Data-driven decision generation systems, methods, apparatus, and electronic devices

By working together with the processing module, evidence graph construction and calculation module, multi-objective decision-making module and verification module, the problem of interpretability arbitration mechanism for logical conflicts between multiple sources of information and comprehensive balancing of multiple business objectives is solved, thereby improving the credibility and practical applicability of the decision.

CN121684026BActive Publication Date: 2026-11-10CHINA MOBILE GROUP DESIGN INST +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511781762.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-11-29
Publication Date
2026-11-10
Estimated Expiration
2045-11-29

AI Technical Summary

Technical Problem

In existing technologies, relying on a single model for black-box decision-making makes it difficult to achieve an interpretable arbitration mechanism for logical conflicts between multiple sources of information and a comprehensive balance of multiple business objectives, resulting in low credibility of decision results and difficulty in implementing them in actual engineering.

Method used

Through the coordinated operation of the processing module, evidence graph construction and calculation module, multi-objective decision-making module and verification module, the system achieves standardized integration of multi-source data, precise control of evidence consistency and verification-driven iterative optimization, thereby improving the credibility and practical applicability of target decisions.

Benefits of technology

Through the coordinated operation of the collaborative modules, standardized integration of multi-source data and precise control of evidence consistency were achieved, thereby improving the credibility and practical applicability of decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121684026B_ABST
    Figure CN121684026B_ABST
Patent Text Reader

Abstract

This disclosure presents a data-driven decision generation system, method, apparatus, and electronic device. The system includes a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module. The processing module processes acquired multi-source data to obtain a set of evidence units. The evidence graph construction and calculation module constructs an evidence graph based on the evidence unit set and calculates the consistency score of each evidence unit in the evidence graph. The multi-objective decision-making module generates at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and obtains the target decision based on the rearrangement results of the at least two candidate decisions. If the verification result of the target decision does not meet a preset verification threshold, the verification module updates the evidence unit set based on the verification result and reprocesses the multi-source data until the verification result of the target decision meets the preset verification threshold. This achieves the integration of multi-source data and improves the credibility and practical applicability of the target decision.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and in particular to a data decision generation system, method and apparatus, and electronic device. Background Technology

[0002] With the continuous expansion of network scale and the rapid penetration of various emerging services, the coverage of wireless networks, the number of users accessing the network, and the demand for data transmission are growing exponentially. As a result, the impact of increased disconnection rates, handover failures, and abnormal interference during network operation on communication quality is becoming increasingly significant.

[0003] In related technologies, multi-source information is integrated by constructing knowledge graphs or linear reasoning processes; however, relying on a single model for black-box decision-making makes it difficult to achieve an interpretable arbitration mechanism for logical conflicts between multi-source information and a comprehensive balance of multiple business objectives, resulting in low credibility of decision results and making it difficult to implement in actual engineering. Summary of the Invention

[0004] This disclosure provides a data decision generation system, method, apparatus, and electronic device to solve problems in related technologies. Through the coordinated linkage of a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module, it achieves standardized integration of multi-source data, precise control of evidence consistency, and verification-driven iterative optimization, thereby improving the credibility and practical applicability of target decisions.

[0005] According to a first aspect of this disclosure, a data decision generation system is provided, the system comprising: a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module; wherein; The processing module is used to standardize the acquired multi-source data to obtain a set of evidence units, and send the set of evidence units to the evidence graph construction and calculation module. The evidence graph construction and calculation module is used to construct an evidence graph based on the received set of evidence units, calculate the consistency score of each evidence unit in the evidence graph, and send the evidence graph and the consistency score of each evidence unit to the multi-objective decision module. The multi-objective decision-making module is used to receive the evidence graph and the consistency scores of each evidence unit, generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, obtain the target decision according to the rearrangement result of the at least two candidate decisions, and send the target decision to the verification module. The verification module is used to receive the target decision, and if the verification result of the target decision does not meet the preset verification threshold, update the evidence unit set according to the verification result, and reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold.

[0006] In some embodiments of this disclosure, the processing module is further configured to, The multi-source data is collected and subjected to data cleaning, field mapping, and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data, and alarm data; Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree; Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

[0007] In some embodiments of this disclosure, the evidence graph construction and calculation module is further used for, Using the evidence units as evidence nodes, construct the evidence diagram corresponding to each evidence node; Based on a predefined rule base, edge relationships are constructed between the evidence nodes; wherein, the edge relationships include at least one of support, conflict, reference, temporal association, and association with the same entity; The consistency score of each evidence unit is calculated based on the edge relationship and the initial weight corresponding to the evidence node.

[0008] In some embodiments of this disclosure, the multi-objective decision module is further used to, Based on the evidence graph and the consistency score, at least two candidate decisions are generated; Select a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives; The candidate decisions in the non-dominated solution set are rearranged based on preset business rules to obtain the target decision.

[0009] In some embodiments of this disclosure, the verification module is further configured to, Perform decision verification on the target decision and generate the verification result; wherein, the decision verification includes at least one of tool verification, rule verification and cross-verification; The verification results are quantified into node-level feedback factors, and the weights of the corresponding evidence nodes in the evidence graph are updated based on the node-level feedback factors. When the verification result does not meet the preset verification threshold, a retrieval instruction is generated based on the missing information identified in the verification result, and the retrieved new evidence unit is added to the evidence unit set to update the evidence unit set.

[0010] In some embodiments of this disclosure, the system further includes an audit module, wherein the audit module is used to, Record the entire process of standardizing the multi-source data to obtaining the target decision; When calculating the consistency score and filtering the non-dominated solution set, save the corresponding parameter snapshot; Based on the operation log and the parameter snapshot, the reasoning path and evidence chain of the target decision are traced.

[0011] According to a second aspect of this disclosure, a data decision generation method is provided, comprising: The acquired multi-source data is standardized to obtain a set of evidence units; An evidence graph is constructed based on the set of evidence units, and the consistency score of each evidence unit in the evidence graph is calculated. Based on the evidence graph and the consistency scores of each evidence unit, at least two candidate decisions are generated, and the target decision is obtained based on the rearrangement results of the at least two candidate decisions. If the verification result of the target decision does not meet the preset verification threshold, the evidence unit set is updated according to the verification result, and the multi-source data is reprocessed until the verification result of the target decision meets the preset verification threshold.

[0012] In some embodiments of this disclosure, the standardization process of the acquired multi-source data to obtain a set of evidence units includes: The multi-source data is collected and subjected to data cleaning, field mapping, and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data, and alarm data; Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree; Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

[0013] In some embodiments of this disclosure, the step of constructing an evidence graph based on the set of evidence units and calculating the consistency score of each evidence unit in the evidence graph includes: Using the evidence units as evidence nodes, construct the evidence diagram corresponding to each evidence node; Based on a predefined rule base, edge relationships are constructed between the evidence nodes; wherein, the edge relationships include at least one of support, conflict, reference, temporal association, and association with the same entity; The consistency score of each evidence unit is calculated based on the edge relationship and the initial weight corresponding to the evidence node.

[0014] In some embodiments of this disclosure, generating at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and obtaining the target decision based on the rearrangement results of the at least two candidate decisions, includes: Based on the evidence graph and the consistency score, at least two candidate decisions are generated; Select a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives; The candidate decisions in the non-dominated solution set are rearranged based on preset business rules to obtain the target decision.

[0015] In some embodiments of this disclosure, the step of updating the evidence unit set according to the verification result when the verification result of the target decision does not meet the preset verification threshold, and reprocessing the multi-source data until the verification result of the target decision meets the preset verification threshold, includes: Perform decision verification on the target decision and generate the verification result; wherein, the decision verification includes at least one of tool verification, rule verification and cross-verification; The verification results are quantified into node-level feedback factors, and the weights of the corresponding evidence nodes in the evidence graph are updated based on the node-level feedback factors. When the verification result does not meet the preset verification threshold, a retrieval instruction is generated based on the missing information identified in the verification result, and the retrieved new evidence unit is added to the evidence unit set to update the evidence unit set.

[0016] In some embodiments of this disclosure, after the multi-source data has been reprocessed until the verification result of the target decision meets the preset verification threshold, the method further includes: Record the entire process of standardizing the multi-source data to obtaining the target decision; When calculating the consistency score and filtering the non-dominated solution set, save the corresponding parameter snapshot; Based on the operation log and the parameter snapshot, the reasoning path and evidence chain of the target decision are traced.

[0017] According to a third aspect of this disclosure, a data decision generation apparatus is provided, comprising: The processing unit is used to standardize the acquired multi-source data to obtain a set of evidence units; A construction unit is used to construct an evidence graph based on the set of evidence units and to calculate the consistency score of each evidence unit in the evidence graph. A generation unit is configured to generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and to obtain a target decision based on the rearrangement results of the at least two candidate decisions. An update unit is used to update the evidence unit set according to the verification result when the verification result of the target decision does not meet the preset verification threshold, and to reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold.

[0018] In some embodiments of this disclosure, the processing unit includes: The processing module is used to collect the multi-source data and perform data cleaning, field mapping and disambiguation on the multi-source data to obtain preprocessed data; wherein, the multi-source data includes at least performance management counters, drive test logs, configuration data and alarm data; The first calculation module is used to perform multi-dimensional feature calculation on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage and knowledge base matching degree. The first construction module is used to generate corresponding evidence units for each preprocessed data based on the feature vector, and generate corresponding initial weights for each evidence unit, and construct the evidence unit set through all the evidence units.

[0019] In some embodiments of this disclosure, the building unit includes: The second construction module is used to construct the evidence diagram corresponding to each evidence node, using the evidence unit as the evidence node; The third construction module is used to construct the edge relationships between the evidence nodes based on a predefined rule base; wherein the edge relationships include at least one of support, conflict, reference, temporal association and same entity association; The second calculation module is used to calculate the consistency score of each evidence unit based on the edge relationship and the initial weight corresponding to the evidence node.

[0020] In some embodiments of this disclosure, the generation unit includes: A generation module is used to generate the at least two candidate decisions based on the evidence graph and the consistency score; A filtering module is used to filter out a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives; The rearrangement module is used to rearrange the candidate decisions in the non-dominated solution set based on preset business rules to obtain the target decision.

[0021] In some embodiments of this disclosure, the updating unit includes: An execution module is used to perform decision verification on the target decision and generate the verification result; wherein, the decision verification includes at least one of tool verification, rule verification, and cross-verification; An update module is used to quantify the verification result into a node-level feedback factor and update the weight of the corresponding evidence node in the evidence graph based on the node-level feedback factor. An addition module is used to generate a retrieval instruction based on the missing information identified in the verification result when the verification result does not meet the preset verification threshold, and to add the retrieved new evidence unit to the evidence unit set to update the evidence unit set.

[0022] In some embodiments of this disclosure, the apparatus further includes: The recording unit is used to record the entire process of standardizing the multi-source data until the target decision is obtained after the update unit reprocesses the multi-source data until the verification result of the target decision meets the preset verification threshold. A storage unit is used to save a corresponding parameter snapshot when calculating the consistency score and filtering the non-dominated solution set; The tracing unit is used to trace the reasoning path and evidence chain of the target decision based on the operation log and the parameter snapshot.

[0023] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the second aspect of the preceding embodiments.

[0024] According to a fifth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the second aspect of the preceding description.

[0025] According to a sixth aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described in the second aspect of the foregoing.

[0026] In summary, the data decision generation system, method, apparatus, and electronic device provided in this disclosure include: a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module. The processing module is used to standardize the acquired multi-source data to obtain a set of evidence units, and then send the evidence unit set to the evidence graph construction and calculation module. The evidence graph construction and calculation module is used to construct an evidence graph based on the received evidence unit set, calculate the consistency score of each evidence unit in the evidence graph, and send the evidence graph and the consistency scores of each evidence unit to the multi-objective decision-making module. The multi-objective decision-making module is used to receive the evidence graph and the consistency scores of each evidence unit, and, based on the evidence graph and... The consistency scores of each evidence unit generate at least two candidate decisions. Based on the rearrangement of these two candidate decisions, the target decision is obtained and sent to the verification module. The verification module receives the target decision and, if the verification result of the target decision does not meet the preset verification threshold, updates the evidence unit set based on the verification result and reprocesses the multi-source data until the verification result of the target decision meets the preset verification threshold. Through the coordinated operation of the processing module, the evidence graph construction and calculation module, the multi-target decision module, and the verification module, the standardized integration of multi-source data, the precise control of evidence consistency, and the verification-driven iterative optimization are achieved, thereby improving the credibility and practical applicability of the target decision.

[0027] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0028] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram of the structure of a data decision generation system provided in an embodiment of the present disclosure; Figure 2 A schematic diagram illustrating the dependencies between modules provided in an embodiment of this disclosure; Figure 3 This is a flowchart illustrating a processing module provided in an embodiment of the present disclosure; Figure 4 This is a flowchart illustrating an evidence graph construction and calculation module provided in an embodiment of this disclosure. Figure 5 A flowchart illustrating a multi-objective decision-making module provided in an embodiment of this disclosure; Figure 6 This is a flowchart illustrating a verification module provided in an embodiment of the present disclosure; Figure 7 This is a schematic diagram of the structure of another data decision generation system provided in an embodiment of the present disclosure; Figure 8 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure; Figure 9 A flowchart illustrating another data decision generation method provided in this disclosure embodiment; Figure 10 A flowchart illustrating another data decision generation method provided in this disclosure embodiment; Figure 11 A flowchart illustrating another data decision generation method provided in this disclosure embodiment; Figure 12 A flowchart illustrating another data decision generation method provided in this disclosure embodiment; Figure 13 A flowchart illustrating another data decision generation method provided in this disclosure embodiment; Figure 14 This is a schematic diagram of the structure of a data decision generation device provided in an embodiment of the present disclosure; Figure 15 This is a schematic diagram of another data decision generation device provided in an embodiment of the present disclosure; Figure 16 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0029] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0030] The following description, with reference to the accompanying drawings, describes a data decision generation system, method, apparatus, and electronic device according to embodiments of the present disclosure.

[0031] Figure 1 This is a schematic diagram of the structure of a data decision generation system provided in an embodiment of the present disclosure.

[0032] like Figure 1 As shown, the system includes: a processing module 11, an evidence graph construction and calculation module 12, a multi-objective decision-making module 13, and a verification module 14; wherein, The processing module 11 is used to standardize the acquired multi-source data to obtain a set of evidence units, and send the set of evidence units to the evidence graph construction and calculation module 12.

[0033] In some embodiments, multi-source data refers to a data set that comes from different acquisition channels, has different data formats, or different semantic rules; standardization processing refers to a series of operations performed on multi-source data, including unified format conversion, redundant information removal, data integrity repair, and semantic consistency calibration, with the aim of transforming heterogeneous data into structured data that can be processed uniformly; after standardization processing, an evidence unit is the smallest data unit that can independently express a single fact or information and can be verified individually, and multiple evidence units are combined according to preset rules to form an evidence unit set; after the processing module 11 acquires multi-source data through the data acquisition interface, it executes the standardization processing flow, encapsulates each processed structured data fragment into an evidence unit, summarizes all evidence units to form an evidence unit set, and sends it to the evidence graph construction and calculation module 12 through the inter-module data transmission interface.

[0034] The evidence graph construction and calculation module 12 is used to construct an evidence graph based on the received set of evidence units, calculate the consistency score of each evidence unit in the evidence graph, and send the evidence graph and the consistency scores of each evidence unit to the multi-objective decision module 13.

[0035] In some embodiments, an evidence graph refers to a graphical data structure formed by nodes based on evidence units and edges based on the relationships between evidence units, used to intuitively represent the relationship logic of multiple evidence units; a consistency score refers to a numerical indicator that quantifies the degree of conformity or contradiction between a single evidence unit and other related evidence units at the factual level, based on the relationships between evidence units in the evidence graph; after receiving the set of evidence units, the evidence graph construction and calculation module 12 first maps each evidence unit to an independent node in the evidence graph, establishes edge connections between corresponding nodes of different evidence units based on preset relationship identification rules, and completes the construction of the evidence graph; then, using a preset consistency calculation algorithm, combined with the node distribution and edge connection relationships of the evidence graph, it calculates the consistency score corresponding to each evidence unit to ensure that the consistency score can reflect the credibility association characteristics of the evidence unit; finally, the constructed evidence graph and the consistency scores of each evidence unit are synchronously sent to the multi-objective decision module 13.

[0036] The multi-objective decision module 13 is used to receive the evidence graph and the consistency scores of each evidence unit, generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, obtain the target decision according to the rearrangement result of the at least two candidate decisions, and send the target decision to the verification module 14.

[0037] In some embodiments, candidate decisions refer to multiple potential solutions or conclusions generated for specific business based on the correlation logic and consistency scores of the evidence graph; rearrangement refers to the process of prioritizing multiple candidate decisions according to preset decision evaluation rules, the ranking criteria including but not limited to the strength of evidence support corresponding to the candidate decisions, the aggregation results of consistency scores, etc. After receiving the evidence graph and consistency scores, the multi-objective decision module 13 extracts key information from the evidence graph and quantitative features of the consistency scores based on the preset decision generation logic, and combines them to form at least two candidate decisions; then it starts the rearrangement process, prioritizes and ranks all candidate decisions according to preset rules, selects the optimal candidate decision in the ranking results as the target decision, and sends the target decision to the verification module 14 through the data transmission channel.

[0038] The verification module 14 is used to receive the target decision, and if the verification result of the target decision does not meet the preset verification threshold, update the evidence unit set according to the verification result, and reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold.

[0039] In some embodiments, the verification result refers to the quantitative or qualitative conclusion obtained after evaluating the effectiveness, rationality, or accuracy of the target decision through a preset verification mechanism; the preset verification threshold refers to the benchmark value set in advance to determine whether the target decision meets the standard, and the preset verification threshold can be adjusted according to the needs of the business scenario; after receiving the target decision, the verification module 14 starts the preset verification process to evaluate the target decision, generates the corresponding verification result, and compares the verification result with the preset verification threshold; when the verification result meets the preset verification threshold, the target decision is confirmed to meet the standard and output; when the verification result does not meet the preset verification threshold, the reasons for the target decision not meeting the standard are analyzed based on the verification result, the evidence information that needs to be supplemented or corrected is identified, and the original evidence unit set is updated based on the supplemented or corrected evidence information (such as adding new evidence units, deleting invalid evidence units, or correcting the information of existing evidence units, etc.); the updated evidence unit set will trigger the processing module 11 to re-perform the standardization processing of the multi-source data (including the supplemented new data), and sequentially go through the recalculation of the evidence graph construction and calculation module 12 and the regeneration and rearrangement of the multi-target decision module 13 to form a cyclic processing flow until the verification result of the target decision meets the preset verification threshold.

[0040] Through the above structure, the system achieves systematic processing of multi-source data into target decision-making through the coordinated operation of four modules. It solves the problem of heterogeneous multi-source data through standardized processing, clarifies the correlation logic between evidence by building evidence graphs and calculating consistency, ensures the optimality of decision-making through candidate generation and rearrangement for multi-target decision-making, and improves the reliability of decision-making by relying on a verification-driven closed-loop mechanism.

[0041] In summary, the data-driven decision generation system disclosed herein includes: a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module. The processing module standardizes the acquired multi-source data to obtain a set of evidence units, and sends the evidence unit set to the evidence graph construction and calculation module. The evidence graph construction and calculation module constructs an evidence graph based on the received evidence unit set, calculates the consistency score of each evidence unit in the evidence graph, and sends the evidence graph and the consistency scores of each evidence unit to the multi-objective decision-making module. The multi-objective decision-making module receives the evidence graph and the consistency scores of each evidence unit, and, based on the evidence graph and the consistency scores of each evidence unit, performs a multi-objective decision-making process. The system calculates the consistency score of each element, generates at least two candidate decisions, and obtains the target decision based on the rearrangement of these two candidate decisions. The target decision is then sent to the verification module. The verification module receives the target decision and, if the verification result of the target decision does not meet a preset verification threshold, updates the evidence unit set based on the verification result and reprocesses the multi-source data until the verification result of the target decision meets the preset verification threshold. Through the collaborative linkage of the processing module, the evidence graph construction and calculation module, the multi-target decision module, and the verification module, the system achieves standardized integration of multi-source data, precise control of evidence consistency, and verification-driven iterative optimization, thereby improving the credibility and practical applicability of the target decision.

[0042] Figure 2 This is a schematic diagram illustrating the dependencies between modules provided in an embodiment of the present disclosure, such as... Figure 2 As shown, triggered by a user query / task, the processing module first acquires multi-source data, standardizes the multi-source data to obtain a set of evidence units, and sends the evidence unit set to the evidence graph construction and calculation module. After receiving the evidence unit set, the evidence graph construction and calculation module constructs an evidence graph and calculates the consistency score of each evidence unit in the graph. It then sends the evidence graph and the consistency scores of each evidence unit to the multi-objective decision-making module. The multi-objective decision-making module receives the evidence graph and the consistency scores of each evidence unit, generates at least two candidate decisions, and obtains the target decision based on the rearrangement results of the at least two candidate decisions. The target decision is sent to the verification module. After receiving the target decision, the verification module verifies the target decision to generate a verification result and compares the verification result with a preset verification threshold. If the verification result does not meet the preset verification threshold, the verification module updates the evidence unit set according to the verification result and feeds the updated evidence unit set back to the processing module, triggering the processing module to re-standardize the multi-source data. The evidence graph is then reconstructed and the consistency score is calculated by the evidence graph construction and calculation module. The multi-target decision module regenerates candidate decisions and completes the rearrangement, forming an iterative process until the verification result of the target decision meets the preset verification threshold.

[0043] Please continue reading. Figure 1The processing module 11 is also used for, The multi-source data is collected and processed by data cleaning, field mapping and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data and alarm data.

[0044] In some embodiments, the processing module 11 acquires multi-source data through a preset data acquisition interface. The multi-source data includes heterogeneous data such as structured key performance indicators (KPIs), logs, text work orders, maintenance records, configuration and topology data, etc. Then, a standardized processing flow is executed: the multi-source data is cleaned to remove missing values, outliers and duplicate data, and heterogeneous fields from different sources are uniformly mapped to a preset unified template (such as a schema) through field mapping. At the same time, entity disambiguation is performed to ensure that the same entity has a unique identifier in the dataset. This operation constitutes the data normalization process, and the preprocessed data is obtained after processing. All operations in this process are stored in the audit_trail log, recording the input data, processing rules, parameters and processing reasons to ensure auditability and controllability.

[0045] Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree.

[0046] In some embodiments, the processing module 11 performs multi-dimensional feature calculations on the preprocessed data to generate a feature vector {r,c,t,u,k,v}: where v is a placeholder for verification feedback, initially set to 0, and written back by the verification module 11; source reliability r is calculated using a function (such as the Source Reliability function), which comprehensively scores based on parameters such as the source type of the preprocessed data, the integrity of the collection link, the sample size, the data missing rate, and supplier information; time freshness t is calculated using a function (such as the Freshness Score function), which is based on the collection time window of the preprocessed data, combined with configuration parameters (such as Config.τ_by_source, which is the attenuation coefficient corresponding to different sources) and the maximum effective duration (such as Config.H_max); for content similarity c, topic coverage u, and knowledge base matching k, the processing module 11 adopts a large language model (such as Large Language Model). Model (LLM) is a strategy primarily based on enhancement and supplemented by structured computation. If the preprocessed data contains text content (such as work orders or maintenance records), LLM is invoked to perform information extraction and computation. The LLM prompts include three parts: System, Instructions, and Context. System defines it as information extraction and normalization in the wireless network optimization domain, requiring strict output of a single JSON format, prohibiting speculation or output of non-JSON text. Instructions specify that it must extract entities and relationships based on preset ontology and field definitions, calculate scores (where similarity corresponds to c, coverage to u, and kb_match to k), determine conflicts, and reference evidence spans; unknown fields are set to null. Context includes the current task topic, key fields, original text, and knowledge base templates. LLM output strictly follows a preset output schema, which includes fields such as scores, entities, relations, conflicts_detected, and source; all field calculations must reference specific evidence spans. If the preprocessed data is structured KPIs or logs, it is directly processed through functions (such as Structured...). The Sim function calculates content similarity c, the Coverage Score function calculates topic coverage u, and the KBMatch function calculates knowledge base matching k. All feature calculation processes are recorded in the audit_trail log, which records the input, the model used, the parameters, and the reason for the calculation.

[0047] Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

[0048] In some embodiments, the processing module 11 assigns a corresponding feature vector {r,c,t,u,k,v} to each preprocessed data, encapsulating it into an independent Evidence Unit (EU); and simultaneously based on the formula... Calculate the initial weight for each unit of evidence. Where a1, a2, a3, a4, and a5 are configurable or learnable coefficients; the processing module 11 summarizes all generated evidence units and constructs an evidence unit set. During this process, the generation information and initial weight calculation process of each evidence unit are stored in the audit_trail log to ensure the auditability and parameter controllability of the entire process.

[0049] To facilitate understanding, an example is given below: If the original data source is a text work order (document identifier: DOC123), and the task topic is "Analysis of Abnormal Uplink Received Signal Strength Indication in Cell_123", with key fields including "A3 Threshold" and "Switch Time Timer"; after the processing module 11 collects the work order data, it performs data cleaning, field mapping, and disambiguation to obtain preprocessed data; since this data contains text, a large language model is called to perform feature calculation. The context of the large language model is filled with the task topic, key fields, work order text (the text content in the work order with document identifier DOC123 stating "Cell_123's uplink received signal strength indication was higher than the threshold during busy hours from 18:00 to 19:00 on 2025-07-21, and the passive intermodulation test result was positive") and knowledge base template (such as "positive passive intermodulation to increased uplink received signal strength indication"); the large language model outputs JavaScript object representation content conforming to a preset pattern, where the similarity of the scoring items... The degree value is 0.85 (the reason cited is the content of "DOC123#L20-L25" where the uplink received signal strength indication is higher than the threshold), the coverage value is 0.5 (the required fields are ["A3 threshold", "switching timer"], the missing field is ["A3 threshold"]), the knowledge base matching degree value is 0.9 (the matching template is ["template_id_1"]), and the confidence value is 0.8; the entity includes an entity of type "cell" with the identifier "next-generation base station cell identifier-xxx", an entity of type "key performance indicator" with the name "uplink received signal strength indication", etc.; the relation includes a relation of type "reference support" with the reason "passive intermodulation positive to uplink received signal strength indication increase"; the processing module 11 obtains the feature vector based on this output (where the source reliability r=0.92, content similarity c=0.85, time freshness t=0.95, topic coverage u=0.5, knowledge base matching degree k=0.9, and verification feedback v=0), and then uses the formula =0.2×0.92+0.3×0.85+0.2×0.95+0.2×0.5+0.1×0.9 The result is obtained by calculation. =0.789, the feature vector and initial weights are encapsulated into an evidence unit and added to the evidence unit set. It should be noted that the above example is only for illustrative purposes and is not intended to limit the specific content.

[0050] Through the above structure, processing module 11 standardizes multi-source heterogeneous data into evidence units using a unified template schema, ensuring data consistency and comparability; it adopts a strategy of "LLM enhancement + structured computation" to generate features, balancing the semantic processing efficiency of text data with the computational accuracy of structured data; all processes retain audit_trail logs, achieving full-process auditability and parameter controllability; the generated evidence units and initial weights provide a consistent standard and stable convergence basis for the construction of evidence graphs and consistency calculations, avoiding calculation deviations caused by incomparable data.

[0051] Figure 3 This is a flowchart illustrating a processing module provided in an embodiment of the present disclosure, as shown below. Figure 3 As shown, this process is the entire workflow of multi-source evidence standardization and feature generation performed by the processing module: After acquiring multi-source data, the processing module first performs standardization processing (including data cleaning, field mapping, and disambiguation operations) to obtain preprocessed data; then, it completes the multi-dimensional feature calculation of source reliability (r) and time freshness (t) through rules or statistical methods; next, it determines whether the preprocessed data contains text: if it contains text, it calls the corresponding tool to process it to generate feature information of content similarity (c), topic coverage (u), and knowledge base matching degree (k) and records the log, while verifying whether the confidence of the processing result reaches the preset threshold. If it does, the c, u, and k features are retained; if it does not, the feature is marked as low confidence and recorded in the log; if it does not contain text, the c, u, and k features are directly calculated through structured methods and recorded in the log; then, normalization processing is performed on all features, the initial weight corresponding to each preprocessed data is calculated based on the normalized features, and each preprocessed data, corresponding feature, and initial weight are encapsulated into an evidence unit. All evidence units are summarized to construct an evidence unit set, and finally, the evidence unit set and the audit log of the entire process are output.

[0052] Please continue reading. Figure 1 The evidence graph construction and calculation module 12 is also used for, Using the evidence units as evidence nodes, construct the evidence diagram corresponding to each evidence node.

[0053] In some embodiments, after receiving the set of evidence units sent by the processing module 11, the evidence graph construction and calculation module 12 maps each evidence unit to an independent evidence node in the evidence graph. Each evidence node carries a corresponding feature vector {f_i} (including source reliability r, content similarity c, time freshness t, topic coverage u, knowledge base matching degree k, and verification feedback v), and uses the initial weight w_init calculated by the processing module 11 as the initial weight of the evidence node. Based on the evidence nodes, a node set V is constructed to form an evidence graph, which provides a foundation for edge relationship construction and consistency calculation.

[0054] Based on a predefined rule base, edge relationships are constructed between the evidence nodes; wherein, the edge relationships include at least one of support, conflict, reference, temporal association, and association with the same entity.

[0055] In some embodiments, the predefined rule base includes edge relationship determination rules and pre-defined conflict rules. Edge relationship determination rules are used to identify the association type between different evidence nodes. Edge relationships include, but are not limited to, support (support / same_fact), ref_support, conflict (conflict), time-close association (time_close), and same-entity association (same_entity). Pre-defined conflict rules are configurable rules, including but not limited to rule identifier, rule name, trigger conditions (indicators, symbols, thresholds), key performance indicators involved, time window requirements, exception conditions, weight priority, etc. Example rules include "uplink received signal strength indication increases and spectral / passive intermodulation test is negative" and "handover failure rate increases and A3 threshold / handover timer setting does not support neighbor cell measurement," etc. When the key slots or mechanisms of two evidence nodes meet any pre-defined conflict rule, they are determined to be conflicting edge relationships. The evidence graph construction and calculation module 12 traverses all evidence node pairs (i,j) and infers the edge relationship type between nodes through the edge relationship determination rules. If the inference result is a non-empty relationship, it is based on the formula

[0056] Calculate the edge weight W_ij; where base is the global baseline weight (can be 1.0 or set individually according to edge type), and s(type_ij) is the relation type coefficient (support = +1.0, reference support = +0.7, conflict = ...). (Time-series association = +0.3, Same entity association = +0.2, configurable) τ is the time decay function (τ is the time decay constant, consistent with the data freshness strategy). As the source reliability factor, The topic coverage factor is used; after calculating the edge weight W_ij, if its absolute value is less than or equal to the preset edge weight truncation threshold, then... Then, an edge of the corresponding type is established between evidence nodes i and j. The edge carries the relationship type and edge weight information. All edges constitute the edge set E of the evidence graph, thus completing the construction of the evidence graph (V,E).

[0057] The consistency score of each evidence unit is calculated based on the edge relationship and the initial weight corresponding to the evidence node.

[0058] In some embodiments, the consensus score calculation employs a multi-round propagation iteration approach, starting with initialization, where the initial weights of the evidence nodes are set. Let w_init be the initial weight output by processing module 11; then enter the iteration loop, with the upper limit of the iteration rounds set to K_max (if it can be set to more than 3 rounds). In each iteration, perform the following operations: First, degree normalization processing: for each evidence node i, calculate the sum of the edge weights of all its associated edges. ( To minimize the value and avoid a denominator of 0); the second step is to calculate the β coefficient for the current round using β annealing. (β is the initial β value, η is the annealing coefficient); the third step is neighbor aggregation calculation, based on the formula Calculate the next round consensus score for node i, where, The edge weights after degree normalization. Let α be the consistency score of evidence node i in round k+1, which is used to quantify the credibility of the node in the current evidence graph structure. Let α be the initial weight retention coefficient of the node. Let N(i) be the set of neighboring nodes of evidence node i, that is, all evidence nodes that are connected to i by an edge. Let j be the weight of neighbor node j in the kth round; the fourth step is to update the weights using the formula... Update the node weights, where γ is the damping coefficient, the smooth function uses the min-max or sigmoid method, and the clip function is used to truncate the weight values ​​within the [0,1] interval; the iteration terminates when any of the following conditions is met: the maximum change in weight of all evidence nodes. ,in, The preset weight change threshold; the average change in consistency across all evidence nodes. ,in, A preset consistency change threshold is set; optionally, the Top-N node ranking remains unchanged within k consecutive rounds; if a new evidence unit or edge relationship is added, only the k-hop neighborhood of the new node is subjected to K rounds of local iteration (K is much smaller than K_max), while the weight values ​​of the remaining nodes are maintained from the previous round. After the incremental update iteration converges, the final weight w_i and consistency score cons(i) of each evidence node are obtained; based on the consistency score of the evidence node, it can be calculated using the formula... The consistency score for candidate solutions can also include a conflict penalty term. (λ is the conflict penalty coefficient), Consistency(x) is the basic consistency score of candidate solution x, nodes(x) is the set of all evidence nodes involved in candidate solution x, and the node consistency scores are aggregated by summation; λ is the conflict penalty coefficient (default 0.8), which is used to adjust the degree of influence of conflict on the consistency score. The larger λ is, the more obvious the deduction of the score by the conflict edge; E_conflict(x) is the set of all conflict edges involved in candidate solution x; W_pq: the edge weight of the conflict edge (p,q), |W_pq| takes the absolute value to unify the direction of conflict penalty; w_p, w_q: the final weight values ​​of the nodes p and q at both ends of the conflict edge. The node weights are merged by mean, so that the conflict deduction between high weight nodes is more significant. This consistency score is used for downstream candidate decision reordering.

[0059] To facilitate understanding, the following example illustrates the situation: If there are three evidence units (EU1, EU2, EU3), their evidence node attributes are as follows: EU1 (PM displays an abnormally high uplink received signal strength indicator, r=0.95, u=0.8, ... =0.75); EU2 (spectral / passive intermodulation test positive, r=0.98, u=0.9, =0.78); EU3 (Physical cell identifier conflict, r=0.9, u=0.7, =0.65); The process of constructing edge relationships and calculating consistency scores in the evidence graph construction and calculation module 12 is as follows: First, infer the edge relationship type: EU1 and EU2 are support relationships, EU1 and EU3 are ref support relationships, EU2 and EU3 are time-close relationships, and the time difference Δt≈0 among the three; Second, calculate the edge weights: set base=1.0, τ=24h (Δt≈0, so g_time=1), s Given (support)=1.0, s(ref_support)=0.7, and s(time_close)=0.3, we calculate W_12=1.0×1.0×1×(0.95+0.98) / 2×(0.8+0.9) / 2≈0.82, W_13=1.0×0.7×1×(0.95+0.90) / 2×(0.8+0.7) / 2≈0.49, and W_23=1.0×0.3×1×(0.98+0.90) / 2×(0.9+0.7) / 2≈0.49. 0.7) / 2≈0.23; Finally, perform a round of consistency calculation: set α=0.6, β=0.4, γ=0.5, η=0, and calculate cons(1)=0.6×0.75+0.4×(0.82 / 0.82×0.78+0.49 / (0.82+0.49)×0.65)≈0.833, cons(2)=0.6×0.78+0.4×(0.82 / 0.82×0.75+0.23 / (0.82+0.23)×0.65)≈0 0.775, cons(3) = 0.6 × 0.65 + 0.4 × (0.49 / (0.82 + 0.49) × 0.75 + 0.23 / (0.23 + 0.49) × 0.78) ≈ 0.608; This result shows that the support relationship between EU1 and EU2 results in a high consistency score, mutually reinforcing the support for the hypothesis of "abnormal uplink received signal strength indication". EU3 provides secondary support through reference support and time series correlation, and its consistency score is relatively low. Candidate decision rearrangement can be based on this consistency score. It should be noted that the above examples are only illustrative and do not limit the specific content.

[0060] Through the above structure, the evidence graph construction and calculation module 12 organizes the evidence units into a multi-relationship evidence graph, explicitly expressing various relationships between evidence, thus solving the problem of difficulty in arbitrating contradictions in multi-source data; the multi-round consistency calculation based on the graph structure integrates intra-node information and edge information, providing an interpretable structured basis for cross-source fusion and rearrangement; the calculation process adopts stabilization methods such as degree normalization, damping, and annealing to ensure rapid convergence of iterations and controllable results, and the incremental update mechanism reduces online overhead; the entire process supports auditability, and each consistency score can be traced back to the specific evidence node and edge weight, providing a reliable quantitative basis for decision-making.

[0061] Figure 4 This is a flowchart illustrating an evidence graph construction and calculation module provided in an embodiment of this disclosure, as shown below. Figure 4 As shown, this process is the entire workflow of the evidence graph construction and calculation module in constructing the evidence graph and calculating the consistency score: After receiving the set of evidence units output by the processing module, the evidence graph construction and calculation module performs edge relationship reasoning on the evidence nodes in the set to determine the type of edge relationship between each evidence node (including support, reference support, conflict, temporal association, etc.); if a valid edge relationship can be inferred, the corresponding edge weight (W_ij) is calculated for different types of edge relationships, and the corresponding type of edge (such as support, reference support, conflict, temporal association) is established to complete the construction of the evidence graph; then, the consistency calculation is initialized, the initial weight of the evidence node is set to the initial weight (w_init) output by the processing module, and a multi-round propagation iteration process is entered (the upper limit of the number of iterations is K_max): in each iteration, degree normalization, damping, and annealing are performed in sequence, a single round of propagation calculation is performed to update the consistency score of the node, and then the calculation result is smoothed; then, it is determined whether the change in node weight or consistency score meets the preset iteration termination condition ( If the conditions are not met, the iteration continues; if the conditions are met, the iteration terminates. The final output is the constructed evidence graph, as well as the final weight and consistency score of each evidence unit.

[0062] Please continue reading. Figure 1 The multi-objective decision module 13 is also used for, Based on the evidence graph and the consistency score, the at least two candidate decisions are generated.

[0063] In some embodiments, candidate decisions refer to optional combinations of evidence subsets, reasoning branches, or execution plans. After receiving the evidence graph (including evidence nodes, edge relationships, and edge weights) and the consistency scores of each evidence unit, the multi-objective decision module 13 combines the association relationships of evidence nodes, the distribution of supporting / conflicting edges, and the consistency score aggregation features in the evidence graph with a preset decision generation logic to form at least two candidate decisions. During the generation process, it is ensured that different candidate decisions contain different combinations of evidence subsets or reasoning paths, providing sufficient optional space for multi-objective screening and rearrangement. Each candidate decision is associated with its set of evidence nodes, set of edge relationships, and corresponding initial feature information.

[0064] Select a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives.

[0065] In some embodiments, the screening process includes, but is not limited to, constraint filtering, multi-indicator calculation, normalization and direction unification, and dominance relationship determination. First, constraint filtering is performed based on the constraints set by the Service Level Agreement (SLA), including constraints such as budget limits, time limits, and minimum quality thresholds. Candidate decisions that violate these constraints (such as those exceeding the budget, exceeding the time limit, or having correctness or consistency below the minimum quality threshold) are eliminated, retaining the set of feasible candidate decisions. Second, a multi-indicator vector is calculated. For each feasible candidate decision x, an original indicator vector is calculated using the indicator function set F. Where, Correctness is the validator pass rate or factual consistency (determined by tool validation and cross-validation results), and Consistency is the aggregated consistency score of the evidence nodes involved in the candidate decision (such as the mean or quantiles). Nodes(x) is the set of evidence nodes involved in candidate decision x; Robustness is the score for consistency across time windows, consistency during busy hours and throughout the day, and stability against noisy / missing data; Latency is the end-to-end latency estimate (taking the mean); Cost is the estimate of call cost or engineering cost; then normalization and direction unification are performed, using the formula... Normalize each indicator (ε is the minimum value to avoid a denominator of 0; the clip function truncates the result to the [0,1] interval), and convert Latency and Cost to a higher value (conversion formula is...). , The remaining indicators (Correctness, Consistency, Robustness) are directly normalized to obtain a unified normalized indicator vector. Finally, dominance is determined: if the normalized indicator vector Vy corresponding to candidate decision y is greater than or equal to the normalized indicator vector Vx corresponding to candidate decision x in each dimension, and Vy has at least one dimension greater than Vx, then y is said to dominate x. All feasible candidate decisions are traversed, and dominated candidate decisions are eliminated. The remaining non-dominated candidate decisions constitute the non-dominated solution set (i.e., the Pareto front). The support gain and conflict penalty for each candidate decision are calculated. Where E_support(x) is the set of support edges involved in candidate decision x. To normalize the edge weights, w_p and w_q are the final weights of the two endpoints p and q of the supporting edge, used to quantify the strength of mutual support within the candidate decisions; conflict penalty. , where E_conflict(x) is the set of conflict edges involved in candidate decision x, used to quantify the strength of the internal conflict relationship of candidate decision.

[0066] The candidate decisions in the non-dominated solution set are rearranged based on preset business rules to obtain the target decision.

[0067] In some embodiments, the preset business rules include tie-breaking rules, diversity rules, and business preference rules. The re-ranking process adopts a heuristic selection strategy: first, the secondary ranking alternative score is calculated, based on the normalized index vector Vec_n(x), and then a comprehensive score formula is used. Calculate the score for each candidate decision (where b1 to b5 are configurable weights, with default values ​​of b1=0.4, b2=0.3, b3=0.1, b4=0.1, and b5=0.1; Latency_norm and Cost_norm are normalized values ​​without inversion; the Coverage metric derived from topic coverage u or missing field rate can be added if necessary); then calculate the reference weight w_next using the formula... The algorithm generates (sigma is the clip function or sigmoid function to ensure numerical stability, p1 to p8 are configurable coefficients with values ​​ranging from [0,2], r, c, t, u, k, v are the feature vector components of the evidence unit, S(x) is the support gain, and C(x) is the conflict penalty); finally, a heuristic selection is performed, breaking ties by sorting the scores in descending order, maximizing the diversity of candidate decisions by using (Kolmogorov-Smirnov, KS) distance or Euclidean distance, and adjusting the sorting order in combination with business preferences (such as prioritizing low cost or low latency), finally obtaining an ordered list of candidate decisions, and taking the candidate decision ranked first as the target decision.

[0068] To facilitate understanding, an example is given below: Suppose there are three candidate decisions x1, x2, and x3, and the screening and rearrangement process is as follows: Constraint filtering: Set SLA constraints as budget ≤ 1.0, time limit ≤ 0.80s, and minimum quality threshold (Correctness ≥ 0.84, Consistency ≥ 0.83). All three candidate decisions meet the budget and minimum quality requirements, but x3 has a Latency of 0.90s > 0.80s and is therefore eliminated. The feasible candidate decision set is {x1, x2}. Raw metric calculations: The raw metrics for x1 are Correctness=0.84, Consistency=0.84, Robustness=0.68, Latency=0.70s, and Cost=1.0; the raw metrics for x2 are Correctness=0.86, Consistency=0.90, Robustness=0.72, Latency=0.50s, and Cost=0.8. Normalization and Direction Unification: Based on the extrema of the feasible set (Correctness∈[0.84,0.86], Consistency∈[0.84,0.90], Robustness∈[0.68,0.72], Latency∈[0.50,0.70], Cost∈[0.8,1.0]), the normalized values ​​are calculated using the norm(m) formula, and are unified to be as large as possible: the normalized index vector of x1 is Correctness'=0.00, Consistency'=0.00, Robustness'=0.00, Latency'=0.50, Cost'=0.00; the normalized index vector of x2 is Correctness'=1.00, Consistency'=1.00, Robustness'=1.00, Latency'=1.00, Cost'=1.00; Domination relationship determination: The normalized index vector of x2 is ≥ x1 in each dimension, and at least one dimension needs to be greater than it. Therefore, x2 dominates x1, and the non-dominated solution set is {x2}. Rearrangement and target decision determination: If the non-dominated solution set contains only x2, x2 is directly used as the target decision; if the non-dominated solution set contains multiple candidate decisions (e.g., x2 and x3 are both non-dominated solutions when there is no SLA constraint), the score is calculated according to the comprehensive score formula (score of x2 = 0.4×1.00 + 0.3×1.00 + 0.1×1.00 - 0.1×0.00 - 0.1×0.3333 ≈ 0.767, score of x3 = 0.4×1.00 + 0.3×0.00 + 0.1×0.25 - 0.1×1.00 - 0.1×0.00 ≈ 0.325), sorted in descending order of score, and x2 is taken as the target decision.

[0069] It should be noted that the above is only an illustrative example and does not limit the specific content.

[0070] Through the above structure, the multi-objective decision-making module 13 generates multiple candidate decisions based on evidence graphs and consistency scores. It eliminates unsuitable candidates through constraint filtering and selects non-dominated efficient solutions using Pareto fronts, avoiding suboptimal choices caused by blind comparisons. It ensures the comparability of multi-objective indicators through normalization and direction unification, and quantifies the internal logical consistency of candidate decisions by combining support gain and conflict penalty. Based on heuristic rearrangement of preset business rules, it takes into account multi-objective trade-offs, business preferences, and decision diversity. The final target decision can adapt to the constraints and needs of different scenarios, improving the rationality of the decision and its adaptability to practical applications.

[0071] Figure 5 This is a flowchart illustrating a multi-objective decision-making module provided in an embodiment of the present disclosure, as shown below. Figure 5 As shown, this process is the entire workflow of candidate decision screening and rearrangement performed by the multi-objective decision-making module: The multi-objective decision-making module generates a set of candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and then performs constraint filtering (based on the constraints set by the service level agreement) to eliminate candidate decisions that violate the constraints; then, it performs multi-indicator calculations on the remaining feasible candidate decisions and completes the normalization and direction unification of the indicators; then, it performs a dominance relationship determination on the candidate decisions: if a candidate decision is dominated by other candidate decisions, it is eliminated; if a candidate decision is not dominated by other candidate decisions, it is included in the non-dominated solution set; finally, based on preset business rules, it performs heuristic selection and ranking of candidate decisions in the non-dominated solution set to obtain the target decision.

[0072] Please continue reading. Figure 1 The verification module 14 is also used for, The target decision is validated, and the validation result is generated; wherein the decision validation includes at least one of tool validation, rule validation, and cross-validation.

[0073] In some embodiments, decision verification employs a parallel execution approach using multiple types of validators, covering tool verification, rule verification, cross-validation, sampling, and probe / monitoring. All validators adhere to a unified standard interface (strict JSON format) for unified management by the orchestrator. Tool verification calls external systems or experimental tools (such as spectral / PIM testers, topology configuration query systems, counter query tools, etc.) to obtain measured data. Rule verification makes judgments based on a preset deterministic rule base (including thresholds, time series, and logical consistency criteria). Cross-validation supplements evidence through consistency comparisons across sources and indicators. Figure 1Consistency verification; random sampling for high-risk conclusions involves manual sampling or scoring; probes / monitoring collect real-time data through QoE user plane probes, proactive testing, etc.; the verifier's input includes trace ID, candidate ID, query requirements, evidence graph snapshot ID, a list of evidence nodes involved in the target decision, and constraints (such as time limits, budget); the output verification results include verification pass status, indicators (fact consistency, citation coverage, etc.), missing field list, evidence spans, remarks, verification cost, latency, confidence level, and severity level. For example, for the target decision "Cell_123 uplink RSSI anomaly caused by PIM interference", the tool verification calls the PIM tester to obtain test results, rule verification matches the knowledge base rule "PIM positive to ULRSSI increase", cross-verifies the time sequence consistency between the PM counter and the drive test data, and finally aggregates all verifier outputs to form a complete verification result. It should be noted that the above examples are only illustrative and do not limit the specific content.

[0074] The verification results are quantified into node-level feedback factors, and the weights of the corresponding evidence nodes in the evidence graph are updated based on the node-level feedback factors.

[0075] In some embodiments, the quantization and update process includes feedback aggregation, node attribution assignment, weight and consistency recalculation, etc. First, feedback aggregation is performed, mapping the outputs of each validator to candidate-level score vectors (including factual consistency, reference coverage, etc.) and a list of missing fields. Then, the process is further refined using formulas. Calculate the node-level feedback factor v_i (v_i ranges from [-1, 1]), where w_k is the validator weight, conf_k is the confidence level of the validator output, and sgn_k(i) ∈ {-1, 0, +1} represents the negative, no, or positive impact of the verification result on evidence node i. The clip function is used to truncate the result to the interval [-1, 1]. Next, perform node attribution allocation, using Shapley-lite or normalized weighting, to distribute the candidate-level verification impact to the corresponding evidence nodes according to the attribution weight A_i(x) (based on the evidence node importance, node weight w_i, and edge weight aggregation result). Finally, perform weight and consistency updates based on the formula... Update the weights of the evidence nodes, where w_prev is the node's weight before the update, r (source reliability), c (content similarity), t (time freshness), u (topic coverage), and k (knowledge base matching degree) are the node's original features, local_conflict is the weighted sum of the conflicting edges connected to the node (calculated according to the edge weight W_ij), and q1 to q7 are configurable coefficients (e.g., q1=0.5, q2=0.1, q3=0.1, q4=0.1, q5=0.1, q6=0.1, q7=0.3); and simultaneously based on the formula Perform a consistency recalculation (α=0.6, β=0.4, η=0.2, To ensure that the node consistency score matches the updated weight, the weight is adjusted accordingly.

[0076] When the verification result does not meet the preset verification threshold, a retrieval instruction is generated based on the missing information identified in the verification result, and the retrieved new evidence unit is added to the evidence unit set to update the evidence unit set.

[0077] In some embodiments, the preset verification thresholds include a verification failure rate threshold τ_fail (default 0.3 to 0.5) and a quality threshold τ_quality (correctness / consistency threshold, default 0.7 to 0.85). A targeted re-retrieval is initiated when the verification result meets any of the following triggering conditions: verification failure rate greater than τ_fail, correctness / consistency less than τ_quality, missing field list length |M(x)|>0, or high intensity of conflict rule hits. During the targeted re-retrieval process, a retrieval subtask is first generated item by item based on the missing field list of the verification result (e.g., missing "A3_threshold" field), clearly defining the retrieval target (e.g., "retrieving A3 threshold configuration data for Cell_123"). Then, tool routing is executed to select the data source with the lowest cost and shortest latency, while respecting budget and time constraints. Finally, the retrieved new evidence units (NewEUs) are added to the evidence unit set, with each new evidence unit carrying a trace_id and source metadata. For the updated evidence unit set, incremental graph building is performed: nodes are created for the new evidence units, local edges (k-hop neighborhoods, k defaults to 2) are constructed between them and the original evidence nodes, and local propagation is performed according to the consistency calculation logic in 5.2 to update the weights and consistency scores of the relevant nodes. At the same time, a snapshot of the evidence graph (snapshot_id) is retained, and it is committed only when the updated quality meets the standards; otherwise, it is rolled back to the previous stable state to ensure the stability and reliability of the evidence graph.

[0078] One feasible approach is to use a fixed, open-loop "retrieval-generation-verification" process, or to pre-determine a fixed number of iterations.

[0079] To facilitate understanding, an example is given below: Suppose the target decision is "the uplink RSSI anomaly during busy hours in Cell_123 on 2025-07-21 was caused by PIM interference", and the execution process of verification module 14 is as follows: Decision validation: The tool was activated for validation (calling the PIM tester), rule validation (matching the rule "PIM positive to ULRSSI increase"), and cross-validation (comparing the PM counter with the road test data). The validator output showed "passes=true, factual consistency=0.87, missing_fields=[A3_threshold], confidence level=0.82", and the validation failure rate=0.1<τ_fail=0.3, but key fields were missing. Feedback Quantification and Weight Update: Assuming the relevant evidence nodes are EU1 (RSSI abnormal) and EU2 (PIM positive), calculate the node-level feedback factors: v1=clip((0.80.821+0.90.851) / (0.80.82+0.90.85),-1,1)≈0.91, v2=clip((0.80.821+0.90.851) / (0.80.82+0.90.85),-1,1)≈0.91; Update node weights: for EU1, w_prev=0.7915, local_conflict=0, w_next=clip(0.7915+0.50.91+0.10.95+0.10.85+0.10.95+0.10.8+0.10.9-0.3*0,0,1)=0.9815; Targeted re-retrieval: Due to the missing "A3_threshold" field, a retrieval command was generated to "retrieve A3 threshold configuration data of Cell_123". The new evidence unit EU4 (A3_threshold=-90dBm, r=0.96, u=0.92, w_init=0.82) was retrieved. Incremental update: Create a node for EU4, construct its same entity association edge with EU1 and EU2 (s=0.2), perform one round of local propagation, update the consistency scores of relevant nodes, and complete the update of the evidence unit set and evidence graph.

[0080] It should be noted that the above examples are merely illustrative and do not limit the specific content.

[0081] Through the above structure, the verification module 14 ensures the correctness and reliability of the target decision through multiple types of verification, the node-level feedback factor realizes the accurate update of the evidence node weights, and the targeted re-retrieval and incremental mapping mechanism actively supplements the evidence when the decision does not meet the standards, forming a closed loop of "verification-feedback-update-supplementation" to achieve self-correction and gradual convergence; at the same time, the incremental update only performs a limited number of rounds of propagation on local nodes, taking into account both update efficiency and numerical stability, and avoiding the resource consumption and increased latency caused by full graph recalculation.

[0082] Figure 6 This is a flowchart illustrating a verification module provided in an embodiment of the present disclosure, as shown below. Figure 6 As shown, this process is the entire workflow of the verification module executing the verification-driven closed loop and re-retrieval: After receiving the target decision output by the multi-objective decision-making module, the verification module performs decision verification on the target decision (including tool verification, rule verification, cross-verification, sampling, etc.) and aggregates the verification results to generate verification results; then, the verification results are quantified into node-level feedback factors, and the weights and consistency scores of the corresponding evidence nodes in the evidence graph are updated based on the feedback factors; then, it is determined whether the verification results meet the preset verification threshold: if not, a retrieval instruction is generated based on the missing information identified in the verification results, new evidence units are obtained through targeted re-retrieval and added to the evidence unit set, the evidence graph is incrementally updated and local consistency is recalculated, and then the multi-objective decision-making module is returned to regenerate and rearrange the candidate decisions; if compliant, the target decision is output, and the audit module records the entire operation log.

[0083] Figure 7 A schematic diagram of the structure of a data decision generation system provided in an embodiment of this disclosure is further shown.

[0084] like Figure 7 As shown, the system also includes an audit module 15, wherein the audit module 15 is used for, Record the entire process of standardizing the multi-source data to obtaining the target decision.

[0085] In some embodiments, the operation log covers the entire chain of data processing, evidence graph construction, multi-objective decision-making, and verification closure, including but not limited to the source information of multi-source data (data source identifier, collection time, format type), standardization processing records (execution steps and results of data cleaning, field mapping, and disambiguation processing), evidence unit generation information (calculation process and values ​​of feature vectors and initial weights w_init), evidence graph construction records (calculation basis of node creation, edge relationship type, and edge weights W_ij), the iterative process of consistency calculation (node ​​weights and consistency scores in each iteration), candidate decision generation and screening records (constraint filtering results, multi-indicator calculation values, normalization results, and dominance relationship determination process), verification results and feedback update records (verifier output, node-level feedback factor v_i, and values ​​before and after weight updates), and re-retrieval and incremental update records (retrieval instructions, new evidence unit information, and local propagation results of incremental graph construction). The operation log is recorded in a unified format, including timestamps, operation module identifiers, operator / system identifiers, operation content, and associated data identifiers (such as evidence unit ID, candidate decision ID, and evidence graph snapshot ID), ensuring that each step of the operation is traceable and verifiable.

[0086] When calculating the consistency score and filtering the non-dominated solution set, save the corresponding parameter snapshot.

[0087] In some embodiments, the parameter snapshot includes two categories: consistency calculation parameters and non-dominated solution set filtering parameters. The consistency calculation parameters cover iterative parameters such as α (0.6), β (0.4), η (0.2), γ (0.5), K_max (3 to 5), δ_w (1e-3), and δ_c (1e-3), edge weight calculation parameters such as s(type) (relationship type coefficient), τ (time decay constant, 24h), and base (global baseline weight), as well as the configuration of processing logic such as degree normalization, smoothing, and truncation. The non-dominated solution set filtering parameters... The solution set selection parameters cover SLA constraints (budget cap, time limit, minimum quality threshold), multi-indicator calculation methods, min / max values ​​in the normalization formula, dominance relationship determination rules, and weights b1 to b5 in the comprehensive score formula (default b1=0.4, b2=0.3, b3=0.1, b4=0.1, b5=0.1). Parameter snapshots are bound to the corresponding stage's evidence graph snapshots and candidate decision sets, recording the default values ​​or user-covered values ​​of the parameters to ensure consistent calculations and the reproducibility of the non-dominated solution set selection process.

[0088] Based on the operation log and the parameter snapshot, the reasoning path and evidence chain of the target decision are traced.

[0089] In some embodiments, the tracing process is achieved by associating operation logs and parameter snapshots: reasoning path tracing can restore the generation logic of the target decision, including but not limited to the generation basis of candidate decisions, the judgment results of constraint filtering, the calculation process of multi-indicator normalization and dominance relationship judgment, and the rule application order of heuristic rearrangement; evidence chain tracing can locate all evidence units on which the target decision depends, including but not limited to the source data, feature vectors and initial weights of each evidence unit, the edge relationship type and edge weights between evidence units, the iterative calculation process of consistency scores, and the feedback evaluation of each evidence unit in the verification stage; at the same time, the audit module 15 supports querying operation logs and parameter snapshots by conditions such as target decision ID, evidence unit ID, and time range, generating tracing reports, clarifying the evidence support, calculation parameters and execution logic corresponding to each reasoning step of the target decision, and realizing full-link tracing of "decision-evidence-parameter-operation".

[0090] To facilitate understanding, the following example illustrates the process: Regarding the target decision of "Uplink RSSI Anomaly Analysis for Cell_123", the tracing process of audit module 15 is as follows: Operation log query: Querying by target decision ID, the multi-source data includes DOC123 text work order and PIM test data. Standardization processing performed field mapping (mapping "ULRSSI" to "Uplink Received Signal Strength Indication"). The initial weight of evidence unit EU1, w_init=0.789 (calculated as 0.2×0.92+0.3×0.85+0.2×0.95+0.2×0.5+0.1×0.9). Parameter snapshot retrieval: The consistency calculation parameters are α=0.6, β=0.4, K_max=3, and the SLA constraints for screening the non-dominated solution set are budget ≤1.0 and time limit ≤0.8s; Reasoning path tracing: Reconstruct the screening process of candidate decision x2. The correctness of x2 is 0.86 and the consistency is 0.90. After normalization, the vector is better than x1 in each dimension. Therefore, it is judged as a non-dominated solution. According to the comprehensive score formula, the score is 0.767. The first one in the ranking becomes the target decision. Evidence chain tracing: The target decision depends on EU1 (RSSI abnormal) and EU2 (PIM positive). The edge relationship between the two is a support relationship, with edge weight W_12=0.82, consistency scores cons(1)=0.833 and cons(2)=0.775, and verification stage feedback factors v1=0.91 and v2=0.91. After the weight update, they are 0.9815 and 0.978 respectively. There are no conflicting edges and missing field supplementary verification records.

[0091] It should be noted that the above examples are merely illustrative and do not limit the specific content.

[0092] Through the above structure, the audit module 15 achieves complete traceability of the target decision-making reasoning path and evidence chain through full-process operation logs and key parameter snapshots, ensuring that the decision-making process is auditable and reproducible; at the same time, it provides a reliable basis for compliance verification and problem investigation of decisions. When a decision deviates, it can quickly locate whether the problem is in data processing, parameter configuration, rule application or verification, thereby improving the reliability and operational efficiency of the system.

[0093] Corresponding to the data decision generation system described above, this invention also proposes a data decision generation method. Since the method embodiments of this invention correspond to the system embodiments described above, details not disclosed in the method embodiments can be referred to in the system embodiments described above, and will not be repeated here.

[0094] Figure 8 This is a flowchart illustrating a data decision generation method provided in an embodiment of the present disclosure.

[0095] like Figure 8 As shown, the method includes steps 101-104.

[0096] Step 101: Standardize the acquired multi-source data to obtain a set of evidence units.

[0097] In some embodiments, multi-source data refers to a data set that comes from different collection channels and has different formats and semantics, including at least performance management data, drive test logs, configuration data, alarm data, etc.; the standardization process is as follows: perform data cleaning on the multi-source data to remove missing values, outliers and duplicate data, map heterogeneous fields to a preset unified data model through field mapping, perform entity disambiguation to ensure that the identifier of the same entity is unique, and complete the normalization of time and unit. The above operations are used to obtain preprocessed data; then generate a corresponding evidence unit (i.e., the smallest data unit that can independently express a single fact and can be verified separately) for each preprocessed data, and combine all evidence units according to preset rules to construct an evidence unit set.

[0098] Step 102: Construct an evidence graph based on the set of evidence units, and calculate the consistency score of each evidence unit in the evidence graph.

[0099] In some embodiments, when constructing the evidence graph, each evidence unit is mapped to an independent node in the evidence graph. The relationships between evidence units are identified based on predefined association rules, and corresponding types of edges are established (edge ​​relationships include, but are not limited to, at least one of support, conflict, reference, temporal association, and same entity association). The edges carry relationship type and edge weight information, and the nodes and edges together constitute the evidence graph. When calculating the consistency score, the initial weight of the evidence unit is first set to the initial weight. Then, multiple rounds of propagation iterations are performed: in each round, the edge weights of the nodes are normalized, an annealing coefficient is introduced to adjust the contribution intensity of neighbor aggregation, the consistency score is calculated based on the initial weight of the node and the weighted information of the neighbor nodes, and then the node weight is updated through a damping coefficient and a truncation function. When the change in node weight or the change in consistency score meets a preset threshold, or the number of iterations reaches the upper limit, the process terminates, and the consistency score corresponding to each evidence unit is obtained.

[0100] Step 103: Based on the evidence graph and the consistency scores of each evidence unit, generate at least two candidate decisions, and obtain the target decision based on the rearrangement results of the at least two candidate decisions.

[0101] In some embodiments, when generating candidate decisions, at least two candidate decisions for the business problem are formed by combining the node relationships, edge weight distribution, and consistency scores of the evidence graph with a preset decision generation logic. The reordering process is as follows: First, candidate decisions are filtered according to the constraints of the service level agreement (such as budget, time limit, minimum quality threshold) to retain the feasible set; then, multi-dimensional indicators (including correctness, consistency, robustness, latency, and cost) are calculated for the feasible candidate decisions, and the indicators are normalized and unified to the direction of "the larger the better"; dominated candidate decisions are eliminated by determining the dominance relationship to obtain the non-dominated solution set; finally, the non-dominated solution set is sorted according to preset business rules (such as business preferences and diversity requirements), and the candidate decision ranked first is taken as the target decision.

[0102] Step 104: If the verification result of the target decision does not meet the preset verification threshold, update the evidence unit set according to the verification result and reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold.

[0103] In some embodiments, when verifying the target decision, at least one decision verification operation, such as tool verification, rule verification, or cross-validation, is performed to generate a verification result containing the verification result, missing fields, and confidence level. This result is then compared with a preset verification threshold (such as verification failure rate or quality threshold). If the verification result does not meet the threshold, the verification result is quantified as a node-level feedback factor, and the weight of the corresponding evidence unit is updated. Simultaneously, a retrieval instruction is generated based on the missing information identified in the verification result. New evidence units are obtained through targeted re-retrieval, and the new evidence units are added to the evidence unit set to complete the update. Subsequently, based on the updated evidence unit set, all processes, such as constructing the evidence graph, are re-executed until the verification result of the target decision meets the preset verification threshold.

[0104] The above methods can achieve standardized integration of multi-source heterogeneous data, clarify the logical relationships between evidence with the help of evidence graphs, select the optimal decision through multi-objective rearrangement, and optimize the decision quality by relying on the verification-driven closed-loop mechanism. This effectively solves the problems of difficult integration of multi-source data, unexplainable decision-making process, and insufficient reliability of results, and improves the logic, controllability and engineering implementation capability of data-driven decision-making.

[0105] In summary, the data-driven decision generation method disclosed herein includes: standardizing the acquired multi-source data to obtain a set of evidence units; constructing an evidence graph based on the evidence unit set and calculating the consistency score of each evidence unit in the evidence graph; generating at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and obtaining a target decision based on the rearrangement results of the at least two candidate decisions; updating the evidence unit set based on the verification results and reprocessing the multi-source data until the verification results of the target decision meet the preset verification threshold when the verification results of the target decision do not meet the preset verification threshold through the coordinated linkage of the processing module, the evidence graph construction and calculation module, the multi-target decision module, and the verification module, thereby improving the credibility and practical applicability of the target decision.

[0106] Figure 9 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 9 As shown, the method includes the following steps: Step 201: Collect the multi-source data and perform data cleaning, field mapping and disambiguation on the multi-source data to obtain preprocessed data; wherein, the multi-source data includes at least performance management counters, drive test logs, configuration data and alarm data.

[0107] Step 202: Perform multi-dimensional feature calculation on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage and knowledge base matching degree.

[0108] Step 203: Generate corresponding evidence units for each preprocessed data based on the feature vector, and generate corresponding initial weights for each evidence unit. Construct the evidence unit set through all the evidence units.

[0109] Figure 10 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 10 As shown, the method includes the following steps: Step 301: Using the evidence unit as evidence node, construct the evidence diagram corresponding to each evidence node.

[0110] Step 302: Based on a predefined rule base, construct edge relationships between the evidence nodes; wherein the edge relationships include at least one of support, conflict, reference, temporal association, and association with the same entity.

[0111] Step 303: Calculate the consistency score of each evidence unit based on the edge relationship and the initial weight corresponding to the evidence node.

[0112] Figure 11 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 11 As shown, the method includes the following steps: Step 401: Based on the evidence graph and the consistency score, generate the at least two candidate decisions.

[0113] Step 402: Select a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives.

[0114] Step 403: Rearrange the candidate decisions in the non-dominated solution set based on preset business rules to obtain the target decision.

[0115] Figure 12 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 12 As shown, the method includes the following steps: Step 501: Perform decision verification on the target decision and generate the verification result; wherein the decision verification includes at least one of tool verification, rule verification and cross-verification.

[0116] Step 502: Quantify the verification result into a node-level feedback factor, and update the weight of the corresponding evidence node in the evidence graph based on the node-level feedback factor.

[0117] Step 503: When the verification result does not meet the preset verification threshold, a retrieval instruction is generated based on the missing information identified in the verification result, and the retrieved new evidence unit is added to the evidence unit set to update the evidence unit set.

[0118] Figure 13 A flowchart illustrating a data decision generation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 13 As shown, the method includes the following steps: Step 601: Record the entire process of standardizing the multi-source data and obtaining the target decision.

[0119] Step 602: When calculating the consistency score and filtering the non-dominated solution set, save the corresponding parameter snapshot.

[0120] Step 603: Based on the operation log and the parameter snapshot, trace the reasoning path and evidence chain of the target decision.

[0121] Corresponding to the data decision generation method described above, this invention also proposes a data decision generation apparatus. Since the apparatus embodiments of this invention correspond to the method embodiments described above, details not disclosed in the apparatus embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0122] Figure 14 This is a schematic diagram of the structure of a data decision generation device provided in an embodiment of the present disclosure, as shown below. Figure 14 As shown, it includes: a processing unit 51, a construction unit 52, a generation unit 53, and an update unit 54.

[0123] Processing unit 51 is used to standardize the acquired multi-source data to obtain a set of evidence units; Construction unit 52 is used to construct an evidence graph based on the evidence unit set and calculate the consistency score of each evidence unit in the evidence graph; The generation unit 53 is used to generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and to obtain the target decision based on the rearrangement results of the at least two candidate decisions; The update unit 54 is used to update the evidence unit set according to the verification result when the verification result of the target decision does not meet the preset verification threshold, and to reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold.

[0124] In summary, the data decision generation device provided in this disclosure includes: standardizing acquired multi-source data to obtain a set of evidence units; constructing an evidence graph based on the evidence unit set and calculating the consistency score of each evidence unit in the evidence graph; generating at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and obtaining a target decision based on the rearrangement results of the at least two candidate decisions; updating the evidence unit set based on the verification results and reprocessing the multi-source data if the verification result of the target decision does not meet the preset verification threshold, until the verification result of the target decision meets the preset verification threshold; through the coordinated linkage of the processing module, the evidence graph construction and calculation module, the multi-target decision module, and the verification module, the device achieves standardized integration of multi-source data, precise control of evidence consistency, and verification-driven iterative optimization, thereby improving the credibility and practical applicability of the target decision.

[0125] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 15 As shown, the processing unit 51 includes: The processing module 511 is used to collect the multi-source data and perform data cleaning, field mapping and disambiguation processing on the multi-source data to obtain preprocessed data; wherein, the multi-source data includes at least performance management counters, drive test logs, configuration data and alarm data; The first calculation module 512 is used to perform multi-dimensional feature calculation on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage and knowledge base matching degree. The first construction module 513 is used to generate corresponding evidence units for each preprocessed data based on the feature vector, and generate corresponding initial weights for each evidence unit, and construct the evidence unit set through all the evidence units.

[0126] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 15 As shown, the building unit 52 includes: The second construction module 521 is used to construct the evidence diagram corresponding to each evidence node, with the evidence unit as the evidence node; The third construction module 522 is used to construct edge relationships between the evidence nodes based on a predefined rule base; wherein the edge relationships include at least one of support, conflict, reference, temporal association and same entity association; The second calculation module 523 is used to calculate the consistency score of each evidence unit based on the edge relationship and the initial weight corresponding to the evidence node.

[0127] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 15 As shown, the generation unit 53 includes: The generation module 531 is used to generate the at least two candidate decisions based on the evidence graph and the consistency score; The filtering module 532 is used to filter out a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives; The rearrangement module 533 is used to rearrange the candidate decisions in the non-dominated solution set based on preset business rules to obtain the target decision.

[0128] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 15 As shown, the update unit 54 includes: The execution module 541 is used to perform decision verification on the target decision and generate the verification result; wherein, the decision verification includes at least one of tool verification, rule verification and cross-verification; The update module 542 is used to quantify the verification result into a node-level feedback factor and update the weight of the corresponding evidence node in the evidence graph based on the node-level feedback factor. The addition module 543 is used to generate a retrieval instruction based on the missing information identified in the verification result when the verification result does not meet the preset verification threshold, and to add the retrieved new evidence unit to the evidence unit set to update the evidence unit set.

[0129] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 15 As shown, the device further includes: The recording unit 55 is used to record the entire process of standardizing the multi-source data until the target decision is obtained after the update unit 54 reprocesses the multi-source data until the verification result of the target decision meets the preset verification threshold. The storage unit 56 is used to save the corresponding parameter snapshot when calculating the consistency score and filtering the non-dominated solution set; The tracing unit 57 is used to trace the reasoning path and evidence chain of the target decision based on the operation log and the parameter snapshot.

[0130] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0131] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0132] Figure 16 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0133] like Figure 16 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 602 or loaded from storage unit 608 into RAM (Random Access Memory) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An I / O (Input / Output) interface 605 is also connected to the bus 604.

[0134] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0135] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as data decision generation methods. For example, in some embodiments, the data decision generation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform the aforementioned data decision generation method by any other suitable means (e.g., by means of firmware).

[0136] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0137] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0139] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0140] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0141] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0142] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0143] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0144] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A data-driven decision generation system, characterized in that, The system includes: a processing module, an evidence graph construction and calculation module, a multi-objective decision-making module, and a verification module; wherein... The processing module is used to standardize the acquired multi-source data to obtain a set of evidence units, and send the set of evidence units to the evidence graph construction and calculation module. The evidence graph construction and calculation module is used to construct an evidence graph based on the received set of evidence units, calculate the consistency score of each evidence unit in the evidence graph, and send the evidence graph and the consistency score of each evidence unit to the multi-objective decision module. The multi-objective decision-making module is used to receive the evidence graph and the consistency scores of each evidence unit, generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, obtain the target decision according to the rearrangement result of the at least two candidate decisions, and send the target decision to the verification module. The verification module is used to receive the target decision, and if the verification result of the target decision does not meet the preset verification threshold, update the evidence unit set according to the verification result, and reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold. The processing module is further used for, The multi-source data is collected and subjected to data cleaning, field mapping, and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data, and alarm data; Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree; Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

2. The system according to claim 1, characterized in that, The evidence graph construction and calculation module is also used for, Using the evidence units as evidence nodes, construct the evidence diagram corresponding to each evidence node; Based on a predefined rule base, edge relationships are constructed between the evidence nodes; wherein, the edge relationships include at least one of support, conflict, reference, temporal association, and association with the same entity; The consistency score of each evidence unit is calculated based on the edge relationship and the initial weight corresponding to the evidence node.

3. The system according to claim 1, characterized in that, The multi-objective decision-making module is also used for, Based on the evidence graph and the consistency score, the at least two candidate decisions are generated; Select a non-dominated solution set from the at least two candidate decisions; wherein the non-dominated solution set represents the optimal trade-off relationship when there is no absolutely optimal solution for the at least two candidate decisions under all optimization objectives; The candidate decisions in the non-dominated solution set are rearranged based on preset business rules to obtain the target decision.

4. The system according to claim 1, characterized in that, The verification module is also used for, Perform decision verification on the target decision and generate the verification result; wherein, the decision verification includes at least one of tool verification, rule verification and cross-verification; The verification results are quantified into node-level feedback factors, and the weights of the corresponding evidence nodes in the evidence graph are updated based on the node-level feedback factors. When the verification result does not meet the preset verification threshold, a retrieval instruction is generated based on the missing information identified in the verification result, and the retrieved new evidence unit is added to the evidence unit set to update the evidence unit set.

5. The system according to claim 1, characterized in that, The system also includes an audit module, wherein the audit module is used for, Record the entire process of standardizing the multi-source data to obtaining the target decision; When calculating the consistency score and filtering the non-dominated solution set, save the corresponding parameter snapshot; Based on the operation log and the parameter snapshot, the reasoning path and evidence chain of the target decision are traced.

6. A data-driven decision generation method, characterized in that, The method includes: The acquired multi-source data is standardized to obtain a set of evidence units; An evidence graph is constructed based on the set of evidence units, and the consistency score of each evidence unit in the evidence graph is calculated. Based on the evidence graph and the consistency scores of each evidence unit, at least two candidate decisions are generated, and the target decision is obtained based on the rearrangement results of the at least two candidate decisions. If the verification result of the target decision does not meet the preset verification threshold, the evidence unit set is updated according to the verification result, and the multi-source data is reprocessed until the verification result of the target decision meets the preset verification threshold. The standardization process for the acquired multi-source data to obtain a set of evidence units includes: The multi-source data is collected and subjected to data cleaning, field mapping, and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data, and alarm data; Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree; Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

7. A data decision generation device, characterized in that, The device includes: The processing unit is used to standardize the acquired multi-source data to obtain a set of evidence units; A construction unit is used to construct an evidence graph based on the set of evidence units and to calculate the consistency score of each evidence unit in the evidence graph; A generation unit is configured to generate at least two candidate decisions based on the evidence graph and the consistency scores of each evidence unit, and to obtain a target decision based on the rearrangement results of the at least two candidate decisions. An update unit is used to update the evidence unit set according to the verification result when the verification result of the target decision does not meet the preset verification threshold, and to reprocess the multi-source data until the verification result of the target decision meets the preset verification threshold. The processing unit is further used for, The multi-source data is collected and subjected to data cleaning, field mapping, and disambiguation to obtain preprocessed data; wherein the multi-source data includes at least performance management counters, drive test logs, configuration data, and alarm data; Multidimensional feature calculations are performed on the preprocessed data to obtain a feature vector; wherein the feature vector includes at least source reliability, content similarity, time freshness, topic coverage, and knowledge base matching degree; Based on the feature vector, a corresponding evidence unit is generated for each preprocessed data, and a corresponding initial weight is generated for each evidence unit. The evidence unit set is constructed through all the evidence units.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 6.

Citation Information

Patent Citations

  • Decision-making result determination method, equipment and device and readable storage medium

    CN109585014A

  • Storage information verification system based on consensus mechanism

    CN120602228A