A multi-factor value evaluation method and device for intelligence resource preferred configuration

Abstract

The present disclosure belongs to the technical field of intelligent information retrieval, and provides a multi-factor value evaluation method and device for intelligence resource optimal configuration. The method comprises: acquiring a query context corresponding to an intelligence auxiliary decision-making task; in response to the query context, scoring a to-be-evaluated intelligence in parallel in multiple intelligence value scoring factors to obtain a factor score corresponding to each intelligence value scoring factor; performing confidence evaluation on each factor score to obtain a corresponding confidence; modifying a corresponding basic weight according to the confidence of each factor score to obtain a corresponding modified weight; and obtaining a comprehensive score of the to-be-evaluated intelligence according to all the factor scores and the modified weights thereof, wherein the comprehensive score is used for priority sorting of the to-be-evaluated intelligence.

Description

Technical Field

[0001] This disclosure belongs to the field of intelligent information retrieval technology, and in particular relates to a multi-factor value evaluation method and apparatus for the optimal allocation of intelligence resources. Background Technology

[0002] In intelligence processing workflows geared towards intelligence decision-making tasks, decision-making bodies typically need to quickly identify high-value intelligence and allocate resources based on task requirements. However, existing intelligence value assessment methods generally suffer from the following shortcomings: ① They rely solely on a single or limited number of general dimensions (such as keyword matching or popularity) for scoring, failing to simultaneously consider task relevance, timeliness, and source reliability; ② The scores for each dimension are directly summed linearly with fixed weights, ignoring the differences in confidence levels between different dimensions, resulting in low-confidence factors lowering the overall assessment accuracy; ③ There is a lack of dynamic feedback mechanisms between the scoring results and the task context, making it difficult to adjust priorities in real time based on decision queries. These problems cause intelligence ranking to deviate from task requirements, reducing the efficiency of subsequent intelligence distribution and decision resource allocation. Summary of the Invention

[0003] This disclosure provides a multi-factor value assessment method and apparatus for the optimal allocation of intelligence resources, which can effectively solve the above-mentioned problems.

[0004] This disclosure is implemented as follows:

[0005] Firstly, this disclosure provides a multi-factor value assessment method for the optimal allocation of intelligence resources, the method comprising:

[0006] Obtain the query context corresponding to the intelligence-assisted decision-making task;

[0007] In response to the query context, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated.

[0008] The confidence level of each factor score is assessed to obtain the corresponding confidence level;

[0009] Based on the confidence level of each factor score, the corresponding base weights are adjusted to obtain the corresponding adjusted weights;

[0010] Based on all the factor scores and their adjusted weights, a comprehensive score is obtained for the intelligence to be evaluated, and the comprehensive score is used to prioritize the intelligence to be evaluated.

[0011] Secondly, this disclosure provides a multi-factor value assessment device for the optimal allocation of intelligence resources, the device comprising:

[0012] The context acquisition module is used to acquire the query context corresponding to intelligence-assisted decision-making tasks.

[0013] A multi-factor scoring module is used to respond to the query context and score the intelligence to be evaluated in parallel on multiple intelligence value scoring factors to obtain the factor score corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated.

[0014] The confidence assessment module is used to assess the confidence of each factor score and obtain the corresponding confidence level.

[0015] The correction coefficient acquisition module is used to correct the corresponding basic weights based on the confidence level of each factor score to obtain the corresponding corrected weights.

[0016] The optimization module is used to obtain a comprehensive score of the intelligence to be evaluated based on all the factor scores and their modified weights, and the comprehensive score is used to prioritize the intelligence to be evaluated.

[0017] Thirdly, this disclosure provides an electronic device, including:

[0018] Memory, the memory storing execution instructions; and

[0019] A processor that executes execution instructions stored in the memory, causing the processor to perform the method described in the first aspect.

[0020] 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.

[0021] Compared with the prior art, the beneficial effects of this disclosure are:

[0022] This disclosure provides a multi-factor value assessment method for the optimal allocation of intelligence resources. First, intelligence is scored in parallel using multiple factors. Then, the confidence level of each factor score is assessed. The confidence level is used to correct the closed-loop fusion of the basic weights of each factor, weakening low-confidence dimensions and strengthening high-confidence dimensions. This effectively eliminates misjudgments caused by traditional fixed-weight linear summation, improving the robustness and adaptability of the comprehensive score. Simultaneously, the comprehensive scoring process is driven by the query context, enabling dynamic assessment of intelligence value and ensuring that the optimal allocation of intelligence resources matches task requirements in real time. Attached Figure Description

[0023] Figure 1 This is a flowchart of S100, a multi-factor value assessment method for the optimal allocation of intelligence resources, provided in an embodiment of this disclosure.

[0024] Figure 2 This is a system architecture diagram of the multi-factor value assessment method S100 for the optimal allocation of intelligence resources provided in this embodiment of the disclosure.

[0025] Figure 3 This is a schematic diagram of the structure of a multi-factor value assessment device 1000 for the optimal allocation of intelligence resources provided in this embodiment of the disclosure. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] 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.

[0029] 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.

[0030] 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.

[0031] 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)."

[0032] 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.

[0033] Example 1

[0034] Please refer to Figure 1 This disclosure provides a multi-factor value assessment method S100 for the optimal allocation of intelligence resources.

[0035] Specifically, method S100 includes:

[0036] S102, Obtain the query context corresponding to the intelligence-assisted decision-making task;

[0037] S104, in response to the query context, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated.

[0038] S106, Calculate the confidence level of each factor score to obtain the corresponding confidence level;

[0039] S108, Based on the confidence level of each factor score, the corresponding basic weight is adjusted to obtain the corresponding adjusted weight;

[0040] S110, Based on all the factor scores and their modified weights, a comprehensive score for the intelligence to be evaluated is obtained, and the comprehensive score is used to prioritize the intelligence to be evaluated.

[0041] In some implementations, the method S100 is used for the optimal allocation of intelligence resources in fields such as finance, energy, supply chain, and public policy.

[0042] In step S102, the query context is the instruction of the input intelligence management system used to describe the intelligence requirement, and its form includes natural language text, structured field templates, etc.

[0043] For example, natural language text: "Determine whether the Federal Reserve is likely to announce an interest rate hike of 25 basis points or more next week."

[0044] Structured fields: {Subject: Federal Reserve, Action: Interest rate hike, Magnitude: >25 bp, Time window: [T, T+7d]}.

[0045] Contextualized recommendation: The title and key entities of the "policy details page" currently opened by the user are directly used as the query context to achieve "related intelligence recommendation".

[0046] The query context for intelligence-assisted decision-making tasks can be obtained through any one or a combination of the following methods:

[0047] 1. Natural Language Requirements Form

[0048] Create a new "Intelligence Request Form" in the intelligence-assisted decision-making system interface, enter the task description in natural language, and the background will generate a standard query vector through NLU parsing.

[0049] 2. Digital Twin Large Screen Interaction

[0050] On the digital twin dashboard, select the area of ​​interest (such as the Yangtze River Delta / Persian Gulf / chip supply chain hub) and select the event type from the drop-down menu (e.g., policy regulation, price fluctuations, logistics disruptions, technology export controls). The system will automatically output a structured requirement template containing "region-node-event tags".

[0051] 3. Permanent Subscription Conditions

[0052] Users pre-subscribe to keywords, regions, or platform types, and the backend caches these as persistent query contexts, periodically matching them with the intelligence stream.

[0053] 4. Early Warning Rule Engine

[0054] The rules engine outputs a Boolean logical expression based on a preset threshold (e.g., "Federal Reserve AND Interest Rate Hike AND 25-50 bp"). This expression becomes the real-time query context, triggering push notifications within seconds.

[0055] The query context automatically expires when the task's lifecycle ends.

[0056] The method S100 relies on a query context-driven mechanism to explicitly inject task intent into the intelligence value assessment link. In the comprehensive scoring stage, it dynamically weights and fuses multi-factor scores, so that the Top-K ranking drifts in real time with task requirements, effectively avoiding the "one-list-for-all" resource mismatch caused by static scoring due to the lack of context.

[0057] In step S104, during the intelligence value assessment, multiple intelligence value scoring factors constitute a multi-dimensional complementary perspective on intelligence value. Each factor scores in parallel from independent dimensions, forming a complementary structure with low correlation between dimensions and high information gain. This structure significantly broadens the coverage space of value signals and effectively avoids the loss of high-value intelligence in the selection process due to single-dimensional / few-dimensional scoring, thereby improving the recall rate of high-value intelligence.

[0058] In some implementations, the intelligence value scoring factor further includes:

[0059] The novelty factor is used to assess the novelty of the intelligence being evaluated relative to historical data.

[0060] Pattern disturbance factor is used to assess the potential impact of the intelligence to be assessed on the macro-system pattern.

[0061] Technology evolution signal factor, used to assess the value level of technology evolution in the intelligence to be evaluated.

[0062] By employing a six-dimensional parallel scoring system covering six key intelligence concerns—semantics, timeliness, credibility, innovation, strategic perspective, and technology—the subsequent comprehensive scoring reflects both the quality of the content itself and adapts to the specific priorities of different task scenarios. Furthermore, by dynamically adjusting the base weights of the six-dimensional factors, the same intelligence can reveal its corresponding value level for different tasks.

[0063] In some embodiments, the method S100 further includes:

[0064] Obtain the intelligence type label of the intelligence to be evaluated;

[0065] Based on the intelligence type label, the base weight of at least one of the intelligence value scoring factors is adjusted.

[0066] Intelligence type labels can be generated by multi-label classifiers, keyword rules, or manual methods, and then the corresponding weight adjustment table can be called based on the label to achieve scene adaptive fusion.

[0067] For example, a pre-trained language model (such as RoBERTa-base) can be followed by a linear mapping layer. The input is an intelligence title, summary, and keywords, and the output is a multi-label logical value. The training set uses historically labeled intelligence and incorporates samples automatically labeled with a domain dictionary. The loss function chosen is BCE With Logits Loss.

[0068] Table 1 shows an example of weight adjustment.

[0069]

[0070] For example, if the intelligence report states that "the digital yuan smart contract is being used for commodity settlement for the first time," and the current scenario is "financial transaction," then the weight of "market dynamics disruption" will be increased to highlight its potential impact on interest rates and exchange rates. If the scenario is "scientific research tracking," then the weight of "technology evolution signals" will be increased to focus on the innovative value of the contract itself.

[0071] In some implementations, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors, including:

[0072] Extract at least one entity and its initial attributes of the intelligence to be evaluated;

[0073] The entity and its initial attributes are matched with the knowledge graph to obtain extended attributes of the entity features, wherein the extended attributes include the corresponding initial attributes;

[0074] Generate structured vectors based on all entities and their extended attributes;

[0075] The intelligence to be evaluated is scored on the semantic relevance factor based on the structured vector.

[0076] The entities and their initial attributes extracted from the intelligence are mapped to nodes in the knowledge graph after being linked, transforming the original strings into queryable and scalable graph entities.

[0077] Specifically, the extended attributes are a subset of entity attributes in the knowledge graph that have been filtered for task relevance.

[0078] For example, using keywords from the current query context as anchors, one- or two-hop path filtering is performed within the knowledge graph, retaining only attribute edges whose relevance to the task topic is greater than a threshold, forming a "task-related attribute subset". This "task-related attribute subset" mechanism avoids the tail noise of vectorization from overwhelming key features and reduces the input dimensionality of subsequent scoring models, improving computational efficiency and evaluation accuracy.

[0079] An entity's initial or extended attributes include task-related relationship attributes between the entity and other co-existing entities.

[0080] In some implementations, the same "entity linking—attribute expansion—vectorization" process is performed on the query context as on the intelligence text to generate a structured vector corresponding to the query context. This vector has the same dimension and distribution as the structured vector of the intelligence text and can be directly used for cosine similarity calculation, thereby ensuring that the semantic relevance factor scoring is performed within a unified attribute space.

[0081] In some implementations, the knowledge graph employs a financial policy knowledge graph. For example, the nodes of a financial policy knowledge graph include central banks, benchmark interest rates, and monetary tools, while the edges include quantitative transmission relationships such as "interest rate hike → US Treasury yield ↑ → US dollar index ↑". By introducing a knowledge graph, the matching of intelligence and query context goes beyond the text content. Instead, it maps identified entities to the attribute space extended by the graph, achieving semantic alignment in the "tool-chain-impact" dimension. This accurately captures deep semantics and significantly improves the professional accuracy of semantic relevance factors. For instance, "Federal Reserve" in the text is mapped to a computable vector: <Subject: Federal Reserve, Policy Tool: Interest Rate Hike, Impact Chain: Liquidity → US Dollar Index, Quantitative Transmission: 25 bp → +1.2%>.

[0082] The attribute chain fields provided by the knowledge graph are complete and have stable sources, resulting in high confidence. Furthermore, extended attributes can increase semantic similarity with the query context, giving higher weight to domain-specific scores in subsequent fusion stages. This improves the matching accuracy between selected high-value intelligence and current task requirements, enabling the priority of high-value intelligence and precise resource allocation.

[0083] In some implementations, the specific calculation process of the semantic relevance factor S1 can be broken down into two steps: first, obtain the "basic semantic similarity", then obtain the "entity attribute enhancement score" from the knowledge graph, and finally fuse them according to weights.

[0084] Step 1, Basic semantic similarity:

[0085] The intelligence text and query context are input into a language model that has been pre-trained in a specialized domain (e.g., MIL-RoBERTa). The [CLS] vector is used to calculate the cosine similarity, which is then mapped to the [0, 1] interval and denoted as... .

[0086] Step 2, Entity Attribute Enhancement Score:

[0087] a) Entity Joining: Perform NER and entity joining on the query context and intelligence respectively to obtain two sets of nodes. , .

[0088] b) Subgraph expansion: with Using the seed as a reference, perform a 1-hop expansion within the knowledge graph to extract a subset of "task-related attributes".

[0089] c) Attribute Vectors: For each entity, perform embedding and pooling on the filtered attribute-value pairs to obtain the query attribute extension vector and the intelligence attribute extension vector, respectively. Only the attributes related to the query are retained. Directly connected attribute nodes. For the intelligence side, only nodes directly connected to... Directly connected attribute nodes. d) Enhancement score: Calculate the cosine similarity between the query attribute extension vector and the intelligence attribute extension vector, then map it to the interval [0, 1] and denote it as... .

[0090] Weight fusion:

[0091] Calculation formula: .in, .

[0092] Specifically, =0.6, =0.4.

[0093] In some implementations, the information heat factor S2 is based on the characteristics of intelligence dissemination, and its calculation formula is as follows: .

[0094] The hour difference between the time the intelligence was released and the current time.

[0095] Specifically, =100. Statistical analysis of 10,000 historical intelligence entries showed a median readership of 95 and a mean of 104. Rounding down to 100 was taken as the "zero decay baseline," ensuring that S2≈1 represents the "average popularity."

[0096] =2 indicates that the reference behavior has a higher weight on value.

[0097] For example, in high-frequency financial scenarios, =0.35, corresponding to the exponential decay rate of a "2-day half-life".

[0098] The calculation results of the above formula are mapped to [0, 1] after min-max truncation to obtain the factor score S2.

[0099] In some implementations, a credibility scoring system is established based on source credibility mapping and multi-source cross-validation, which includes different categories of sources such as official media, authoritative institutions, and professional media, in order to obtain the factor score S3 corresponding to the authority factor.

[0100] Specifically, a three-tiered process of "classification, mapping, and cross-validation" is adopted: first, the sources of intelligence are classified into multiple categories and assigned prior scores. Different categories are formed based on the credibility of the source. Then, they are further categorized according to the historical accuracy over the past 90 days. The credibility score of intelligence obtained from a single source is determined by dynamic correction of time-delay decay. Calculation formula: . Indicates the order of a single source. It indicates the number of days since the intelligence was released until the current time.

[0101] If the same intelligence event occurs within 24 hours from more than one independent source with different top-level domains and high credibility (e.g., confirmed by category), and the event triple overlap is not less than the threshold, then an additional 0.02 cross-validation score (capped at 0.1) is added for each additional source. The final output score is truncated to [0, 1] and used as the factor score S3. Calculation formula: . Indicates the quantity from a single source. This indicates the number of unique and highly reliable sources that are distinct from the domain's top-level domain.

[0102] In some implementations, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including:

[0103] The intelligence to be evaluated is matched with the novel keyword library to identify novel keywords and obtain the fragment containing the novel keywords as the first candidate novel fragment;

[0104] If the novel keyword is not identified, obtain the maximum semantic similarity between each sentence of the intelligence to be evaluated and the historical intelligence database. If the minimum value of the maximum semantic similarity is less than the first threshold, obtain the second candidate novel fragment according to the corresponding sentence.

[0105] Input the first candidate novel fragment or the second candidate novel fragment into the large language model, output the novelty probability, and if the novelty probability is not less than the second threshold, assign a base novelty score of 1.0, and perform exponential decay according to the relative time difference to obtain the factor score corresponding to the novelty factor.

[0106] Construct a thesaurus containing 500 novel keywords from various professional fields.

[0107] Taking financial novelty keywords as an example, sources can include: 1. Frequently used expressions such as "first time," "pilot," "innovation," and "breakthrough" in announcements from the central bank, the China Securities Regulatory Commission, and stock exchanges; 2. Naming rules for financial instruments (such as the first appearance of "Science and Technology Innovation Board 50 Options"); 3. Policy tool codes and regulatory event terminology (such as "T+0 reform").

[0108] Candidate novel fragments are continuous clauses with no more than 64 tokens that have been detected by keyword matching or low similarity. They are used to provide the large model with secondary confirmation of whether they have appeared for the first time globally.

[0109] Specifically, the first candidate novel fragment is the clause containing the hit word, then expands it by 5 words to the left and right, and truncates it to the beginning / end of the sentence without crossing paragraphs.

[0110] The second candidate novel fragment is selected by comparing each clause in the historical intelligence database with the least similar clause.

[0111] The acquisition of the second candidate novel fragment involves physical sentence segmentation and sentence-by-sentence comparison. Physical sentence segmentation involves dividing the information to be evaluated into a list of sentences using periods, question marks, and exclamation marks. The maximum semantic similarity is calculated by using Fin-BERT, which is pre-trained in the Fin-BERT database, to calculate the cosine similarity between each sentence in the list and all sentences in the past 30 days' historical database. If multiple sentences have the same value, the shortest sentence is selected to reduce noise during secondary verification.

[0112] The historical intelligence database includes: assessed intelligence from the past 30 days, plus summaries of announcements from Open Market Access / Wind.

[0113] Specifically, the first threshold is set to 0.65.

[0114] Specifically, the large language model is a 7-B parameter generative model, which is further pre-trained on financial corpora and fine-tuned with LoRA, and can output novelty probabilities in an offline environment.

[0115] Example of a second confirmation prompt for a large language model:

[0116] The following text was published in 20XX-XX. Please determine whether its content is appearing for the first time in publicly available media / official announcements worldwide.

[0117] {{candidate novel fragment}};

[0118] Output: {"First probability": P} novel}

[0119] Specifically, the second threshold is set to 0.8.

[0120] The core task of secondary verification is to verify whether the candidate novel fragment has indeed appeared for the first time in publicly reported global news. The model internally gives the probability by comparing the time-semantic distribution in the self-training data, no longer relying on keywords, thereby eliminating false novelty such as "re-publishing with a different title" or "old data in a new package".

[0121] If the second confirmation fails, the basic novelty score is set to 0.

[0122] Factor scoring S4 by relative time difference Exponential decay is employed to ensure that the novelty value of newly emerging intelligence rapidly decreases over time. The calculation formula is as follows: .in, and The meaning and values ​​of are as described above.

[0123] By employing a four-level cascade process of "keyword matching → semantic comparison → secondary confirmation by large model → time decay," we eliminate the repackaging of old news and ensure that truly first-time technological, policy, or market signals receive high novelty scores, while avoiding misjudgments caused by time drift.

[0124] In some implementations, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including:

[0125] The intelligence to be evaluated is parsed into event tuples of <subject, predicate, object, time>.

[0126] Input the event tuples into the dynamic system event graph to update the graph structure and edge weights;

[0127] The updated dynamic system event graph is inferred using a graph neural network, and the system disturbance index is output as the factor score corresponding to the pattern disturbance factor.

[0128] Dynamic system event graphs are a type of temporal knowledge graph that possesses the following three core mechanisms:

[0129] 1. Configurable time window: Only retain industry events within the predetermined duration (duration is set according to industry needs);

[0130] 2. Weight decay: Edge weights decay exponentially day by day, and the impact of old events naturally fades away;

[0131] 3. Incremental Updates: New intelligence is added to the database in real time, and the graph structure is updated locally as needed, without the need for a full reconstruction.

[0132] These three mechanisms work together to provide graph neural networks with a "snapshot of the current situation," enabling the same event to receive differentiated risk scores under different macroeconomic environments.

[0133] The composition and operation of a dynamic system event diagram are as follows:

[0134] 1. Nodes, including: policy entities (central bank, ministries, exchanges), key enterprises / organizations (OPEC, power grids, wafer fabs), and infrastructure / regions (Port of Hormuz, Shanghai nickel warehouses, New York Federal Reserve).

[0135] 2. Edges are generated from event tuples (subject, predicate, object, time) parsed from intelligence, for example, "Federal Reserve - Interest Rate Hike - Federal Funds Rate - 20XX - XX". Pairs with the same node and the same predicate are merged into one edge, and a list of timestamps is retained.

[0136] 3. Edge weight (0-1 floating point), its calculation formula is as follows: .in, This represents the intensity of the event, for example, an interest rate hike of 1.0, a statement of intent of 0.2, and sanctions of 0.8. The edge weights decay exponentially. λ = 0.35 / 24 h, corresponding to a half-life of approximately 2 days. =Current date - Last event date (days). Reset when a new event is added to the database. And clear .

[0137] Dynamic mechanism:

[0138] Incremental insertion: New information is entered into the database → tuples are parsed → edges are added / weights are updated.

[0139] Scrolling window: Only retains events within the most recent preset time window, and automatically deletes older events.

[0140] Calculation output:

[0141] After each update, the graph neural network (GNN) runs a forward propagation on the latest graph, outputs the risk embedding of each node, and takes the cosine distance of the target node pair as the system disturbance index, which is used as the factor score S5 corresponding to the pattern disturbance factor.

[0142] By using a closed loop of event tuples, dynamic graphs, and GNNs, the degree of disturbance of intelligence to the balance of power can be quantified in real time, enabling dynamic assessment of the degree of disturbance and dynamic adjustment of intelligence priority ranking. This ensures that high-value intelligence during the period of power transition is automatically placed at the top, significantly improving the timeliness and hit rate of intelligence optimization.

[0143] In some implementations, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including:

[0144] The intelligence to be evaluated is matched with the technology evolution dictionary to identify evolution information, which includes technical entities and technical action words in the same sentence, as well as the inverse document frequency of the technical entities;

[0145] Dependency parsing is performed on each statement containing the evolution information to obtain the co-occurrence strength score of the evolution information;

[0146] Based on the co-occurrence intensity scores of all the evolution information and the inverse document frequency of the corresponding technical entity, the factor score corresponding to the technical evolution signal factor is obtained.

[0147] The inverse document frequency is obtained by smoothing the total number of historical documents and the number of documents containing the entity by 1, and then taking the logarithm.

[0148] Dependency resolution is used to determine the number of shortest syntactic path edges between technical action words and technical entities within the same statement. A smaller number of edges indicates higher co-occurrence strength. The co-occurrence strength score is jointly determined by the dependency arc length penalty and the role weight of the technical action word. Specifically, the calculation formula is as follows: Among them, the dependent arc length penalty .in, This represents the shortest dependency arc from the technical action term to the technical entity; only those ≤3 are counted, and those >3 are considered irrelevant. The role weights of technical action terms are preset by domain experts and can be fine-tuned online. For example, "first release, mass production" is assigned 1.0, "upgrade, improvement" is assigned 0.8, and "application, deployment" is assigned 0.6.

[0149] In some implementations, the same statement is considered the same clause. The same clause refers to a text segment bounded by commas, semicolons, periods, exclamation marks, question marks, and their following capital letters, and which does not contain coordinating conjunctions. Co-occurrence strength scores are calculated only when the technical entity and the technical action word are in the same clause. Cross-clause phrases are considered irrelevant, and their co-occurrence strength scores are set to 0. This is because words within the same clause share the predicate core, resulting in semantic focus, while cross-clause phrases often involve topic shifts such as "discussing technology A first, then action B," reducing the causal confidence of the "technology-action" relationship.

[0150] The formula for calculating factor score S6 is: .in, Indicates the number of evolving information items in the intelligence. It indicates the order of evolution information.

[0151] The technology evolution signal factor, through the co-occurrence of "technical entities in the same clause + technical action words" and superimposed with TF-IDF weights, restores technical intelligence from "text characters" to "technical events," thereby accurately capturing breakthrough and scarce signals, capturing the value of "technological advancement," and quantifying the degree of "technological advancement." Integrating factor scoring S6 into value assessment fills the gap in traditional intelligence configuration schemes' insensitivity to technical intelligence, automatically prioritizing revolutionary technological intelligence in the optimal ranking process.

[0152] Step S108 involves a confidence assessment of each factor score, including: data integrity check, stability analysis of the calculation process, and verification of the reasonableness of the results.

[0153] In some implementations, a confidence level assessment is performed on each of the factor scores to obtain the corresponding confidence level, including:

[0154] The confidence level of the factor score is obtained by multiplying the data integrity sub-indicator, the calculation stability sub-indicator, and the result reasonableness sub-indicator based on the factor score. The method for obtaining the data integrity sub-indicator includes:

[0155] Based on the list of required fields corresponding to the factor scores, an existence check is performed on each of the required fields to obtain the missing rate of each required field;

[0156] The missing value is obtained by weighted summation based on the missing rate of all the required fields and their corresponding key weights;

[0157] The calculation formula for the data integrity sub-index is: Data integrity sub-index = 1 - missing degree;

[0158] The method for obtaining the stability sub-index includes:

[0159] After applying a slight perturbation to the same input data, the factor score is calculated k times to obtain k score results. The coefficient of variation (CV) is calculated, and max(0,1-CV) is used as the sub-index of computational stability.

[0160] The method for obtaining the result reasonableness sub-index includes:

[0161] The factor scores are input into a pre-trained isolated forest anomaly detection model to obtain reasonable probabilities, which are then used as a sub-index of the reasonableness of the results.

[0162] Existence checks include determining if the field object exists and is not null or entirely NaN. For each required field, an existence check is performed. If the field object is None, an empty string, an empty set, or entirely NaN, the field is considered missing and the missing rate is recorded as 1; otherwise, the missing rate is recorded as 0.

[0163] The magnitude of the slight disturbance is much smaller than the operational tolerance error, ensuring that the operational interpretation of the same intelligence remains unchanged, and is only used to detect the numerical stability of the calculation process.

[0164] Mild disturbances include any one or a combination of the following:

[0165] 1. Apply Gaussian noise with a relative amplitude not exceeding ±1% to the input numerical fields. , and will Cut to [ 0.03, +0.03].

[0166] 2. For the embedded vector or continuous features, add independent Gaussian noise element by element. ;

[0167] 3. For the neural network model, during the inference phase, set the Dropout layer to 0.1 and repeat the forward computation;

[0168] 4. For discrete text input, randomly mask no more than 2% of the tokens before encoding.

[0169] After perturbation, perform scoring k≥5 times to obtain... Then calculate CV = σ / μ. Calculate the stability sub-indices. And clamp to [0,1].

[0170] The results showed that the rationality sub-index passed the pre-trained isolated forest anomaly detection model, which is dedicated to each factor. Obtain, model Dedicated to S i The model uses S i The historical normal scores are used as training samples. The historical results of the same factor scores in the past 24 months are taken, and only ≥10,000 samples that have been manually verified as "normal" are retained.

[0171] During online inference, the current factor score and its corresponding input vector are input into the model, for example, a three-dimensional input vector. ,in, This represents the data integrity sub-index, outputs the normal probability p, and uses p as the result reasonableness sub-index. Furthermore, when p is below the threshold, the confidence level of the rating factor is set to 0. It also reflects whether the "score itself" and the "accompanying quality" are outliers, avoiding misjudgment based on a single threshold.

[0172] The confidence level of factor scoring is assessed by jointly evaluating three dimensions: data integrity, computational stability, and result reasonableness, resulting in a confidence level C. i This enables automatic masking of incomplete or abnormal scores, ensuring that subsequent priority configuration results can be reproduced.

[0173] Step S108, correct the weight calculation formula: .in, The corresponding base weights are used. The base weights are adjusted and normalized in real time using confidence levels, which significantly compresses the weights of low-confidence factor scores, while maintaining the contribution of high-confidence factor scores close to their original weights. This achieves an adaptive adjustment of "enhancing credible dimensions and suppressing untrustworthy dimensions" before fusion, effectively reducing the probability of missorting due to noise or missing data, and improving the accuracy and robustness of intelligence priority ranking.

[0174] Step S110, the formula for calculating the overall score V: The modified and normalized weights are used to weight and fuse the scores of each factor. The comprehensive score V is monotonically comparable in the [0,1] interval and can be directly mapped to intelligence priority. This makes the ranking results highly consistent with the task requirements, significantly reduces the misranking rate, and supports fast ranking, providing an immediate and reliable quantitative basis for the optimal allocation of intelligence resources.

[0175] In some implementations, cross-dimensional consistency checks are performed to compare whether the scores of each factor are contradictory. If samples with "high scores but low credibility" or "extreme imbalance" are found, they are automatically downgraded or marked for manual review.

[0176] Specifically, if V > 0.8 and there exists a factor score Ci < 0.3, it is judged as "high-value suspicious" and a second-level review is triggered.

[0177] , If (Smax) If Smin > 0.7 and Smax > 0.8, it is considered "extremely unbalanced" and the output is reduced in weight. The overall score after weight reduction is... ,in, This is the penalty coefficient.

[0178] Based on the actual operation logs and third-party manual evaluation reports for 30 days after the system went online, the Spearman rank correlation coefficient between the comprehensive score of the method S100 and the score of the human experts reached 0.83, which is 29% higher than the traditional linear weighted method. The ranking is highly consistent with the expert judgment, the useful information rate of the selected high-value intelligence is increased by 26%, the proportion of inflated intelligence is significantly reduced, and the optimal allocation of intelligence resources is achieved.

[0179] The method S100 integrates technologies such as deep learning, knowledge graphs, and graph neural networks to construct a complete evaluation system that includes six dimensions: semantic relevance, information popularity, authority, novelty, pattern disturbance degree, and technological evolution signal. Compared with the existing technologies with 1-3 dimensions, the evaluation completeness is improved by more than 200%, which can comprehensively reflect the multiple value characteristics of intelligence in professional fields and achieve accurate quantitative evaluation of intelligence value.

[0180] The complete four-layer architecture of method S100 is as follows: Figure 2 As shown, the system includes a user interface layer, a business logic layer, a data access layer, and an infrastructure layer. The diagram also illustrates the call relationships between these layers.

[0181] Example 2

[0182] Query context: The latest breakthrough in the Federal Reserve's digital monetary policy.

[0183] Intelligence pending evaluation: The Federal Reserve's digital dollar (FedCoin) successfully completed its first smart contract settlement test in a retail payment scenario on [Date], marking a crucial step for global central bank digital currencies from the wholesale to the consumer level. This test employed privacy computing and a programmable monetary framework.

[0184] Valuation process:

[0185] S1, Preprocessing stage:

[0186] Entity identification: FedCoin, smart contracts, retail payments, privacy computing.

[0187] Knowledge Graph Enhancement: FedCoin {Type: CBDC, Innovation: Programmability, Development Stage: Pilot}.

[0188] S2, Factor score calculation for each factor:

[0189] Semantic relevance factor: S1=0.85 (high relevance, high matching degree of monetary policy keywords);

[0190] Information popularity factor: S2 = (50000 + 2×200) / (100 + 0.1×1) = 502.0 / 100.1 = 0.92 (high popularity);

[0191] Authority factor: S3 = 0.90 (Official information, high credibility);

[0192] Novelty factor: S4 = 0.95 ("First-time" keyword, extremely high novelty);

[0193] Disruption factor: S5 = 0.70 (related to the global race for central bank digital currencies);

[0194] Technology evolution signal factor: S6 = 0.88 (breakthrough in core technologies of the next generation payment system).

[0195] S3, Confidence assessment: The confidence level of each scoring factor is above 0.85.

[0196] S4, Dynamic Fusion:

[0197] The overall score V = (0.25×0.85×0.90 + 0.15×0.92×0.88 + 0.20×0.90×0.92 +0.15×0.95×0.87 + 0.15×0.70×0.85 + 0.10×0.88×0.86) / normalization factor = 0.843, classifying it as high-value intelligence and prioritizing its submission to the intelligence-assisted decision-making system interface. Furthermore, a six-dimensional radar chart and confidence level watermark will be added, and a briefing (Word / PPT) will be generated to further support decision-making.

[0198] Example 3

[0199] This disclosure provides a multi-factor value assessment device 1000 for the optimal allocation of intelligence resources.

[0200] The evaluation device 1000 may include corresponding modules that execute one or more steps of the flowchart described above for the multi-factor value evaluation method for optimal allocation of intelligence resources. 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.

[0201] Specifically, such as Figure 3 As shown, the evaluation device 1000 includes:

[0202] The context acquisition module 1002 is used to acquire the query context corresponding to the intelligence-assisted decision-making task.

[0203] The multi-factor scoring module 1004 is used to respond to the query context and score the intelligence to be evaluated in parallel on multiple intelligence value scoring factors to obtain the factor score corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated.

[0204] The confidence assessment module 1006 is used to assess the confidence of each factor score to obtain the corresponding confidence level.

[0205] The correction coefficient acquisition module 1008 is used to correct the corresponding basic weights based on the confidence level of each factor score to obtain the corresponding corrected weights.

[0206] The preferred module 1010 is used to obtain a comprehensive score of the intelligence to be evaluated based on all the factor scores and their modified weights, and the comprehensive score is used to prioritize the intelligence to be evaluated.

[0207] 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 execute the above-described multi-factor value assessment method for optimal configuration of intelligence resources.

[0208] This disclosure also provides a readable storage medium storing execution instructions, which, when executed by a processor, are used to implement the above-described multi-factor value assessment method for optimal allocation of intelligence resources.

[0209] The hardware architecture of the evaluation device 1000, implemented using a processor-based hardware approach, 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. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0210] Bus 1100 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 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.

[0211] 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).

[0212] 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).

[0213] 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.

[0214] 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.

[0215] 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.

[0216] 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.

[0217] 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 multi-factor value assessment method for the optimal allocation of intelligence resources, characterized in that, The method includes: Obtain the query context corresponding to the intelligence-assisted decision-making task; In response to the query context, the intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated. The confidence level of each factor score is assessed to obtain the corresponding confidence level; Based on the confidence level of each factor score, the corresponding base weights are adjusted to obtain the corresponding adjusted weights; Based on all the factor scores and their adjusted weights, a comprehensive score is obtained for the intelligence to be evaluated, and the comprehensive score is used to prioritize the intelligence to be evaluated.

2. The method as described in claim 1, characterized in that, The intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors, including: Extract at least one entity and its initial attributes from the intelligence; The entity and its initial attributes are matched with the knowledge graph to obtain the extended attributes of the entity, wherein the extended attributes include the corresponding initial attributes; Generate structured vectors based on all entities and their extended attributes; Based on the structured vector, the intelligence to be evaluated is scored in parallel on the semantic relevance factors.

3. The method as described in claim 1, characterized in that, For each of the aforementioned factor scores, a confidence level assessment is performed to obtain the corresponding confidence level, including: The confidence level of the factor score is obtained by multiplying the data integrity sub-indicator, the calculation stability sub-indicator, and the result reasonableness sub-indicator based on the factor score. The method for obtaining the data integrity sub-indicator includes: Based on the list of required fields corresponding to the factor scores, an existence check is performed on each of the required fields to obtain the missing rate of each required field; The missing value is obtained by weighted summation based on the missing rate of all the required fields and their corresponding key weights; The calculation formula for the data integrity sub-index is: Data integrity sub-index = 1 - missing degree; The method for obtaining the stability sub-index includes: After applying a slight perturbation to the same input data, the factor score is calculated k times to obtain k score results. The coefficient of variation (CV) is calculated, and max(0,1-CV) is used as the sub-index of computational stability. The method for obtaining the result reasonableness sub-index includes: The factor scores are input into a pre-trained isolated forest anomaly detection model to obtain reasonable probabilities, which are then used as a sub-index of the reasonableness of the results.

4. The method as described in claim 1, characterized in that, The intelligence value scoring factors also include: a novelty factor, used to determine the degree of novelty of the intelligence to be evaluated relative to historical data; The intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including: The intelligence to be evaluated is matched with the novel keyword library to identify novel keywords and obtain the fragment containing the novel keywords as the first candidate novel fragment; If the novel keyword is not identified, obtain the maximum semantic similarity between each sentence of the intelligence to be evaluated and the historical intelligence database. If the minimum value of the maximum semantic similarity is less than the first threshold, obtain the second candidate novel fragment according to the corresponding sentence. Input the first candidate novel fragment or the second candidate novel fragment into the large language model, output the novelty probability, and if the novelty probability is not less than the second threshold, assign a base novelty score of 1.0, and perform exponential decay according to the relative time difference to obtain the factor score corresponding to the novelty factor.

5. The method as described in claim 1, characterized in that, The intelligence value scoring factors also include: the pattern disturbance factor, which is used to assess the potential impact of the intelligence to be assessed on the macro-system pattern; The intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including: The intelligence to be evaluated is parsed into event tuples of <subject, predicate, object, time>. Input the event tuples into the dynamic system event graph to update the graph structure and edge weights; The updated dynamic system event graph is inferred using a graph neural network, and the system disturbance index is output as the factor score corresponding to the pattern disturbance factor.

6. The method as described in claim 1, characterized in that, The intelligence value scoring factors also include: a technology evolution signal factor, used to assess the value level of technology evolution in the intelligence to be evaluated; The intelligence to be evaluated is scored in parallel on multiple intelligence value scoring factors to obtain factor scores corresponding to each intelligence value scoring factor, including: The intelligence to be evaluated is matched with the technology evolution dictionary to identify evolution information, which includes technical entities and technical action words in the same sentence, as well as the inverse document frequency of the technical entities; Dependency parsing is performed on each statement containing the evolution information to obtain the co-occurrence strength score of the evolution information; Based on the co-occurrence intensity scores of all the evolution information and the inverse document frequency of the corresponding technical entity, the factor score corresponding to the technical evolution signal factor is obtained.

7. The method as described in claim 1, characterized in that, The method further includes: Obtain the intelligence type label of the intelligence to be evaluated; Based on the intelligence type label, the base weight of at least one of the intelligence value scoring factors is adjusted.

8. A multi-factor value assessment device for the optimal allocation of intelligence resources, characterized in that, The device includes: The context acquisition module is used to acquire the query context corresponding to intelligence-assisted decision-making tasks. A multi-factor scoring module is used to respond to the query context and score the intelligence to be evaluated in parallel on multiple intelligence value scoring factors to obtain the factor score corresponding to each intelligence value scoring factor. The intelligence value scoring factors include: a semantic relevance factor, used to evaluate the semantic matching degree between the intelligence to be evaluated and the query context; an information popularity factor, used to evaluate the dissemination popularity of the intelligence to be evaluated within the current time window; and an authority factor, used to evaluate the credibility of the source of the intelligence to be evaluated. The confidence assessment module is used to assess the confidence of each factor score and obtain the corresponding confidence level. The correction coefficient acquisition module is used to correct the corresponding basic weights based on the confidence level of each factor score to obtain the corresponding corrected weights. The optimization module is used to obtain a comprehensive score of the intelligence to be evaluated based on all the factor scores and their modified weights, and the comprehensive score is used to prioritize the intelligence to be evaluated.

9. An electronic device, characterized in that, 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 1-7.

10. A readable storage medium, characterized in that, 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 1-7.

Images