An open source intelligence evaluation method and device based on context information

Abstract

This disclosure belongs to the field of open-source intelligence assessment technology. This disclosure provides an open-source intelligence assessment method and apparatus based on context information. The method includes: in response to an open-source intelligence assessment task, determining at least one cognitive dimension; obtaining context information corresponding to the intelligence; under the constraints of the context information, retrieving corresponding entity information and its relationship information from an intelligence knowledge graph according to each cognitive dimension; obtaining a first dimension score for each cognitive dimension based on the entity information and its relationship information, and integrating the score and the context information to generate an assessment conclusion.

Description

Technical Field

[0001] This disclosure belongs to the field of open source intelligence assessment technology, and in particular relates to an open source intelligence assessment method and apparatus based on context information. Background Technology

[0002] With the diversification and increasing complexity of modern financial intelligence sources, traditional intelligence assessment methods generally neglect the crucial role of "context" in intelligence assessment when faced with multi-dimensional, multi-contextual, and multi-task requirements. This leads to significant biases in assessment results regarding timeliness, regional applicability, and task relevance. For example, the same intelligence regarding the approval and launch of a cross-border payment platform may have drastically different tactical value, strategic impact, and technological maturity in the context of emerging market expansion versus tightening regulations. Furthermore, traditional systems cannot dynamically adjust their assessment logic based on time, geography, and task context, resulting in assessment results that are detached from actual business needs.

[0003] Furthermore, existing methods suffer from structural deficiencies in the division of cognitive dimensions, often assessing intelligence value from a single-dimensional perspective, resulting in one-sided assessments that are difficult to support high-level command and decision-making. Summary of the Invention

[0004] This disclosure provides an open-source intelligence assessment method and apparatus based on contextual information, which can effectively solve the above-mentioned problems.

[0005] This disclosure is implemented as follows: Firstly, an open-source intelligence assessment method based on contextual information, the method comprising: In response to an open-source intelligence assessment task, at least one cognitive dimension is identified, which includes at least one of a technical dimension, a tactical dimension, and a strategic dimension. Obtain context information corresponding to the intelligence, including time context, geographic context, and task context; Under the constraints of the aforementioned contextual information, the corresponding entity information and its relational information are retrieved from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relation edges as the corresponding entity information and relation information; Based on the entity information and its relationship information of each cognitive dimension, a first dimension score for that cognitive dimension is obtained, and the score and the context information are integrated to generate an evaluation conclusion, wherein the score includes the first dimension scores of all cognitive dimensions.

[0006] Secondly, an open-source intelligence assessment device based on contextual information, the device comprising: The dimension acquisition module is used to determine at least one cognitive dimension in response to an open-source intelligence assessment task, wherein the cognitive dimension includes at least one of a technical dimension, a tactical dimension, and a strategic dimension; The context acquisition module is used to acquire context information corresponding to the intelligence, including time context, geographic context, and task context. The knowledge acquisition module is used to retrieve corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, under the constraints of the context information, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relation edges as the corresponding entity information and relation information; An evaluation module is used to obtain a first dimension score for each cognitive dimension based on the entity information and its relationship information, and to integrate the score and the context information to generate an evaluation conclusion, wherein the score includes the first dimension scores for all cognitive dimensions.

[0007] Thirdly, this disclosure provides an electronic device, including: Memory, the memory storing execution instructions; and A processor that executes execution instructions stored in the memory, causing the processor to perform the method described in the first aspect.

[0008] Fourthly, this disclosure provides a readable storage medium storing executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect.

[0009] Compared with the prior art, the beneficial effects of this disclosure are: This disclosure provides an open-source intelligence multi-agent collaborative evaluation method and apparatus based on contextual information. By combining cognitive dimensions and contextual information, it accurately extracts relevant entities and relationships from an intelligence knowledge graph, providing a systematic, interpretable, and traceable solution for open-source intelligence evaluation. Through deep coupling of a three-dimensional cognitive framework of "technology-application-strategy" with contextual information of "time-geography-task," it achieves second-level task-driven pruning in the knowledge graph, quickly locating the most relevant subset of entities and relationships. Compared with existing technologies, this method significantly improves the accuracy, timeliness, scenario adaptability, and decision support capabilities of the evaluation. Especially in complex and ever-changing geopolitical and financial intelligence, competitive intelligence, and risk warning scenarios, through context-driven dynamic knowledge pruning, the method can generate evaluation conclusions with higher confidence and more complete evidence chains across cognitive dimensions in real time. Attached Figure Description

[0010] Figure 1 This is a flowchart of the open-source intelligence assessment method S100 based on context information provided in this embodiment of the disclosure.

[0011] Figure 2 This is a flowchart of S200, an open-source intelligence multi-agent collaborative evaluation method based on context information provided in this embodiment.

[0012] Figure 3 This is a schematic diagram of the structure of the open-source intelligence assessment device 1000 based on context information provided in this embodiment.

[0013] Figure 4 This is a schematic diagram of the structure of the open-source intelligence multi-agent collaborative evaluation device 2000 based on context information provided in this embodiment. Detailed Implementation

[0014] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0015] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] Unless otherwise stated, the exemplary implementations / embodiments shown are to be understood as providing exemplary features of various details that provide ways in which the technical concepts of this disclosure can be implemented in practice. Therefore, unless otherwise stated, the features of various implementations / embodiments may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concepts of this disclosure.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure. The singular forms “a,” “the,” and “the” used in the embodiments of this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0018] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0019] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0020] The terms "first" and "second" used herein are merely to distinguish similar objects and do not represent a specific ordering of the objects. Understandably, the specific order or sequence of "first" and "second" can be interchanged where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein.

[0021] Example 1 Please refer to Figure 1 This disclosure provides an open-source intelligence assessment method S100 based on contextual information. Specifically, the method S100 includes: S102, in response to the open source intelligence assessment task, identify at least one cognitive dimension, said cognitive dimension including at least one of a technical dimension, a tactical dimension and a strategic dimension; S104, Obtain context information corresponding to the intelligence, the context information including time context, geographical context and task context; S106, under the constraints of the context information, retrieve the corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension; S108, based on the entity information and its relationship information of each cognitive dimension, obtain the first dimension score of the cognitive dimension, and integrate the score and the context information to generate an evaluation conclusion, wherein the score includes the first dimension scores of all cognitive dimensions.

[0022] In some implementations, the open-source intelligence is financial intelligence. Examples include "a sovereign wealth fund increases its holdings in a certain industry ETF" or "a country's central bank issues new regulations for real-time payment systems."

[0023] In step S102, a three-tiered cognitive framework model is established, including a strategic cognitive layer, a tactical cognitive layer, and a technical cognitive layer. This is combined "as needed," with at least one level activated to form a cognitive framework for intelligence assessment. By establishing a mapping relationship between the cognitive framework and the assessment task, dynamic selection of the cognitive framework driven by the task is achieved. The technical cognitive dimension focuses on the technological value and maturity of intelligence. The tactical cognitive dimension focuses on the practical deployment and execution effectiveness of intelligence. The strategic cognitive dimension focuses on the macro-ecological impact and long-term structural influence of intelligence.

[0024] The dynamic selection of the cognitive framework involves extracting expert experience and modeling the cognitive process. More specifically, the modeling includes at least one of the following: a financial keyword library, a cognitive dimension label library, and a pre-trained classification model of the cognitive framework. These models are continuously maintained and iterated. For example, a dual-threshold strategy of online confidence decline alerts and expert sampling review is used to ensure that the cognitive framework evolves in sync with the real financial situation, avoiding "model aging" that leads to evaluation distortion.

[0025] In some implementations, in response to an open-source intelligence assessment task, at least one cognitive dimension is identified, including: The intelligence is matched with a pre-built database of keywords in the financial field to obtain keywords; Based on the keywords, at least one of the cognitive dimensions is determined.

[0026] In some implementations, the keywords include financial keywords within the intelligence content. Specifically, financial keywords may be: Intelligence type keywords, such as product innovation, market fluctuations, and policy releases; Geographic keywords, such as Asia-Pacific region, global; Institutional / asset keywords, such as sovereign wealth funds, quantitative funds, and central bank digital currencies; Strategic keywords, such as arbitrage strategy, cross-border settlement, and risk hedging.

[0027] In some implementations, the keywords include financial keywords in the intelligence content, as well as task keywords extracted based on the user type / task type field in the assessment task.

[0028] Correspondingly, the input for the evaluation task can be intelligence with user type fields, such as trader, researcher, or compliance officer. Task type fields are then formed based on the user type fields, and task keywords are further extracted.

[0029] Key terms for the tasks include, for example, liquidity monitoring, portfolio adjustment, and macroeconomic early warning.

[0030] Task type fields, such as "Assess the fintech maturity of this information disclosure," "Determine whether this intelligence has immediate value for an upcoming trading strategy," and "Determine whether this intelligence foreshadows a significant change in monetary policy in a region." Inputs for assessment tasks can also be the intelligence and task type fields.

[0031] The input for an evaluation task can also be the intelligence content and the task type field.

[0032] In some implementations, at least one cognitive dimension is determined based on the keywords, including: The keywords are matched with a pre-set cognitive dimension tag library to obtain cognitive dimension tags; Based on the cognitive dimension labels, at least one of the cognitive dimensions is determined.

[0033] The cognitive dimension tag library and tag-frame lookup table were pre-compiled by financial experts. First, the mapping between keywords and tags in the cognitive dimension tag library is used to obtain the tags corresponding to each keyword. Then, tag combinations are used to look up the tag-frame lookup table to determine the cognitive frame containing at least one cognitive layer.

[0034] For example, the intelligence content (in short): A country's central bank is promoting the international interconnection of real-time payment systems; Keywords: central bank, cross-border, payment; Tags: Regulatory agencies, cross-border business, payment systems; Cognitive Framework: Activating the strategic and tactical cognitive layers.

[0035] The "two-level mapping" of keywords-tags-frameworks solidifies expert human experience into computable rules, completes the screening of cognitive dimensions, and meets the hard requirement of "explainability" in relevant domain scenarios.

[0036] In some implementations, at least one cognitive dimension is determined based on the keywords, including: The keywords are input into a pre-trained classification model trained on labeled data in the financial field to determine at least one of the cognitive dimensions.

[0037] The classification model outputs the probability that the intelligence value belongs to each cognitive dimension. If the probability of any cognitive dimension is greater than the threshold, the corresponding cognitive layer is activated.

[0038] In step S104, the time context represents the time period during which the intelligence needs to be evaluated, such as the time period that the evaluation task focuses on. For example, its source could be the valid time period corresponding to an entity in the intelligence text, user input, etc. The geographic context represents the geographic range during which the intelligence needs to be evaluated. For example, its source could be geographic entities in the intelligence text, user input, etc. The task context represents the evaluation objective of the evaluation task. For example, its source could be the aforementioned user type / task type field, task keywords, etc.

[0039] Contextual information can be obtained by performing NER (Network Errata) on the intelligence content in the financial field, or by sentence-level matching to extract contextual keywords from a pre-built contextual keyword library.

[0040] The scope of contextual information can be based on the preset user type / task type field or included in the task field, used to define snapshot windows in knowledge graphs and temporal geographs.

[0041] In step S106, the dynamic update mechanism of the knowledge graph includes: entity evolution tracking (monitoring changes in financial products, institutional structures, and business layouts); dynamic relationship adjustment (updating the strength and type of relationships between entities based on the latest market events); and incremental attribute updates (supplementing and correcting knowledge entity attributes based on newly acquired intelligence information). Relationship information can be one or more. Relationship information can be used to characterize the weighted relationships between connected entity information. Specifically, the acquisition process includes: obtaining seed entities based on intelligence, then extending them outwards by a certain number of hops, removing duplicates, and obtaining the extended relationship edges as relationship information, along with the entities connected to those edges.

[0042] In step S106, under the constraints of the context information, the corresponding entity information and its relationship information are retrieved from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relational edges.

[0043] Each edge in the knowledge graph has a pre-built index key =<geo-cell, time-slot, edge-type> This includes geographic grid encoding, time-based bucket numbering, and edge predicate enumeration, enabling the graph engine to locate all edge blocks that simultaneously satisfy the temporal, spatial, and cognitive dimensions in a single sequential read.

[0044] The entity type and relationship type are determined based on the cognitive dimension. Specifically, the cognitive dimension type whitelist is converted into search conditions, edges that match the cognitive dimension type are scanned, and all non-matching predicates are skipped, which can greatly reduce disk read volume (in one embodiment, it is reduced by 70%).

[0045] For example: Technical dimensions: Node type = {Payment network, Financial product, Algorithm model, …}, Edge type = {Product-technical characteristics, System-integration, Model-data source, …}; Tactical Dimension: Node Type = {Financial Institutions, Market Events, Regulatory Authorities, …}, Edge Type = {Institution-Region, Event-Participants, Policy-Scope of Impact, …}.

[0046] The whitelist of cognitive dimension types is continuously maintained by experts, and for example, it is further written according to each cognitive sub-dimension. Whitelist filtering includes determining whether an edge entity matches the whitelist during scanning, whether the entity at the edge's starting point matches the whitelist, and whether the entity at the edge's ending point matches the whitelist; otherwise, it is discarded.

[0047] Knowledge in a knowledge graph, after being filtered through a triple cross-referencing of cognitive dimensions and spatiotemporal context, can yield different knowledge subnetworks from different task perspectives, thus producing significantly different but evidence-based assessment conclusions that are more aligned with task requirements. For example, intelligence: A country's central bank announces a pilot program for cross-border digital currency clearing.

[0048] Search results: 30+ days in Southeast Asia + tactical understanding, knowledge sub-networks acquired: digital currency - clearing channels × regional banks - business coverage × cross-border capital flows.

[0049] Search: 365 days + global + strategic understanding, knowledge sub-networks acquired: digital currency - impact of monetary policy × international payment alliances - interoperability × regulatory policies - degree of openness.

[0050] In some implementations, under the constraints of the context information, retrieving corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension includes: Based on the task context, determine the number of relational edge hops of the seed entity.

[0051] By using a task-hop count mapping table, k-hop is limited to the depth truly required by the task, enabling adaptive knowledge acquisition and controllable query complexity. For example, research-oriented and policy-oriented tasks require different depths, allowing for task-driven, flexible edge hop counts. The upper limit of the hop count determines the longest causal chain required by the task, ensuring no critical remote nodes are missed and no irrelevant branches are pulled in, thus reducing the size of some task edges (including the resulting entity nodes) and avoiding resource waste caused by full queries.

[0052] In some implementations, the intelligence knowledge graph includes multiple sub-graphs, each of which achieves information fusion through entity alignment and relation mapping; the sub-graphs include a financial product and institution knowledge graph, a macroeconomic knowledge graph, and a regulatory policy knowledge graph; Under the constraints of the aforementioned contextual information, the corresponding entity information and its relational information are retrieved from the intelligence knowledge graph according to each cognitive dimension, including: The technical dimension only calls the knowledge graph of financial products and institutions, the tactical dimension only calls the knowledge graph of macroeconomics, and the strategic dimension only calls the knowledge graph of regulatory policies.

[0053] Entity alignment specifically involves unifying entity URIs. The same entity reuses its base URI across multiple sub-graphs. If the same URI in multiple sub-graphs conflicts in terms of attributes, the latest timestamp takes precedence. For example, the entity "fin:DigitalBond" in the Financial Products and Institutions Knowledge Graph is aligned with the entity "fin:DigitalBond" in the Macroeconomic Knowledge Graph using the unified URI "fin:product / digital-bond". Alignment attributes include: issuing entity, scenario, and compliance status.

[0054] Relational mapping connects semantically identical or upstream / downstream edges in different subgraphs through OWL equivalence or sub-attribute declarations, thereby piecing together the originally relationally fragmented subgraphs into a "virtual large graph" containing all knowledge for cross-layer reasoning.

[0055] Using a corresponding sub-graph for each cognitive dimension can isolate noise, and single-graph queries reduce recall, ensuring that the knowledge used is more aligned with the evaluation perspective. This improves accuracy while reducing computational load and increasing response time. It also increases the interpretability of the scoring, eliminating the need to explain, for example, "market fluctuations affected the technical assessment," and shortens the scoring tracing path.

[0056] In step S108, entity information is converted into entity vectors and relation information is converted into relation vectors. The feature dimensions of the two types of vectors are the same. Then, they are convoluted into fixed-length graph vectors by GNN and sparsely represented by adjacency matrix.

[0057] Then input at least one independent rating model under this cognitive dimension to obtain the first dimension rating.

[0058] In some implementations, a first dimension score for each cognitive dimension is obtained based on the entity information and its relationship information for each cognitive dimension, including: The entity information and relationship information of all the cognitive dimensions are fused into a micromap; Based on the micromap, a first dimension score is obtained for each of the cognitive dimensions.

[0059] For assessment tasks that activate multiple cognitive layers, knowledge from different cognitive dimensions is interconnected and partially integrated before assessment of each cognitive dimension. This avoids the evidence used for each cognitive dimension being fragmented and forming isolated islands, thereby reducing the potential inherent conflicts between the first-dimensional scores of each cognitive dimension. This results in a highly consistent and interpretable assessment of the value of financial intelligence (in one example, expert consensus increased from 0.74 to 0.91).

[0060] In some implementations, each of the cognitive dimensions corresponds to a plurality of cognitive sub-dimensions.

[0061] For example, the cognitive sub-dimensions under the technology cognitive dimension include: the value of fintech, strategic impact, and timeliness. The cognitive sub-dimensions under the tactical cognitive dimension include: regional market impact, the value of fintech, and the value of regulatory guidance.

[0062] Each cognitive sub-dimension corresponds to an independent scoring model. For example, the scoring model for the cognitive sub-dimension of technological value is: Technological Value Score = 0.4 × Technological Breakthrough Level + 0.3 × Technological Maturity + 0.3 × Application Prospects. The scoring model uses preset weights.

[0063] Knowledge used: entity: ofin: CBDC-Link (Central Bank Digital Currency Cross-border Interconnection Platform) ofin: ISO20022 - Gateway (Cross-border Message Gateway) ofin:AML-CoPilot (Intelligent Anti-Money Laundering Compliance Module) side: oCBDC-Link supports the following message standard: ISO20022-Gateway ISO 20022-Gateway Compliance Certification Level: A Number of oCBDC-Link connections to the central bank: 15 oCBDC-Link Integrated Compliance Module: AML-CoPilot oAML-CoPilot Compliance Maturity Level: 8 oCBDC-Link's market penetration rate leadership: 1.2 (20% higher than major competitors).

[0064] The innovation breakthrough score is output by a pre-trained breakthrough regressor: the number of central bank connections (integer: 15) and the Boolean value of the hit keyword "cross-border real-time settlement" are concatenated to form a 2-dim feature vector, which is then input into a lightweight XGBoost regression model. This model has been trained on 2000 fintech samples with minimum MSE and directly outputs a continuous score from 0 to 10 for subsequent technology value scoring. Innovation Breakthrough Score = 9.1.

[0065] The ISO20022 gateway's compliance certification level is converted to 9 points, and the weighted average with the AML module maturity level of 8 points yields a compliance maturity level of 8.6.

[0066] Based on expert rules: It has cross-border instant settlement capabilities and a built-in anti-money laundering module, and has high diffusion potential. Market prospect = 9.4.

[0067] Output score 9.2 / 10 (single-value floating point 0-10, rounded to 1 decimal place).

[0068] The rating includes a confidence level. For example, using the formula Confidence Level = 1 - (Number of missing attributes / Total number of attributes), the confidence level can be used for subsequent conflict resolution. The above rating hits all three criteria: Breakthrough, Maturity, and Prospect, resulting in a 100% confidence level.

[0069] In some implementations, the classification model employs a multi-level classification method. After confirming the cognitive framework, it further outputs the probability of classifying intelligence into each selected cognitive dimension's evaluation dimension based on task context information. The number N of selected evaluation dimensions can be preset. For example, N=3, representing two sub-dimensions under the tactical cognitive dimension and one sub-dimension under the technical cognitive dimension. Furthermore, N can be dynamically assigned by the user / task type field, for example: high-frequency trader, N=2; researcher, N=3; compliance officer, N=1. The same intelligence is automatically matched to the optimal computational load under different tasks, avoiding resource waste or insufficient information caused by a "one-size-fits-all" approach. Each cognitive dimension can include mandatory cognitive sub-dimensions. By outputting the probabilities of each cognitive sub-dimension at once through a multi-level classification model, the top-N sub-dimensions are automatically extracted, improving the response speed of intelligence evaluation.

[0070] In some embodiments, the method further includes: Based on the task context, the cognitive dimension scheduling agent determines at least one cognitive sub-dimension as the evaluation dimension of the evaluation task. Based on the entity information and relationship information of each cognitive dimension, a first dimension score for that cognitive dimension is obtained, including: Based on the entity information and its relationship information of each cognitive dimension, a second dimension score is obtained for each evaluation dimension corresponding to the cognitive dimension, and the second dimension scores of all evaluation dimensions corresponding to the cognitive dimension are used as the first dimension score of the cognitive dimension.

[0071] Pruning the evaluation dimension space based on the task context allows for a more targeted approach when generating natural language reasons for the evaluation results. This involves incorporating the evidence chain of the selected cognitive sub-dimensions into the report, shortening the average generated length, and ensuring that all key reasons fall within the task's domain of concern, thereby reducing reading time and the rate of decision-making errors.

[0072] In some implementations, under the constraints of the context information, retrieving corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension includes: Under the constraints of the aforementioned context information, first entity information and its first relationship information are obtained from the intelligence knowledge graph according to each of the aforementioned evaluation dimensions; Based on the entity information and relationship information of each cognitive dimension, a second dimension score is obtained for each evaluation dimension corresponding to that cognitive dimension, including: Based on the first entity information and its first relationship information, a second dimension score is obtained for each of the evaluation dimensions.

[0073] Based on each evaluation dimension, further corresponding graph knowledge is selected to achieve linear controllability of graph query volume. This ensures coverage of the task's essential needs while avoiding excessive low-relevance knowledge obtained from other cognitive sub-dimensions under the same cognitive dimension, which could lead to score fluctuations. Knowledge that is only related to other cognitive sub-dimensions (including selected and unselected cognitive sub-dimensions) under the same cognitive dimension is no longer introduced. This ensures that the standard deviation of the same intelligence being evaluated multiple times in a short period under the same evaluation dimension is within the expected range.

[0074] In some embodiments, the method further includes: Under the constraints of the aforementioned contextual information, at least one risk coefficient is obtained based on the temporal geopolitical map; The score for the first dimension is corrected based on the risk coefficient.

[0075] The time-series geopolitical map includes: Nodes: Countries / regions, regulatory agencies, and key financial infrastructure; Event edge: Event type + impact intensity; Event types include, for example, currency relationships (currency swaps, interest rate linkage), policy relationships (regulatory agreements, sanctions, macroprudential announcements), technological relationships (payment network interconnection, data exchange), and market events (capital outflows, liquidity shocks, credit defaults).

[0076] Each edge is appended with an effective-ineffective time. A time-series graph attention network is used to calculate node-level risk intensity through multi-step information propagation.

[0077] Methods for constructing sequence graph attention networks include: Node feature design: The static attributes and dynamic states of the macroeconomy and financial institutions are used as node features; Edge weight calculation: The weight of the edge is determined based on the impact intensity and propagation path of the financial event; Time-series aggregation mechanism: Employs an attention mechanism to aggregate financial intelligence information from different time windows.

[0078] Risk coefficients include market volatility intensity coefficient, technology diffusion risk coefficient, compliance deviation coefficient, liquidity tightening multiple, and comprehensive risk coefficient (a weighted average of all / multiple risk coefficients). Each risk coefficient is used to adjust one or more first-dimensional / second-dimensional scores.

[0079] Similar to knowledge graphs, in some implementations, a time retrieval window is determined based on the time context, a geographical retrieval range is determined based on the geographical context, and the event edge type is determined based on the cognitive dimension. A query is performed on the temporal geography graph within the intersection of the time retrieval window and the geographical retrieval range. During the query, a risk subgraph is obtained based on predicate filtering, for example, containing only monetary / policy edges, or multiple edges. The risk subgraph also contains the nodes corresponding to the edges.

[0080] Nodes or edges have pre-defined quantifiable risk values ​​(market volatility intensity coefficient, technology diffusion risk coefficient, etc.), which are then used to... Figure 1 For initial storage, the temporal graph attention network directly reads these values ​​as feature vector input. The risk subgraph formed by nodes with risk attributes and edge vectors is input into the temporal graph attention network. After time decay and neighbor aggregation, the latent vectors of seed nodes are summed and pooled to obtain interpretable node-level risk coefficients, which are used to correct the first dimension score in real time.

[0081] For example, the following risk correction rule can be adopted:

[0082] Where final_dimension represents the corrected score, and dimension represents the original first-dimension score of the corresponding cognitive dimension / the original second-dimension score of the corresponding cognitive sub-dimension. This indicates the first cognitive dimension / sub-dimension associated with it. j Each risk factor (such as market volatility intensity, compliance deviation, etc.) can be considered. To avoid excessive amplification of risk impact, each... Use a truncation rule such as [-0.25, 0.25].

[0083] In some implementations, reasoning is performed based on time-series geopolitical maps to obtain at least one risk coefficient, including: Identify geopolitical risk source nodes from the micromap and determine the initial risk intensity of each risk source node based on the temporal context; Calculate the risk propagation gain based on the graph path distance and edge weights between the risk source node and the target entity of the assessment task; By combining a preset time decay function, the loss of the initial risk intensity along the propagation path is corrected to obtain a comprehensive risk coefficient for a specific cognitive dimension.

[0084] In some implementations, the logic for determining the initial risk intensity includes: Extract the characteristics of emergencies, policy changes, or market fluctuations associated with the geographic context; By comparing historical geopolitical data from the same period, the uncertainty of the current context is calculated using the information entropy model, and then mapped to a risk scalar between 0 and 1 as the initial risk intensity.

[0085] By introducing time-series geopolitical maps and risk coefficients, and quantifying the risk coefficients, a clear basis for adjustment is provided for the assessment, making the assessment results more scientific and reasonable. Dynamic correction of dimensional scores through quantified risk coefficients ensures that the assessment results reflect the latest changes in the financial environment and time. This dynamic correction mechanism enhances the timeliness and adaptability of the assessment, enabling it to promptly capture the impact of changes in the external environment on intelligence value.

[0086] Furthermore, in step S108, the multi-dimensional scoring is deeply coupled with the contextual constraints.

[0087] First, the system performs weighted aggregation and / or conflict resolution on the scores of each dimension (e.g., triggering arbitration rules when the difference between the technical dimension 9.2 and the strategic dimension 6.0 exceeds the threshold); second, it converts time, geography, and task context into limited labels and embeds them into the evaluation conclusions to clarify the applicable boundaries of time and space; finally, it solidifies the knowledge graph retrieval path as an evidence chain, records the basis for score tracing (e.g., entity node ID and relationship hop path), and adaptively generates report templates based on task type.

[0088] By structurally integrating multi-dimensional scoring, contextual constraints, and knowledge graph evidence chains, the assessment conclusions possess a complete logical framework and clear spatiotemporal applicability boundaries, significantly improving interpretability. Simultaneously, the built-in conflict resolution mechanism effectively balances cross-dimensional cognitive biases, ensuring the consistency and robustness of the assessment conclusions. Furthermore, contextual semantic embedding enables the same intelligence to adaptively generate differentiated assessment conclusions based on different task scenarios. Finally, the solidified entity relationship tracing path meets the compliance requirements of financial regulators for a traceable and verifiable assessment process, forming a credible assessment system covering the entire process from data to conclusions.

[0089] The described method S100 successfully replicates the thought process of experts in the relevant domain through multi-level cognitive framework modeling, achieving an assessment consistency of over 92%. Based on a context-adaptive mechanism and employing dynamic knowledge fusion, it can dynamically adjust the evidence used in the assessment according to different assessment scenarios, achieving a scenario adaptation accuracy of 89%, a 28% improvement compared to static methods. The system also exhibits good scalability: its modular system architecture facilitates the integration of new professional domain knowledge and assessment algorithms, supporting rapid deployment and functional expansion.

[0090] In some implementations, to improve the collaborative efficiency of complex intelligence tasks, a multi-agent collaborative analysis architecture is constructed, decoupling the evaluation links such as context parsing, cognitive scheduling, graph retrieval, risk calibration and report generation into autonomous service nodes, and realizing distributed collaboration through standardized interfaces and a shared blackboard mechanism.

[0091] By decomposing complex assessment tasks into independently verifiable and parallel subtasks through a multi-agent collaborative architecture (such as simultaneous execution of graph retrieval and risk calibration), the system can significantly improve its responsiveness to high-timeliness and multi-dimensional conflict scenarios while maintaining assessment accuracy. Furthermore, the mandatory writing of collaborative logs enables the traceability and auditability of the entire assessment process.

[0092] Please refer to Figure 2 This disclosure provides an open-source intelligence multi-agent collaborative evaluation method S200 based on context information. Specifically, the method S200 includes: S202, Construct a multi-agent collaborative analysis architecture, which includes a context parsing agent, a cognitive dimension scheduling agent, a graph retrieval agent, and a report generation agent; S204, in response to the open-source intelligence assessment task, the cognitive dimension scheduling agent determines at least one cognitive dimension, which includes at least one of a technical dimension, a tactical dimension, and a strategic dimension; S206, The context parsing agent obtains context information corresponding to the intelligence, the context information including time context, geographic context and task context; S208, the graph retrieval agent, under the constraints of the context information, retrieves corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relation edges as the corresponding entity information and relation information; S210, the multi-agent collaborative analysis architecture obtains the first dimension score of the cognitive dimension based on the entity information and relationship information of each cognitive dimension, and the report generates an agent integrated score and the context information to generate an evaluation conclusion, wherein the score includes the first dimension scores of all cognitive dimensions.

[0093] In some implementations, the multi-agent collaborative analysis architecture further includes a risk calibration agent; The method further includes: S212, the risk calibration agent, under the constraints of the context information, obtains at least one risk coefficient based on the temporal geopolitical map; S214, Based on the risk coefficient, the risk calibration agent corrects the score of the first dimension.

[0094] In some implementations, the report-generating agent includes an arbitration agent; The method further includes: When there is a conflict or the confidence level is lower than the threshold for the first dimension scores corresponding to different cognitive dimensions, the arbitration agent performs arbitration based on a weighted voting strategy or expert rules, and records the source of conflict, arbitration strategy and final conclusion in the collaborative log to generate a traceable evaluation conclusion.

[0095] In this architecture, multiple agents communicate with each other via a shared blackboard or message bus, and write collaborative logs upon task completion to ensure the evaluation process is traceable and interpretable. This collaborative mechanism can be combined with multi-level classification and scoring models to give the system adaptive and scalable capabilities.

[0096] The first dimension score is the result of multiple agents applying their respective logical rules (graph algorithm, time decay, risk correction) to the same "sub-graph" data and then superimposing the scores.

[0097] For example, the context parsing agent extracts time, geography, and task context from intelligence text and task fields, outputs a standardized context vector, and sends the extraction results to the cognitive dimension scheduling agent. The cognitive dimension scheduling agent matches cognitive dimensions (and further, cognitive sub-dimensions) according to context constraints and distributes the scheduling results to the graph retrieval agent. The graph retrieval agent calls the intelligence knowledge graph based on the cognitive dimension whitelist and hop count rules, generates the corresponding micro-graph, and shares it with the risk calibration agent. The risk calibration agent infers the risk coefficient based on the temporal geography graph and corrects the scores of each cognitive dimension. The report generation agent generates a structured evaluation conclusion based on the score, risk coefficient, and contextual evidence.

[0098] In one example process, the cognitive dimension scheduling agent breaks down the intelligence task of "cross-border digital currency clearing network" into three collaborative links: "innovation maturity, market prospects, and compliance risks." The context parsing agent outputs the context of "the next two years, Southeast Asia, and real-time payment regulation." The cognitive scheduling agent activates the target cognitive dimension of technology and tactics. The graph retrieval agent extracts sub-graphs according to the corresponding whitelist for each target cognitive dimension and writes them into the shared blackboard. The risk calibration agent reads the node IDs in the blackboard, retrieves events such as changes in payment regulation and capital flow controls within the time-series geopolitical graph, and returns the risk coefficient. 0.15, 0.0, 0.15; The report generates a summary score and risk for the intelligent agents, and constructs a reasoning paragraph; If there is a conflict between the market prospects and the recommendations given by risk control, the arbitration intelligent agent performs a vote with confidence as the weight, and records "source of conflict, arbitration strategy, and final conclusion" in the collaboration log, forming a traceable multi-agent interaction sequence.

[0099] Example 2 This example assesses an intelligence report that states "multiple central banks and commercial banks jointly launched a cross-border distributed ledger settlement network."

[0100] Keyword hits: Distributed ledger, cross-border settlement, instant liquidity, regulatory sandbox, quantitative funds, compliance review, risk control. The input classification model outputs the Top-3 cognitive sub-dimensions as technology maturity, market prospects, and compliance risk.

[0101] Simultaneously activate the technical cognition layer and the tactical cognition layer.

[0102] Time context: the next 18 months.

[0103] Geographical context: Europe and Southeast Asia.

[0104] Task Context: Assess the impact of this network on cross-border capital flows and compliance risks.

[0105] The acquired knowledge chain includes: Technological Evolution Path: An Iterative Record from Single Central Bank Pilot Programs to Multi-Central Bank Interconnection; Business application knowledge: Case studies of ledger networks in trade finance and foreign exchange clearing; Compliance strategy knowledge: Regulatory sandbox conditions and AML requirements in various countries.

[0106] The acquired entity information and its relationship information are input into the scoring model corresponding to each cognitive sub-dimension, and the scores of the second dimension are output respectively: technology maturity = 9.4 / 10, market prospects = 9.0 / 10, compliance risk = 7.5 / 10.

[0107] Risk coefficient obtained: Regulatory tightening coefficient = -0.15.

[0108] The second dimension score for market prospects has been revised to 7.6.

[0109] The score for the second dimension of compliance risk has been revised to 6.4.

[0110] Output evaluation report: Real-time business value: High (can significantly reduce cross-border settlement time).

[0111] Leading window: Medium (12-18 months).

[0112] suggestion: 1. Apply for access within the regulatory sandbox framework as soon as possible, and simultaneously deploy and debug the AML model; 2. Complete interface evaluation and performance stress testing with the existing clearing system within 60 days; 3. Continuously monitor core central bank policy meetings and capital flow data, and dynamically adjust liquidity reserves.

[0113] Example 3 This disclosure provides an open-source intelligence assessment device 1000 based on context information.

[0114] The evaluation device 1000 may include corresponding modules that execute one or more steps of the flowchart of the above-described open-source intelligence evaluation method S100 based on context information. Therefore, each or more steps in the flowchart may be executed by a corresponding module, and the evaluation device 1000 may include one or more of these modules. A module may be one or more hardware modules specifically configured to execute a corresponding step, or implemented by a processor configured to execute a corresponding step, or stored in a readable storage medium for processor implementation, or implemented through some combination thereof.

[0115] Specifically, such as Figure 3As shown, the evaluation device 1000 includes: Dimension acquisition module 1002 is used to determine at least one cognitive dimension in response to an open source intelligence assessment task, wherein the cognitive dimension includes at least one of a technical dimension, a tactical dimension and a strategic dimension; The context acquisition module 1004 is used to acquire context information corresponding to the intelligence, the context information including time context, geographic context and task context; The knowledge acquisition module 1006 is used to retrieve corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, under the constraints of the context information, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relation edges as the corresponding entity information and relation information; The evaluation module 1008 is used to obtain a first dimension score for each cognitive dimension based on the entity information and its relationship information of each cognitive dimension, and to integrate the score and the context information to generate an evaluation conclusion, wherein the score includes the first dimension scores of all cognitive dimensions.

[0116] Furthermore, this disclosure provides an open-source intelligence multi-agent collaborative evaluation device 2000 based on context information.

[0117] Specifically, such as Figure 4 As shown, the evaluation device 2000 includes: Architecture building module 2002 is used to build a multi-agent collaborative analysis architecture, which includes a context parsing agent, a cognitive dimension scheduling agent, a graph retrieval agent, a risk calibration agent, and a report generation agent; The dimension acquisition module 2004 is used to respond to the open source intelligence assessment task by determining at least one cognitive dimension by the cognitive dimension scheduling agent, wherein the cognitive dimension includes at least one of the technical dimension, tactical dimension and strategic dimension; The context acquisition module 2006 is used to acquire context information corresponding to the intelligence by the context parsing agent. The context information includes time context, geographic context and task context. The knowledge acquisition module 2008 is used by the graph retrieval agent, under the constraints of the context information, to retrieve corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the entity type and relationship type are determined based on the cognitive dimension. Within the intersection of the time retrieval window and the geographic retrieval range, a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relation edges as the corresponding entity information and relation information; The risk calibration module 2010 is used to obtain at least one risk coefficient based on a time-series geopolitical map under the constraints of the context information, and to correct the first dimension score based on the risk coefficient. The evaluation module 2012 is used to calculate the first dimension score of each of the cognitive dimensions by the multi-agent collaborative analysis architecture based on the entity information and its relationship information, and the report generating agent integrates the score and contextual evidence to generate an evaluation conclusion.

[0118] This disclosure also provides an electronic device, including: a memory storing execution instructions; and a processor or other hardware module executing the execution instructions stored in the memory, causing the processor or other hardware module to perform the evaluation methods S100 / S200 described above.

[0119] This disclosure also provides a readable storage medium storing executable instructions, which, when executed by a processor, are used to implement the above-described evaluation methods S100 / S200.

[0120] The hardware architecture of the evaluation apparatus 1000 / 2000, implemented using a processor-based hardware approach according to this disclosure, can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Buses 1100 / 2100 connect various circuits including one or more processors 1200 / 2200, memories 1300 / 2300, and / or hardware modules. Buses 1100 / 2100 can also connect various other circuits 1400 / 2400, such as peripheral devices, voltage regulators, power management circuits, external antennas, etc.

[0121] Bus 1100 / 2100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component Architecture (EISA) bus, etc. Bus 1100 / 2100 can be divided into address bus, data bus, control bus, etc. For ease of representation, only one connection line is used in this diagram, but this does not indicate that there is only one bus or one type of bus.

[0122] Any process or method description in the flowcharts or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of preferred embodiments of this disclosure includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of this disclosure pertain. The processor performs the various methods and processes described above. For example, the method embodiments of this disclosure may be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some embodiments, part or all of the software program may be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the processor may be configured to perform one of the methods described above by any other suitable means (e.g., by means of firmware).

[0123] The logic and / or steps represented in the flowchart or otherwise described herein may be specifically implemented in any readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-based system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0124] For the purposes of this specification, a "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use in or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM). Furthermore, a readable storage medium can even be paper or other suitable media on which a program can be printed, since a program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in memory.

[0125] It should be understood that various parts of this disclosure can be implemented in hardware, software, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0126] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0127] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a single processing module, or each unit can exist physically separately, or two or more units can be integrated into a single module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a readable storage medium. The storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0128] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A method for open source intelligence evaluation based on context information, characterized in that, The method includes: A multi-agent collaborative analysis architecture is constructed, which includes a context parsing agent, a cognitive dimension scheduling agent, a graph retrieval agent, and a report generation agent. The agents communicate with each other through a shared blackboard or message bus. In response to the open-source intelligence assessment task, the cognitive dimension scheduling agent determines at least one cognitive dimension, which includes at least one of a technical dimension, a tactical dimension, and a strategic dimension. The context parsing agent obtains context information corresponding to the intelligence, including time context, geographic context, and task context. Under the constraints of the context information, the graph retrieval agent retrieves corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the cognitive dimension type whitelist is converted into retrieval conditions. Within the intersection of the time retrieval window and the geographical retrieval range, edges matching the cognitive dimension type are scanned, all non-matching predicates are skipped, and a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relational edges as the corresponding entity information and relational information. The intelligence knowledge graph includes multiple sub-graphs, each of which achieves information fusion through entity alignment and relational mapping. The sub-graphs include a financial product and institution knowledge graph, a macroeconomic knowledge graph, and a regulatory policy knowledge graph. The technical dimension only calls the financial product and institution knowledge graph, the tactical dimension only calls the macroeconomic knowledge graph, and the strategic dimension only calls the regulatory policy knowledge graph. Each edge in the intelligence knowledge graph has a pre-built index key =<geo-cell, time-slot, edge-type> This includes geographic grid coding, time-based bucket numbering, and edge predicate enumeration; The multi-agent collaborative analysis architecture obtains a first dimension score for each cognitive dimension based on the entity information and relationship information of each cognitive dimension, and generates an integrated agent score and an evaluation conclusion based on the context information in the report, wherein the score includes the first dimension scores of all cognitive dimensions.

2. The method of claim 1, wherein, Each of the aforementioned cognitive dimensions includes multiple cognitive sub-dimensions; The method further includes: Based on the task context, at least one of the cognitive sub-dimensions is determined as the evaluation dimension of the evaluation task; Based on the entity information and relationship information of each cognitive dimension, a first dimension score for that cognitive dimension is obtained, including: Based on the entity information and its relationship information of each cognitive dimension, a second dimension score is obtained for each evaluation dimension corresponding to the cognitive dimension, and the second dimension scores of all evaluation dimensions corresponding to the cognitive dimension are used as the first dimension score of the cognitive dimension.

3. The method of claim 1, wherein, The method further includes: Under the constraints of the context information, at least one risk coefficient is obtained based on the temporal geography graph, wherein the temporal geography graph includes geographic nodes and event edges, the event edges are used to characterize the event type and impact intensity between two geographic nodes, and each event edge is appended with an effective-ineffective time. The score for the first dimension is corrected based on the risk coefficient.

4. The method of claim 1, wherein, In response to the open-source intelligence assessment task, at least one cognitive dimension was identified, including: The intelligence is matched with a pre-built database of keywords in the financial field to obtain keywords; Based on the keywords, at least one of the cognitive dimensions is determined.

5. An open source intelligence evaluation device based on context information, characterized by, The device includes: An architecture building module is used to build a multi-agent collaborative analysis architecture, which includes a context parsing agent, a cognitive dimension scheduling agent, a graph retrieval agent, a risk calibration agent, and a report generation agent. In response to the open-source intelligence assessment task, the cognitive dimension scheduling agent determines at least one cognitive dimension, which includes at least one of a technical dimension, a tactical dimension, and a strategic dimension. The context parsing agent obtains context information corresponding to the intelligence, including time context, geographic context, and task context. Under the constraints of the context information, the graph retrieval agent retrieves corresponding entity information and its relationship information from the intelligence knowledge graph according to each cognitive dimension, including: The time retrieval window is determined based on the time context, the geographical retrieval range is determined based on the geographical context, and the cognitive dimension type whitelist is converted into retrieval conditions. Within the intersection of the time retrieval window and the geographical retrieval range, edges matching the cognitive dimension type are scanned, all non-matching predicates are skipped, and a query is performed on the intelligence knowledge graph to obtain matching entity nodes and their relational edges as the corresponding entity information and relational information. The intelligence knowledge graph includes multiple sub-graphs, each of which achieves information fusion through entity alignment and relational mapping. The sub-graphs include a financial product and institution knowledge graph, a macroeconomic knowledge graph, and a regulatory policy knowledge graph. The technical dimension only calls the financial product and institution knowledge graph, the tactical dimension only calls the macroeconomic knowledge graph, and the strategic dimension only calls the regulatory policy knowledge graph. Each edge in the intelligence knowledge graph has a pre-built index key =<geo-cell, time-slot, edge-type> This includes geographic grid coding, time-based bucket numbering, and edge predicate enumeration; The multi-agent collaborative analysis architecture obtains a first dimension score for each cognitive dimension based on the entity information and relationship information of each cognitive dimension, and generates an integrated agent score and an evaluation conclusion based on the context information in the report, wherein the score includes the first dimension scores of all cognitive dimensions.

6. An electronic device, comprising: include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the method according to any one of claims 1-4.

7. A readable storage medium characterized by, The readable storage medium stores execution instructions, which, when executed by a processor, are used to implement the method described in any one of claims 1-4.

Images