically including one or more of the following: verification of the mandatory content of the request parameters, data type verification, and SQL injection pattern risk detection.
[0014] As a further preferred embodiment of the parametric fusion analysis method for multi-source heterogeneous data of the present invention, the conversion strategy includes a script-based conversion strategy and a dictionary mapping conversion strategy; The script-based conversion strategy is used to execute custom function logic. The dictionary mapping conversion strategy is used to map field encoded values to corresponding display names.
[0015] As a further preferred embodiment of the parametric fusion analysis method for multi-source heterogeneous data of the present invention, when the configured data source type is identified as LLM, the dynamic template contains prompt words or context information sent to the large language model; the HTTP request body generated in step 3 contains the dynamic template after parameterized replacement. As a further preferred embodiment of the parametric fusion analysis method for multi-source heterogeneous data of the present invention, it further includes: taking the response returned by the large language model inference service, passing it through the unified mapping in step 2 and the transformation and arrangement in step 4 in sequence to obtain a structured model output, and associating and fusing the model output with records from other data sources, and including them together in the fused data view.
[0016] A parametric fusion analysis device based on a parametric fusion analysis method for multi-source heterogeneous data, comprising: Configuration Acquisition Module: Used to acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence; Heterogeneous execution adaptation module: used to select the corresponding execution path according to the data source type identifier, perform data acquisition operation and obtain raw response data; and according to the mapping rule corresponding to the data source type identifier, uniformly map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Dataset parameterization module: used to receive request parameters, validate the request parameters, and after the validation is successful, use the request parameters to replace the placeholders in the dynamic template to generate executable content, and trigger the heterogeneous execution adaptation module to execute the executable content; Transformation orchestration module: used to obtain the record list generated by the heterogeneous execution adaptation module, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Fusion Output and Traceability Module: Used to output a fused data view containing the final record list and to save traceability information, which includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
[0017] As a further preferred embodiment of the parametric fusion analysis device for multi-source heterogeneous data of the present invention, the mapping rules built into the heterogeneous execution adaptation module include: for SQL type data sources, mapping the column names in the row and column structure returned by the relational database to keys, and mapping the corresponding row values to values; For Elasticsearch data sources, align the array of column names returned by the search engine with the array of row data by column; For HTTP or LLM data sources, if the returned JSON data is an object, it is mapped to a single record; if it is an array, each element is mapped to a record.
[0018] As a further preferred embodiment of the parameterized fusion analysis device for multi-source heterogeneous data of the present invention, the dataset parameterization module includes a validation submodule, which is used to perform mandatory validation, data type validation, and SQL injection pattern risk detection on the request parameters; the conversion orchestration module supports conversion strategies including script-based conversion strategies and dictionary mapping conversion strategies; the script-based conversion strategy is used to execute custom function logic; the dictionary mapping conversion strategy is used to map field encoded values to corresponding display names.
[0019] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention reduces access and change costs: adding a new data source type only requires extending the mapping logic of the adaptation layer, and the upper-layer application is unaware of it, greatly reducing repetitive development work; 2. This invention improves dataset reusability: Parameterized datasets can reuse the same data retrieval template in different business scenarios, avoiding redundant development and improving development efficiency; 3. The present invention features a layered conversion and data retrieval system: changes to business rules only require adjustments to the conversion strategy configuration, without modifying the data retrieval code, and the strategy execution is auditable, facilitating rule management; 4. This invention provides a controllable data contract for LLM: it desensitizes and structures the model input, reduces the risk of context explosion and illusion, and ensures data security; 5. The data lineage of this invention is traceable: Each fused view is associated with complete traceability metadata, meeting regulatory compliance requirements. Attached Figure Description
[0020] Figure 1 This is a flowchart of the overall fusion analysis method of the present invention.
[0021] Figure 2 This is a schematic diagram illustrating the heterogeneous data source adaptation and result standardization of the present invention.
[0022] Figure 3 This is a timing diagram for parameter verification and dynamic replacement in this invention.
[0023] Figure 4 This is a schematic diagram of the pluggable conversion strategy chain of the present invention.
[0024] Figure 5 This is a diagram showing the fusion output and traceability relationship of the present invention.
[0025] Figure 6 This is a diagram illustrating the collaborative architecture between the present invention and the large language model.
[0026] Figure 7 This is a diagram showing the modular composition of the device of the present invention. Detailed Implementation
[0027] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings: The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.
[0028] like Figures 1 to 7 As shown, a parametric fusion analysis method for multi-source heterogeneous data includes the following steps: Step 1, Configuration Acquisition: Acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence. Step 2, Heterogeneous Execution and Normalization: Select the corresponding execution path according to the data source type identifier, perform data acquisition operations and obtain raw response data; according to the mapping rules corresponding to the data source type identifier, map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Step 3: Parameterized execution of the dataset. Receive request parameters, validate the request parameters, and if the validation passes, replace the placeholders in the dynamic template with the request parameters to generate executable content. Execute the executable content to obtain the original response data described in Step B. Step 4, Transformation and Arrangement: Obtain the record list generated in Step B, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Step 5, merge output and traceability: output a merged data view containing the final record list, and save traceability information. The traceability information includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
[0029] The data source type identifier includes one or more of the following: SQL for relational databases, Elasticsearch for retrieval analysis engines, HTTP for general HTTP services, and LLM for large language model inference services.
[0030] In step 2, according to the mapping rules corresponding to the data source type identifier, the original response data is uniformly mapped into a list of records in key-value pair format, specifically including the following steps: Step 2.1: When the data source type is identified as SQL, the column names in the row and column structure returned by the relational database are mapped to keys, and the corresponding row values are mapped to values to generate a set of records; Step 2.2: When the data source type is identified as es, the column name array returned by the retrieval analysis engine is aligned with the row data array by column to generate a set of records; Step 2.3: When the data source type is identified as http or LLM, if the returned JSON data is an object, it is mapped to a single record; if the returned JSON data is an array, each array element is mapped to a record.
[0031] In step 3, the request parameters are validated, specifically including one or more of the following: validation of the required fields of the request parameters, validation of the data type, and detection of SQL injection patterns.
[0032] The conversion strategies include script-based conversion strategies and dictionary mapping-based conversion strategies; The script-based conversion strategy is used to execute custom function logic. The dictionary mapping conversion strategy is used to map field encoded values to corresponding display names.
[0033] When the configured data source type is identified as LLM, the dynamic template contains prompt words or context information sent to the large language model; the HTTP request body generated in step 3 contains the dynamic template after parameterized replacement. It also includes: taking the response returned by the large language model inference service, passing it through the unified mapping in step 2 and the transformation and arrangement in step 4 to obtain a structured model output, and then associating and fusing the model output with records from other data sources, and including them in the fused data view.
[0034] A parametric fusion analysis device based on a parametric fusion analysis method for multi-source heterogeneous data, comprising: Configuration Acquisition Module: Used to acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence; Heterogeneous execution adaptation module: used to select the corresponding execution path according to the data source type identifier, perform data acquisition operation and obtain raw response data; and according to the mapping rule corresponding to the data source type identifier, uniformly map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Dataset parameterization module: used to receive request parameters, validate the request parameters, and after the validation is successful, use the request parameters to replace the placeholders in the dynamic template to generate executable content, and trigger the heterogeneous execution adaptation module to execute the executable content; Transformation orchestration module: used to obtain the record list generated by the heterogeneous execution adaptation module, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Fusion Output and Traceability Module: Used to output a fused data view containing the final record list and to save traceability information, which includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
[0035] The mapping rules built into the heterogeneous execution adaptation module include: for SQL type data sources, mapping column names in the row and column structure returned by the relational database to keys, and mapping corresponding row values to values; For Elasticsearch data sources, align the array of column names returned by the search engine with the array of row data by column; For HTTP or LLM data sources, if the returned JSON data is an object, it is mapped to a single record; if it is an array, each element is mapped to a record.
[0036] The dataset parameterization module includes a validation submodule, used to perform mandatory field validation, data type validation, and SQL injection pattern risk detection on request parameters; the conversion orchestration module supports conversion strategies including script-based conversion strategies and dictionary mapping conversion strategies; the script-based conversion strategy is used to execute custom function logic; the dictionary mapping conversion strategy is used to map field encoded values to corresponding display names.
[0037] This invention proposes a four-layer logical architecture: First layer: Heterogeneous execution adaptation layer; The execution path is selected based on the data source type identifier (SQL / ES / HTTP / LLM), the connection and call are completed, and the raw response is obtained. Then, responses of different formats are uniformly mapped to a structured list of key-value pairs. Specific mapping rules: Relational type: Column name → key, row value → value; Search and analysis engine: Parses the array of column names and the array of row data, aligning them by column; HTTP JSON object: mapped to a single record; HTTP JSON array: Each element maps to a record.
[0038] Second layer: Dataset parameterization layer: The dataset configuration binds the dataset identifier, data source, and dynamic template (SQL or HTTP request body, including placeholders such as ${param}). It performs mandatory field validation, type validation, and SQL injection pattern detection on the request context parameters. Once successful, the placeholders are replaced with the actual parameter values, generating executable content.
[0039] Third layer: Transition and arrangement layer: Based on the configuration, strategies such as script-based transformation (JavaScript / Groovy engine executes custom functions), dictionary mapping transformation (encoding → display name), and conditional filtering are applied in sequence to obtain the transformed list of records.
[0040] Fourth layer: Fusion output and traceability layer: Output a merged data view (JSON list or table file), and save traceability information such as data source identifier, dataset identifier, transformation strategy type, and timestamp.
[0041] Optional extension: Collaboration with large language models; Configure the data source to point to the HTTP interface of the LLM inference service, with the request body containing de-identified and parameterized hints or RAG context. The LLM response is then processed through the same normalization and transformation pipeline before being associated and merged with the structured data.
[0042] Example 1 (Basic Fusion Scenario): Suppose there is a relational database (MySQL) and an HTTP service (returning JSON arrays), and you need to merge the data from both.
[0043] Data source configuration: Source A is of type SQL, and the connection string is jdbc:mysql: / / ...; Source B is of type HTTP, and the URL is https: / / api.example.com / orders.
[0044] Dataset configuration: Dataset DS1 is bound to source A, and the dynamic SQL template is SELECT * FROM orders WHERE order_id = ${orderId}; Dataset DS2 is bound to source B, and the dynamic template is {"query":"${keyword}"}.
[0045] Request parameters: {orderId: "1001", keyword: "laptop"}.
[0046] After successful verification, the executable SQL and request body are obtained by replacement.
[0047] After execution, source A returns a result set, mapped as [{order_id:1001, amount:299}]; source B returns a JSON array, mapped as a similar structure.
[0048] Configure conversion strategy: Apply a currency conversion script (JavaScript) to the amount field to convert USD to CNY; apply a dictionary mapping (1 → "Paid") to the status field.
[0049] Output the merged view and record traceability information.
[0050] Example 2 (Scenario of Collaboration with Large Models); Configure an HTTP data source of type LLM, pointing to an LLM inference service. The dataset template contains the prompt: "Analyze the cause of the anomaly in order ${orderId}, order details: ${orderInfo}". Here, ${orderInfo} comes from the query results of the aforementioned relational database. First, retrieve the order details from the database using a parameterized dataset, replace them, and then call the LLM interface. Associate the text returned by the LLM with the original data by order ID, finally outputting a fused view containing "order information + model analysis conclusions".
[0051] Example 3 (Safety Control): The parameter validation phase includes SQL injection pattern detection (such as detecting dangerous patterns like OR 1=1, DROP, etc.) and HTTP domain whitelist verification. Sensitive connection information (database password, API key) is centrally managed by the configuration center and is not exposed in plaintext in the template.
[0052] The complete design of the four-layer logical architecture, especially the parameterized dataset and pluggable transformation orchestration as independent layers.
[0053] The parameterized dataset mechanism enables declarative reuse and runtime dynamic replacement of data retrieval logic.
[0054] A pluggable conversion strategy chain supports the chained execution of script-based and dictionary-mapped strategies, completely decoupled from the data retrieval logic.
[0055] A unified standardization method for heterogeneous responses, covering common formats such as relational, search engine, HTTP, and JSON.
[0056] Controllable collaboration with LLM inference services enables secure fusion of structured data and large model outputs through a unified data contract layer.
[0057] Alternative solutions to the present invention: (1) A unified intermediate representation can be achieved by using a columnar storage format (such as Apache Arrow) instead of a list of key-value pairs.
[0058] (2) Transformation and arrangement can be implemented using a rule engine (such as Drools) or a DSL.
[0059] (3) Interaction with LLM can be achieved by using asynchronous message queues (such as Kafka) instead of synchronous HTTP calls.
[0060] (4) Traceability information can be written into a dedicated bloodline database or data lake.
[0061] It will be understood by those skilled in the art that the above description is merely a preferred embodiment of the invention and is not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still understand the foregoing... The technical solutions described in each example may be modified, or some of the technical features may be replaced with equivalent ones. All modifications and equivalent replacements made within the spirit and principles of the invention shall be included within the scope of protection of the invention. All technical features in this embodiment can be freely combined according to actual needs.
[0062] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A parameterized fusion analysis method of multi-source heterogeneous data, characterized in that, Includes the following steps: Step 1, Configuration Acquisition: Acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence. Step 2, Heterogeneous Execution and Normalization: Select the corresponding execution path according to the data source type identifier, perform data acquisition operations and obtain raw response data; according to the mapping rules corresponding to the data source type identifier, map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Step 3: Parameterized execution of the dataset. Receive request parameters, validate the request parameters, and if the validation passes, replace the placeholders in the dynamic template with the request parameters to generate executable content. Execute the executable content to obtain the original response data described in Step B. Step 4, Transformation and Arrangement: Obtain the record list generated in Step B, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Step 5, merge output and traceability: output a merged data view containing the final record list, and save traceability information. The traceability information includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
2. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, The data source type identifier includes one or more of the following: SQL for relational databases, Elasticsearch for retrieval analysis engines, HTTP for general HTTP services, and LLM for large language model inference services.
3. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, In step 2, according to the mapping rules corresponding to the data source type identifier, the original response data is uniformly mapped into a list of records in key-value pair format, specifically including the following steps: Step 2.1: When the data source type is identified as SQL, the column names in the row and column structure returned by the relational database are mapped to keys, and the corresponding row values are mapped to values to generate a set of records; Step 2.2: When the data source type is identified as es, the column name array returned by the retrieval analysis engine is aligned with the row data array by column to generate a set of records; Step 2.3: When the data source type is identified as http or LLM, if the returned JSON data is an object, it is mapped to a single record; if the returned JSON data is an array, each array element is mapped to a record.
4. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, In step 3, the request parameters are validated, specifically including one or more of the following: validation of the required fields of the request parameters, validation of the data type, and detection of SQL injection patterns.
5. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, The conversion strategies include script-based conversion strategies and dictionary mapping-based conversion strategies; The script-based conversion strategy is used to execute custom function logic. The dictionary mapping conversion strategy is used to map field encoded values to corresponding display names.
6. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, When the configured data source type is identified as LLM, the dynamic template contains prompt words or context information sent to the large language model; the HTTP request body generated in step 3 contains the dynamic template after parameterized replacement.
7. The parametric fusion analysis method for multi-source heterogeneous data according to claim 1, characterized in that, Also includes: The response returned by the large language model inference service is processed sequentially through the unified mapping in step 2 and the transformation and arrangement in step 4 to obtain a structured model output. This model output is then associated and fused with records from other data sources and included in the fused data view.
8. A parametric fusion analysis apparatus based on the parametric fusion analysis method for multi-source heterogeneous data according to any one of claims 1 to 7, characterized in that, Include: Configuration Acquisition Module: Used to acquire data source configuration, dataset configuration, and transformation strategy chain configuration; the data source configuration includes data source type identifier and connection information; the dataset configuration includes dataset identifier, bound data source identifier, and dynamic template containing placeholders; the transformation strategy chain configuration includes at least one transformation strategy arranged in sequence; Heterogeneous execution adaptation module: used to select the corresponding execution path according to the data source type identifier, perform data acquisition operation and obtain raw response data; and according to the mapping rule corresponding to the data source type identifier, uniformly map the raw response data into a record list in key-value pair form, where each record is a set of key-value pairs; Dataset parameterization module: used to receive request parameters, validate the request parameters, and after the validation is successful, use the request parameters to replace the placeholders in the dynamic template to generate executable content, and trigger the heterogeneous execution adaptation module to execute the executable content; Transformation orchestration module: used to obtain the record list generated by the heterogeneous execution adaptation module, and apply each transformation strategy to process the record list in sequence according to the order configured in the transformation strategy chain to obtain the final record list; Fusion Output and Traceability Module: Used to output a fused data view containing the final record list and to save traceability information, which includes at least the currently used data source identifier, dataset identifier, transformation strategy chain type identifier, and timestamp.
9. The parametric fusion analysis device for multi-source heterogeneous data according to claim 7, characterized in that, The mapping rules built into the heterogeneous execution adaptation module include: for SQL type data sources, mapping column names in the row and column structure returned by the relational database to keys, and mapping corresponding row values to values; For Elasticsearch data sources, align the array of column names returned by the search engine with the array of row data by column; For HTTP or LLM data sources, if the returned JSON data is an object, it is mapped to a single record; if it is an array, each element is mapped to a record.
10. The parametric fusion analysis device for multi-source heterogeneous data according to claim 7, characterized in that, The dataset parameterization module includes a validation submodule, which is used to perform mandatory field validation, data type validation, and SQL injection pattern risk detection on request parameters. The conversion orchestration module supports conversion strategies including script-based conversion strategies and dictionary mapping conversion strategies; the script-based conversion strategies are used to execute custom function logic; the dictionary mapping conversion strategies are used to map field encoded values to corresponding display names.