Method and apparatus for assessing vulnerability information field drift and stability based on multiple data sources

By unifying the field space and timeline, the field drift and stability of multi-source vulnerability information are quantified, solving the problem of lack of systematic quantitative evaluation in existing technologies. This enables unified modeling and interpretable evaluation of multi-source vulnerability information, improving the accuracy of vulnerability intelligence fusion and security early warning.

CN123122881APending Publication Date: 2026-11-17ZHEJIANG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611190280.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-06
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies lack a systematic quantitative assessment of field drift and stability in multi-source vulnerability information processing, leading to downstream systems incorrectly adopting delayed or significantly biased field values, which affects the accuracy of vulnerability intelligence fusion, vulnerability knowledge graph construction, attack surface management, and security early warning.

Method used

This paper presents a method for assessing vulnerability information field drift and stability based on multiple data sources. By unifying the field space and timeline, it quantifies the time-sensitivity and structural drift of fields and outputs interpretable assessment results, including field-level risk scores, vulnerability cluster-level stability risk scores, and source responsibility ranking.

Benefits of technology

It achieves unified modeling of multi-source vulnerability information, comprehensively characterizes field stability, supports multiple field types, and generates interpretable evaluation results, which facilitates manual review and automated processing, and improves the accuracy of vulnerability intelligence fusion, vulnerability source quality monitoring, automated early warning, and security operation analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122881A_ABST
    Figure CN123122881A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of cyberspace security technology and discloses a method and apparatus for assessing the drift and stability of vulnerability information fields based on multiple data sources. For original vulnerability information entries from multiple data sources, a unified processing chain is established from the stages of field standardization, cross-source alignment, unified timeline construction, field validity determination, time lag drift quantification, structural divergence quantification, risk fusion, and interpretable output. It can directly address original vulnerability information entries from multiple data sources, and under a unified field space and timeline, uniformly quantify the time lag and structural drift of fields, and output interpretable assessment results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of cyberspace security technology, particularly the fields of vulnerability intelligence processing, multi-source heterogeneous data fusion, data quality analysis, and vulnerability information credibility assessment. Specifically, it relates to a method and apparatus for assessing the drift and stability of vulnerability information fields based on multiple data sources. Background Technology

[0002] As the number of publicly disclosed vulnerabilities continues to grow, various types of data sources have gradually emerged, encompassing vulnerability disclosure, tracking, remediation, and exploitation analysis. These include national-level vulnerability databases, internationally recognized vulnerability databases, open-source community vulnerability bulletin sources, vendor security bulletin sources, third-party threat intelligence platforms, software package ecosystem vulnerability databases, and code hosting platform security bulletin sources. These different sources often publish or maintain corresponding information entries around the same vulnerability; therefore, in practical applications, it is frequently necessary to integrate, align, and assess the credibility of vulnerability information from multiple sources.

[0003] However, due to differences in the construction goals, update mechanisms, collection methods, data models, governance strategies, and downstream use cases of different data sources, the information entries from different sources for the same vulnerability often exhibit significant differences. These differences are not only reflected at the data schema level, such as field naming, field structure, and field nesting levels, but also in aspects such as field content completeness, field appearance time, update speed, expression granularity, semantic accuracy, and reference scope.

[0004] Existing technologies for processing multi-source vulnerability information typically revolve around tasks such as vulnerability deduplication, vulnerability matching, field extraction, vulnerability knowledge graph construction, and vulnerability entity alignment. While these methods can address cross-source correlation issues to some extent, they still have significant shortcomings in assessing field drift and stability for original multi-source vulnerability information entries.

[0005] First, existing methods typically focus more on identifying whether a vulnerability is the same, rather than systematically quantifying the differences in fields across different sources for the same vulnerability. Even when some methods perform field consistency comparisons, they often concentrate on comparing static final values, lacking modeling of the dynamic evolution of fields from missing to available, from coarse to refined, and from errors to revisions.

[0006] Secondly, there are significant differences in the speed at which different sources update fields. For example, one source might provide severity information when the vulnerability is first disclosed, but only later add information on affected versions and fixes; another source might provide a fix earlier, but only later add standard vulnerability identifiers; and some sources fail to provide complete fix information throughout the entire observation window. Comparing only the final field values ​​fails to reflect the true differences in timeliness between sources.

[0007] Furthermore, the structural differences between different types of fields are not the same. Severity fields are typically numeric or ordered enumeration types, description fields are text types, reference fields are set types, and affected range and repair fields are often range or composite structures. Without a difference metric function that is compatible with the field types, it is difficult to evaluate the consistency and stability of multiple types of fields within a unified framework.

[0008] Furthermore, existing technologies often lack interpretable output. Even when a conclusion is reached that a vulnerability entry contains conflicts or inconsistencies, it often fails to specify: which field caused the conflict, which source is lagging in its update, which two sources have the greatest disagreement, or the extent of responsibility a particular source bears for the overall instability. This lack of information significantly impacts the efficiency of subsequent manual review, automated arbitration, and data governance.

[0009] In scenarios such as vulnerability intelligence fusion, vulnerability knowledge graph construction, attack surface management, vulnerability remediation priority determination, and automated security alerts, the stability and reliability of multi-source vulnerability information are of great significance. If field drift and stability cannot be quantitatively assessed, downstream systems are prone to incorrectly adopting lagging or significantly biased field values, leading to distorted alerts, biased remediation recommendations, incorrect asset exposure analysis, and even misjudgments in security operations. Summary of the Invention

[0010] The purpose of this invention is to provide a method and apparatus for assessing the drift and stability of vulnerability information fields based on multiple data sources. This method can directly target original vulnerability information entries from multiple data sources, uniformly quantify the time-sensitivity drift and structural drift of fields under a unified field space and timeline, and output interpretable assessment results.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] Firstly, a method for assessing vulnerability information field drift and stability based on multiple data sources is provided, including:

[0013] Obtain raw vulnerability information entries from multiple data sources, and perform field unification on each raw vulnerability information entry to form a standardized field representation on a unified field set;

[0014] Based on the vulnerability identifier field in the standardized field representation, original vulnerability information entries pointing to the same vulnerability from different data sources are merged into the same vulnerability cluster; for the remaining original vulnerability information entries, the alignment confidence between any two original vulnerability information entries is calculated, and vulnerability clusters are merged based on the alignment confidence;

[0015] For each vulnerability cluster, the release time and update time of the original vulnerability information entries from each source are collected to construct a unified timeline and obtain an ordered unified timeline. Based on the unified timeline, a field status time function for each source is established for any field in the unified field set.

[0016] Using the time in the ordered unified timeline, and based on the field status time function of each source, determine the first available time of the field in each source, and obtain the normalized cross-source time lag drift, the earliest valid source set, and the latest valid source set of the field;

[0017] Aggregate identical fields from the same data source in the vulnerability cluster, perform pairwise comparisons on aggregated fields from different sources, calculate the structural divergence degree, and take the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair.

[0018] The time lag drift and structural divergence are fused according to field weight and dimension weight to obtain field-level risk score and vulnerability cluster-level stability risk score;

[0019] The output includes a structured assessment report containing field-level risk scores, vulnerability cluster-level stability risk scores, the earliest valid source set, the latest valid source set, and the largest divergence source pair.

[0020] Several alternative methods are provided below, but they are not intended as additional limitations on the overall solution above. They are merely further additions or optimizations. Provided there are no technical or logical contradictions, each alternative method can be combined individually with respect to the overall solution above, or multiple alternative methods can be combined with each other.

[0021] Preferably, the step of calculating the alignment confidence between any two original vulnerability information entries for the remaining original vulnerability information entries, and merging vulnerability clusters based on the alignment confidence, includes:

[0022] Based on the vulnerability identifier field, calculate the vulnerability identifier similarity between two original vulnerability information entries;

[0023] Calculate the description similarity between two original vulnerability information entries based on the description field;

[0024] Calculate the temporal proximity between two original vulnerability information entries based on the release time field and the update time field.

[0025] Calculate the similarity of affected scope between two original vulnerability information entries based on the affected scope field;

[0026] Calculate the citation similarity between two original vulnerability information entries based on the citation field;

[0027] The weighted sum of vulnerability identifier similarity, description similarity, temporal proximity, affected scope similarity, and reference similarity is used as the alignment confidence between two original vulnerability information entries;

[0028] Two original vulnerability information entries with an alignment confidence score greater than the alignment threshold are grouped into the same vulnerability cluster.

[0029] Preferably, the step of establishing a field status time function from various sources for any field in a unified field set based on a unified timeline includes:

[0030] Prior to the publication time of the original vulnerability information entry, the fields in the original vulnerability information entry were in an unknown state;

[0031] Between the publication time of the original vulnerability information entry and the first update time, the fields in the original vulnerability information entry are in their initial state values;

[0032] In the original vulnerability information entry, at the... The time of the next update and the first Between the update times, the fields in the original vulnerability information entry are the first... The updated state value , Indicates the number of times the original vulnerability information entry has been updated.

[0033] Preferably, the step of determining the first available time of a field in each source based on the field status time function in the ordered unified timeline includes:

[0034] Take the time from the ordered unified timeline, substitute it into the field status time function, and get the field status value at the current time;

[0035] Determine the validity of the field's status value at the current time based on the set of valid value ranges for the field;

[0036] The current time when the state value first becomes valid is taken as the first available time; if there is no time in the entire ordered unified timeline that makes the state value valid, then the preset upper bound of time is taken as the first available time.

[0037] Preferably, the normalized cross-source time lag drift, the earliest valid source set, and the latest valid source set for the field include:

[0038] The difference between the latest first-access time and the earliest first-access time of a field on a vulnerability cluster is used as the cross-source time lag drift of the field on the vulnerability cluster.

[0039] The cross-source time lag drift is normalized to obtain the normalized cross-source time lag drift.

[0040] Combine one or more data sources that reach their first availability time earliest from all data sources to form the earliest valid source set;

[0041] Combine one or more data sources that reached their first availability time latest from all data sources to form the latest valid source set.

[0042] As a preferred embodiment, the calculation of standardized lag responsibility components is also included, including:

[0043] Construct a set of lagging responsible source pairs based on the Cartesian product of the earliest and latest valid source sets;

[0044]

[0045] In the formula, Indicate source In vulnerability cluster Middle field Standardization lag responsibility weighting For data source In vulnerability cluster Middle field The first available time, For data source In vulnerability cluster Middle field The first available time, For data source collection, Representation field In vulnerability cluster Cross-source time lag drift.

[0046] Preferably, the step of performing pairwise comparisons on the aggregated representations of fields from different sources, calculating the structural divergence degree, and taking the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair includes:

[0047] For the same vulnerability cluster, calculate the aggregated distance value between pairs of different data sources with the same field, and take the mean of all distance values ​​as the structural divergence degree;

[0048] The data sources of the two fields with the highest structural divergence are taken as the source pair with the highest structural divergence in the vulnerability cluster.

[0049] For a field within the same vulnerability cluster, calculate the distance between the aggregated representation under one data source and the aggregated representation under other data sources, and take the average of all distance values ​​as the source outlier of the field under the current data source.

[0050] Preferably, the field-level risk score is a fusion value that integrates time lag drift and structural divergence according to field weights and dimension weights;

[0051] The vulnerability cluster-level stability risk score is the sum of the field-level risk scores of all fields in the same vulnerability cluster.

[0052] As a preferred option, this also includes ranking the source responsibility within the vulnerability cluster, including:

[0053] The weighted sum of the standardized lag responsibility component and the source outlier is taken as the comprehensive responsibility score of the data source on the field in the vulnerability cluster;

[0054] The overall responsibility score of the data source in the vulnerability cluster is obtained by summing the comprehensive responsibility scores of all fields of the same data source in the vulnerability cluster.

[0055] The overall responsibility score is normalized to obtain the normalized source responsibility score of the data source on the vulnerability cluster;

[0056] All data sources are ranked from highest to lowest according to the normalized source responsibility score, resulting in the source responsibility ranking results for data sources within the vulnerability cluster.

[0057] As a preferred option, the method also includes comparing the stability risk scores of vulnerability clusters based on preset risk classification thresholds to obtain the risk level of the vulnerability cluster.

[0058] Secondly, a device for assessing vulnerability information field drift and stability based on multiple data sources is provided, including:

[0059] The data acquisition and standardization module is used to acquire raw vulnerability information entries from multiple data sources and perform field unification on each raw vulnerability information entry to form a standardized field representation on a unified field set.

[0060] The vulnerability alignment and clustering module is used to merge original vulnerability information entries pointing to the same vulnerability from different data sources into the same vulnerability cluster based on the vulnerability identifier field in the standardized field representation; for the remaining original vulnerability information entries, it calculates the alignment confidence between any two original vulnerability information entries and merges vulnerability clusters based on the alignment confidence;

[0061] The unified timeline construction module is used to collect the release and update times of the original vulnerability information entries from various sources for each vulnerability cluster, construct a unified timeline, and obtain an ordered unified timeline. Based on the unified timeline, a field status time function for each source is established for any field in the unified field set.

[0062] The field validity determination and time lag calculation module is used to determine the first available time of a field in each source based on the field status time function of each source, using the time in the ordered unified timeline, and to obtain the normalized cross-source time lag drift of the field, the earliest valid source set, and the latest valid source set;

[0063] The structural divergence calculation module is used to aggregate identical fields from the same data source in the vulnerability cluster, compare the aggregated fields from different sources pairwise, calculate the structural divergence degree, and take the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair.

[0064] The risk fusion module is used to fuse time lag drift and structural divergence according to field weights and dimension weights to obtain field-level risk scores and vulnerability cluster-level stability risk scores.

[0065] The explanation output module is used to output a structured assessment report that includes field-level risk scores, vulnerability cluster-level stability risk scores, the earliest valid source set, the latest valid source set, and the largest divergence source pair.

[0066] The method and apparatus for assessing vulnerability information field drift and stability based on multiple data sources provided by this invention have at least the following advantages compared with the prior art:

[0067] 1. It can directly process raw heterogeneous vulnerability information entries without having to unify all sources to the same schema in advance, thus achieving unified modeling of multi-source vulnerability fields.

[0068] 2. Simultaneously modeling time drift and structural drift, it considers both when a field becomes available and the differences in the final expression of the field, thus providing a more comprehensive characterization of field stability.

[0069] 3. Supports multiple field types, applicable to numeric, text, set, range, enumeration, and composite fields.

[0070] 4. It is interpretable and can generate structured evaluation results including vulnerability cluster-level stability risk score, stability score, vulnerability cluster-level risk level, field-level risk score, field-level contribution, the earliest and latest valid source set of field timeliness, the set of delayed responsibility source pairs, the field with the greatest divergence source pairs, source responsibility score, and source responsibility ranking results, which facilitates manual review, system arbitration, and downstream automated invocation.

[0071] 5. Applicable to various downstream scenarios, serving vulnerability intelligence fusion, vulnerability source quality monitoring, automated early warning, vulnerability knowledge graph construction, and security operation analysis. Specifically, in vulnerability intelligence fusion scenarios, conflicting fields can be automatically marked, downgraded, prioritized for selection, or have a queue for manual review generated based on field-level risk scores, the largest source of disagreement, and source responsibility scores. In vulnerability source quality monitoring scenarios, data sources with long-term update delays, high field missing rates, or significant structural disagreements can be identified based on source responsibility ranking results, and the crawling frequency, synchronization strategy, or quality alarm rules can be adjusted accordingly. In automated early warning scenarios, alarm confidence, alarm priority, and manual review triggering conditions can be adjusted based on vulnerability cluster-level stability risk scores and risk levels to avoid false alarms, missed alarms, or incorrect ranking due to delays or errors in a single source field. In vulnerability knowledge graph construction and asset exposure analysis scenarios, highly unstable fields can be filtered out based on field-level contribution and field stability risk, and trusted field values ​​can be prioritized, thereby reducing the probability of erroneous fields entering the knowledge graph, asset impact analysis, or remediation work order system. Attached Figure Description

[0072] Figure 1 This is a flowchart of the vulnerability information field drift and stability assessment method based on multiple data sources according to the present invention;

[0073] Figure 2 This is a schematic diagram of the structure of the vulnerability information field drift and stability assessment device based on multiple data sources according to the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0076] like Figure 1As shown, this embodiment provides a method for assessing the drift and stability of vulnerability information fields based on multiple data sources. For original vulnerability information entries from multiple data sources, a unified processing chain is established, encompassing field standardization, cross-source alignment, unified timeline construction, field validity determination, quantification of time lag drift, quantification of structural discrepancies, risk fusion, and interpretive output. This enables the assessment of the drift and stability of vulnerability information fields. Specifically, the method includes the following steps:

[0077] S1. Acquisition of original vulnerability information, field extraction and unified field standardization.

[0078] In this step, the raw vulnerability information entries are first obtained from at least two data sources. Data sources may include, but are not limited to, national-level vulnerability databases, internationally recognized vulnerability databases, open-source community vulnerability bulletin sources, vendor security bulletin sources, third-party intelligence platforms, software ecosystem vulnerability bulletin sources, and other data sources that can provide vulnerability information entries.

[0079] Data source constitutes a collection of sources:

[0080]

[0081] For any data source The original set of vulnerability information entries is represented as follows:

[0082]

[0083] In the formula, Indicate source The original collection of vulnerability information entries; Indicate source The Middle One original vulnerability information entry, Each original vulnerability information entry It can be a JSON object or a parsed nested key-value object, as long as the corresponding field values ​​can be extracted.

[0084] Because field definitions, naming conventions, hierarchy, and representation methods differ across sources, a unified set of fields needs to be constructed.

[0085]

[0086] In the formula, Represents a unified set of fields; It represents any field in the unified field set.

[0087] In a preferred embodiment of the present invention, the unified field set includes at least the following fields: vulnerability identifier field. Publication time field Update time field Description field Severity field Affected scope fields Repair information fields and referenced fields In other embodiments, the field may be expanded to include vulnerability type field, affected product field, exploitation status field, vendor field, and component field.

[0088] For any source Any original vulnerability information entry Its standardized field representation is defined as follows:

[0089]

[0090] In the formula, Original vulnerability information entry The corresponding standardized fields, Represents the original vulnerability information entry Middle field The corresponding standardized value is expressed as:

[0091]

[0092] In the formula, This indicates extracting fields from the original entries. Functions corresponding to primitive values; Indicates targeting a field The normalized function.

[0093] Different fields can use different normalization functions, as detailed in this example.

[0094] (1) Normalization of time field.

[0095] For both the publication time and update time fields, it is preferable to uniformly convert them to UTC timestamps or a unified ISO time representation. Let the original time value be... The normalization function is defined as follows:

[0096]

[0097] In the formula, This is a conversion function specifically designed for parsing UTC format time strings. If the original value includes a time zone, it will first perform a time zone conversion; if the original value lacks a time zone, it can be parsed according to the source's default time zone and then uniformly converted to UTC time.

[0098] (2) Describe field normalization.

[0099] For the description field, processing can be performed such as case normalization, noise reduction, stop word filtering, terminology standardization, and special character cleanup. Let the original description text be... The normalization function is defined as follows:

[0100]

[0101] In the formula, This is a conversion function specifically designed for parsing descriptive text. In this embodiment, one specific implementation is as follows:

[0102] To facilitate subsequent similarity calculations and distance calculations for text-based fields, the standardized description text is vectorized. Let the original description text of the original vulnerability information entry be... After processing by the description field normalization function, the standardized description text is obtained. :

[0103]

[0104] The standardized descriptive text is then input into a text encoding function to obtain a descriptive embedding vector. :

[0105]

[0106] in, This describes the field normalization function. The text encoding function can be implemented using TF-IDF vectorization, word vector aggregation, Doc2Vec, BERT, Sentence-BERT, or other text representation models that can output fixed-length semantic representations.

[0107] (3) Normalization of severity field.

[0108] For the severity field, it may take various forms such as severity level, CVSS score, CVSS vector, etc. If the original value is a level label, it can be mapped to an ordered enumeration or a standardized score:

[0109]

[0110] If the original value is the CVSS score, then numerical normalization can be performed:

[0111]

[0112] If the original value is a CVSS vector, the basic score can be parsed first, and then normalized. In the formula... This is a function for mapping grade labels.

[0113] (4) Normalization of the affected range field.

[0114] For fields affected by the scope of the issue, product name, software package name, component name, CPE representation, version range, etc., can be uniformly parsed into a set of standard objects. Let the standardization result be... ,but:

[0115]

[0116] in It can be represented as a product ( )-Version( Constraints on sets:

[0117]

[0118] In some implementations, it can also be broken down into a multi-level set of sub-attributes, such as a product set, a component set, and a version range set.

[0119] (5) Repair information field standardization.

[0120] For the fix information field, fix version number, fix patch link, commit hash, announcement identifier, etc. can be uniformly grouped into a fix object set:

[0121]

[0122] For example:

[0123]

[0124] In the formula, For the first fix version number, This is the second fix version number. For the first commit hash, This is the identifier for the first announcement.

[0125] (6) Normalize the referenced fields.

[0126] For reference fields, reference links, reference announcement links, patch links, analysis article links, etc., can be uniformly converted into a reference collection:

[0127]

[0128]

[0129] To support subsequent field difference calculations, field type functions can also be defined:

[0130]

[0131] in, Representation field Data types, Indicates scalar type. Represents text, Represents a set, Indicates the range of intervals. This indicates an enumeration with a limited type. This indicates a complex nested object type.

[0132] For example:

[0133]

[0134]

[0135]

[0136]

[0137] The output of this step is: all original vulnerability information entries from all sources are mapped to standardized field vectors on a unified field set, providing input for subsequent cross-source alignment and field stability analysis.

[0138] S2, Cross-Source Alignment and Vulnerability Cluster Construction.

[0139] After completing field extraction and standardization, it is necessary to align and merge entries from different data sources that point to the same vulnerability in order to build a vulnerability cluster.

[0140] Suppose that all standardized entries constitute the overall set of entries:

[0141]

[0142] For any two candidate original vulnerability information entries and Record it in the vulnerability identifier field. The standardized values ​​on are respectively and If its vulnerability identifier field can be directly matched, then when:

[0143]

[0144] In this case, it can be determined that the two belong to the same vulnerability.

[0145] However, in real-world scenarios, the following situations are common: some sources have not yet provided a vulnerability identifier field; some sources use non-uniform vulnerability numbers; a single entry contains multiple numbers or number aliases; and different sources do not have completely consistent identifiers for the same vulnerability.

[0146] Therefore, this invention further employs alignment confidence for cross-source alignment. For any two candidate original vulnerability information entries... and The alignment confidence is defined as:

[0147]

[0148] Among them, each weight parameter for:

[0149]

[0150] The definitions of each similarity metric are as follows.

[0151] 1. Similarity of vulnerability identifiers.

[0152]

[0153] In an extended implementation, this item can also be extended by combining an alias mapping table, ID prefix replacement rules, and multiple number intersection relationships.

[0154] 2. Describe the similarity.

[0155] Suppose that the description embedding vectors obtained after text encoding of the standardized description text of two entries are respectively and Then, the description similarity is defined as:

[0156]

[0157] In the formula, It is a small value used to place the denominator as 0.

[0158] 3. Temporal proximity.

[0159] To measure the proximity of two candidate items in the time dimension, an item is defined. The time base is ,entry The time base is Where an item has a standardized publication time field, its publication time is used as the time reference; if the publication time field is missing, the earliest update time is used as the time reference; if the update time is known, the first observation time of the item (the time the item was collected) is used as the time reference. Time proximity is then defined as:

[0160]

[0161] in This is the time decay coefficient.

[0162] 4. Similarity of the affected scope.

[0163] Set up entries and The standardized affected range sets obtained after standardization of the affected range field are as follows: and Where the set of standardized affected scopes includes at least the affected object identifier and the corresponding version constraint, then:

[0164]

[0165] 5. Citation similarity.

[0166] Let the reference sets of the two entries be respectively and ,but:

[0167]

[0168] Pick If a preset alignment threshold is used, then when:

[0169]

[0170] At that time, it was determined that the two entries pointed to the same vulnerability.

[0171] In other embodiments, a dual-threshold strategy may also be used:

[0172]

[0173]

[0174]

[0175] After the above processing, a set of vulnerability clusters is constructed:

[0176]

[0177] Any of the vulnerability clusters This represents a set of cross-origin entries that are determined to point to the same vulnerability instance.

[0178] For any vulnerability cluster Its source set is defined as:

[0179]

[0180] Meanwhile, the source coverage rate is defined as:

[0181]

[0182] This indicator can be used for subsequent risk correction and to assist in judging the credibility of results.

[0183] The output of this step is: a collection of vulnerability clusters. and its corresponding source set .

[0184] S3. Construction of a unified timeline and establishment of field status time functions.

[0185] After completing the vulnerability cluster construction, it is necessary to further establish a unified timeline for each vulnerability cluster in order to depict the temporal evolution of the field across different sources.

[0186] For vulnerability clusters Collect its source set By compiling the release and update times of the corresponding vulnerability entries from various sources, a unified timeline can be constructed:

[0187]

[0188] In the formula, Indicates a vulnerability cluster The set of time points corresponding to the unified timeline; Indicate source In vulnerability cluster The release date of the corresponding vulnerability entry; Indicate source In vulnerability cluster The corresponding vulnerability entry in the middle Next update time; Indicate source In vulnerability cluster The number of times the corresponding vulnerability entry is updated, and .

[0189] All time points in the unified timeline, including the publication time and update time of the corresponding entries from each source, are deduplicated and sorted in ascending order according to chronological order, resulting in:

[0190]

[0191] In the formula, Indicates a vulnerability cluster The unified timeline after deduplication and ascending order; Represents an ordered and unified timeline After deduplication and sorting, the first Each observation time point ; Represents an ordered and unified timeline The total number of time points in the timeline; unless otherwise specified below, the unified timeline refers to the ordered unified timeline. .

[0192] Subsequently, for any field in the unified field set , definition source In time The field status time function is:

[0193]

[0194] In the formula, Indicate source In an orderly and unified timeline Observation time point upper field The state-time function, where ; This indicates that the field was unavailable, unknown, or not disclosed at the corresponding point in time; Indicate source The corresponding field is located after the publication time and before the first update. The initial state value; Indicate source The corresponding entry is in After the next update time, the field The state value, where ; Represents an ordered and unified timeline Any observation point on the time scale.

[0195] If a particular implementation cannot obtain multiple update processes and can only obtain the release time and the final update time, a simplified form can be used:

[0196]

[0197] in, Indicate source The final update time of the corresponding vulnerability entry under this simplified implementation method; Indicates the field after the last update time. The final state value. In this simplified form... Similarly, it represents an orderly and unified timeline. Any observation point on the time, i.e. .

[0198] In a further implementation, to facilitate overall calculation and subsequent statistical analysis, a cluster-level field state tensor can also be defined:

[0199]

[0200] in, Indicates a vulnerability cluster Cluster-level field state tensors in the source, field, and time dimensions are used to reflect vulnerability clusters. The overall status of different sources and fields at different points in time.

[0201] In addition, fields can be defined. At a certain point in time Cross-source availability :

[0202]

[0203] in, Representation field The validity determination function is defined in step S4.

[0204] Further define fields Average availability over the entire ordered unified timeline for:

[0205]

[0206] The output of this step is: an ordered unified timeline. Field status time function and optional cluster-level field state tensor .

[0207] S4. Field validity determination and first availability time calculation.

[0208] After establishing the field status time function, it is necessary to further determine when a field reaches a valid state in different sources. Therefore, this invention defines a field validity determination function.

[0209] For any field The validity determination function is defined as follows:

[0210]

[0211] in, Representation field The set of valid value ranges.

[0212] In a preferred embodiment, the validity of different fields can be determined as follows:

[0213] 1. Description field: When the text length of the description field exceeds the threshold and is not empty, not template text, or invalid placeholder text, it is considered valid.

[0214] 2. Severity field: A field is considered valid if it contains a parsable severity level, CVSS score, or CVSS vector.

[0215] 3. Affected Scope Field: A field is considered valid if it contains at least one resolvable affected object identifier and its version constraint.

[0216] 4. Repair Information Field: A field is considered valid if it contains at least one of the following: repair version, repair patch link, repair commit hash, or repair announcement identifier.

[0217] 5. Referenced fields: A field is considered valid if it contains at least one non-empty and normalizable reference link or referenced object.

[0218] Based on the field validity determination function, define the source. In vulnerability cluster Middle field First available time for:

[0219]

[0220] If field If a valid state is not reached within the entire observation time window, then it is defined as follows:

[0221]

[0222] In the formula, This indicates the upper bound of the preset time, and Later than the orderly unified timeline The last observation point in the data, namely:

[0223]

[0224] Furthermore, fields can be defined. Convergence time That is, the earliest time point when the preset available ratio threshold is reached:

[0225]

[0226] in For fields The convergence threshold is used to represent the field. In vulnerability cluster The criteria for determining when the source set reaches the preset usable ratio.

[0227] The output of this step is: each source In vulnerability cluster Middle field First available time And optional field convergence time .

[0228] S5, Field time lag drift calculation.

[0229] After obtaining the first availability time of each source field, the cross-source time lag drift of the field within the vulnerability cluster is further calculated.

[0230] For any vulnerability cluster and fields Cross-source time lag drift is defined as:

[0231]

[0232] In the formula, Representation field In vulnerability cluster Cross-source time lag drift, this value represents the field The time span between the earliest and latest availability from different sources.

[0233] To ensure comparability of lag values ​​across different fields, this invention normalizes them. Exponential normalization is preferred.

[0234]

[0235] In the formula, Representation field In vulnerability cluster Normalized cross-source time lag drift on the above For fields The time sensitivity coefficient.

[0236] In another embodiment, linear truncation normalization may also be used:

[0237]

[0238] in, For fields The reference hysteresis threshold.

[0239] To pinpoint the specific source of responsibility for timeliness, further define the fields. In vulnerability cluster The earliest valid source set in and the latest valid source set Considering that multiple sources may reach a valid state for the first time at the same time, both the earliest and latest valid sources are represented in set form:

[0240]

[0241]

[0242] When the minimum or maximum value is obtained from only a single source, the earliest valid source set or the latest valid source set degenerates into a single-element set.

[0243] This allows the construction of fields. The source of delayed responsibility is represented by the set:

[0244]

[0245] In the formula, Representation field In vulnerability cluster The set of lagging responsibility source pairs consists of the earliest valid source set and the latest valid source set. This represents the Cartesian product in set operations.

[0246] In addition, to measure the source In the field The degree of timeliness responsibility can also be defined by defining standardized delayed responsibility components:

[0247]

[0248] In the formula, Indicate source In vulnerability cluster Middle field Standardization lag responsibility weighting To prevent smooth terms with a denominator of zero.

[0249] In other embodiments, if a certain source The effective state was not reached within the observation time window, i.e. Then, additional penalties for failure can be added to enhance the timeliness of accountability. :

[0250]

[0251] in, For fields Failure penalty parameters, This is an indicator function.

[0252] In a further implementation, fields may also be defined. Convergence lag:

[0253]

[0254] in, For the fields defined in step S4 In vulnerability cluster Reaching the convergence threshold The convergence time; Used to describe fields The time between the first availability time corresponding to the earliest valid source set and the time between reaching the preset group availability ratio.

[0255] The output of this step is: Field Time Lag Drift. Normalized time lag drift The earliest valid source set Latest valid source set Delayed responsibility sources on the set and optional standardized lag responsibility components Strengthen the sense of timeliness and responsibility and convergence lag .

[0256] S6, Field aggregation represents the construction and structural divergence calculation.

[0257] In addition to time lag analysis, this invention also quantitatively models the differences in the final expression structure of fields across different sources.

[0258] For fields , definition source In vulnerability cluster Field aggregation in the code is represented as follows:

[0259]

[0260] in, Indicate source In vulnerability cluster Middle field The aggregate representation; without causing confusion, it can be abbreviated as . This represents the aggregate function for the field. Different aggregation methods can be used depending on the characteristics of different fields.

[0261] 1. Most recent valid value:

[0262]

[0263]

[0264] 2. Union of historical effective values:

[0265] For collection-type fields, historical union can be used:

[0266]

[0267] 3. Time-weighted aggregate value:

[0268] For fields that change multiple times within the observation window and require comprehensive consideration of both historical and recent states, the source can be determined first. In vulnerability cluster Middle field The set of candidate valid field values ​​is defined as:

[0269]

[0270] in, Representation field In the source and vulnerability clusters The set of valid field values ​​that have appeared within the observation time window; any element in this set can be used as a candidate value for time-weighted aggregation.

[0271] Based on this, the time-weighted aggregate value is defined as:

[0272]

[0273] in, Represents the set of candidate valid field values Any candidate value in; Indicates a point in time The weight can be determined based on the time relevance, duration of status, or business importance; This is an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. The resulting time-weighted aggregate value is the field status value that appears most frequently within the observation time window.

[0274] In practical implementation, the appropriate field aggregation method can be selected based on the field type. For fields such as severity, description, and vulnerability status, where the final revision result is the primary comparison object, the most recent valid value is preferred as the field aggregation representation. For collection-type fields with cumulative characteristics, such as vulnerability identifier, reference, CWE, vulnerability tag, patch link, and affected object set, the union of historical valid values ​​is preferred as the field aggregation representation. For fields that change frequently within the observation window and require consideration of duration or time weight, time-weighted aggregation values ​​can be used as the field aggregation representation. The above aggregation methods can also be combined or switched according to the business scenario.

[0275] After obtaining the aggregate representation, it is necessary to define corresponding field change distance functions for different field types. Representation field The change in distance function.

[0276] (1) Distance between numeric fields:

[0277] For numeric fields such as CVSS scores, the definition is as follows:

[0278]

[0279] In the formula, For fields The preset value range width.

[0280] (2) Distance of enumerated fields:

[0281] For enumerated fields such as severity level, simple differences can be defined:

[0282]

[0283] Ordinal hierarchy differences can also be defined:

[0284]

[0285] in For enumerating level mapping functions, This represents the total number of levels.

[0286] (3) Distance between aggregate fields:

[0287] For collection-type fields such as reference sets, vulnerability tag sets, and CWE sets, Jaccard distance is used:

[0288]

[0289] (4) Distance between text fields:

[0290] For the description field, cosine distance based on the embedding vector is used:

[0291]

[0292] In a further implementation, mixed text distance can also be used:

[0293]

[0294] in, Indicates semantic distance. Indicates lexical distance.

[0295] (5) Version range field distance:

[0296] For the affected range field and the patch version field, if their standardized representation is a set of versions or a set of version ranges... and Then the definition is:

[0297]

[0298] (6) Distance between composite structure fields:

[0299] For composite fields consisting of multiple sub-attributes, a weighted composite distance can be defined:

[0300]

[0301] in:

[0302]

[0303] After completing the field distance definition, the field In vulnerability cluster Structural divergence on Defined as:

[0304]

[0305] To further locate the field In vulnerability cluster The biggest source of structural conflicts in the middle is the definition of fields. The biggest source of disagreement is:

[0306]

[0307] in, Representation field In vulnerability cluster The source of the greatest structural divergence in the middle.

[0308] Additionally, the source can be defined. In the field Outlier of source:

[0309]

[0310] The output of this step is: Field Aggregation Representation Structural Disagreement The biggest source of disagreement is on and source outlier .

[0311] S7. Risk fusion and vulnerability cluster-level stability risk score calculation.

[0312] After calculating the time lag drift and structural divergence, the two are merged to form a field-level risk score and a vulnerability cluster-level risk score.

[0313] For any field Its field-level risk score is defined as:

[0314]

[0315] In the formula, For fields In vulnerability cluster Field-level risk scoring on the top For fields Importance weights; For fields Structural divergence weights; For fields The time lag weight.

[0316] In one implementation, there is:

[0317]

[0318] and:

[0319]

[0320] Calculate the vulnerability cluster-level stability risk score based on the field-level risk scores of all fields:

[0321]

[0322] In the formula, For vulnerability clusters Vulnerability cluster-level stability risk score.

[0323] In some embodiments, to account for the impact of field missing rate, a field missing rate can also be defined:

[0324]

[0325] In the formula, For fields In vulnerability cluster Missing rate of the above field.

[0326] At this point, the field-level risk scoring can be expanded to:

[0327]

[0328] In the formula, This is for the missing penalty weight.

[0329] In a further implementation, to reflect the impact of insufficient source coverage on the assessment results, it can be based on the source coverage rate. The vulnerability cluster-level stability risk score has been revised:

[0330]

[0331] In the formula, This is the coverage penalty coefficient.

[0332] To facilitate horizontal comparisons between different vulnerability clusters, a stability score is defined:

[0333]

[0334] The normalized risk is defined as:

[0335]

[0336] In different business scenarios, scenario-related field weight functions can also be defined:

[0337]

[0338] In the formula, Indicates a vulnerability cluster Stability score, For vulnerability clusters The risk of normalization This indicates the business scenario. For example, in an emergency warning scenario, the severity and repair fields have higher weights, while in an asset exposure analysis scenario, the affected scope field has higher weights.

[0339] The output of this step is: Field-level Risk Score Vulnerability cluster-level risk scoring Optional extended risk score and the revised cluster-level risk score .

[0340] S8. Explainable output and source responsibility positioning.

[0341] After obtaining the field-level risk score and the vulnerability cluster-level risk score, this invention further generates explanatory output to clarify the unstable source fields and their sources of responsibility.

[0342] For any field Its contribution to the total risk of the vulnerability cluster is defined as:

[0343]

[0344] Furthermore, fields can be defined separately. The contribution of timeliness and structure:

[0345]

[0346]

[0347] If missing items are introduced, the missing contribution can also be defined:

[0348]

[0349] In the formula, For fields vulnerability clusters Contribution to total risk.

[0350] Regarding field-level timeliness positioning, utilizing the aforementioned , and Fields can be output directly The earliest valid source set, the latest valid source set, and the lagging responsible source pair set are represented.

[0351] In terms of field-level structure location, utilizing Fields can be output directly In vulnerability cluster The biggest source of disagreement between China and the United States is their relationship.

[0352] In order to quantify responsibility for a single source, this invention defines the source. In vulnerability cluster Middle field The overall responsibility score is as follows:

[0353]

[0354] in, Indicate source In vulnerability cluster Middle field The overall responsibility score; This represents the standardized lag responsibility component defined in step S5; This represents the source outlier degree defined in step S6; It serves as a balance parameter between time-bound liability and structural liability.

[0355] Furthermore, source In vulnerability cluster Overall responsibility score Defined as:

[0356]

[0357] in For fields Responsibility weight:

[0358]

[0359] To facilitate comparisons between different sources, a normalized source responsibility score is further defined:

[0360]

[0361] This allows for a ranking of responsibility by source:

[0362]

[0363] In the formula, Indicates a vulnerability cluster China uses a normalized source responsibility scoring system. The source responsibility ranking results are obtained by sorting from high to low.

[0364] In other embodiments, vulnerability cluster-level stability risk scoring can also be used. Set risk level thresholds. For example:

[0365]

[0366] in, Preset risk classification thresholds; Indicates a vulnerability cluster The risk level is determined accordingly. Vulnerability clusters can be classified into stable, slightly unstable, moderately unstable, and highly unstable levels for subsequent automatic fusion, conflict labeling, or manual review.

[0367] Ultimately, this invention can address vulnerability clusters. The evaluation results are organized into structured evaluation results. Structured evaluation results It includes at least: vulnerability cluster-level stability risk score, stability score, vulnerability cluster-level risk level, field-level risk score, field-level contribution, earliest and latest valid source set for field timeliness, set of delayed responsibility source pairs, field with the greatest divergence source pair, source responsibility score, and source responsibility ranking results.

[0368]

[0369] The output of this step is for vulnerability clusters. Structured evaluation results .

[0370] In other embodiments, the structured evaluation results It can be directly used by downstream vulnerability intelligence processing systems. In vulnerability intelligence fusion scenarios, when multiple source fields in the same vulnerability cluster conflict, it can be based on... The system includes field-level risk scoring, maximum source of disagreement pairings, and source responsibility scoring. It automatically reduces the weight of conflicting fields, selects priority sources, marks conflicts, or generates a queue for review. In source quality monitoring scenarios, it can be based on... The source responsibility score and ranking results are used to issue quality alerts, adjust capture strategies, or optimize synchronization frequency for sources that are lagging behind or have significant discrepancies. In automated early warning and vulnerability handling scenarios, it can be based on... The vulnerability cluster-level stability risk score and risk level control alarm confidence, affect remediation priority ranking, and determine whether to trigger a manual review process; in knowledge graph construction, asset exposure analysis, or vulnerability remediation orchestration scenarios, it can be based on The field-level contribution and risk level are used to filter out highly unstable fields, reducing the probability of erroneous fields being written into downstream databases, knowledge graphs, or work order systems.

[0371] Experimental example: Experimental verification of the method for assessing the drift and stability of multi-source vulnerability information fields.

[0372] To verify the feasibility, risk assessment capability, and interpretation output capability of the multi-source vulnerability information field drift and stability assessment method described in this invention, a corresponding experimental project was constructed, and verification was performed using controllable simulated data and real multi-source vulnerability data. The experimental process corresponds to the processing flow of S1 to S8 described above in this invention, including multi-source vulnerability entry acquisition, field standardization, cross-source vulnerability cluster construction, unified timeline establishment, field validity determination, time lag drift calculation, structural divergence calculation, risk fusion, and structured interpretation result output.

[0373] This experiment was run on a general-purpose computing device with macOS 26.5-arm64 as the operating system and Python version 3.13.7. The main dependencies included numpy 2.3.3, pandas 2.3.3, scikit-learn 1.7.2, and scipy 1.16.2. The controlled simulation data experiment used a random seed of 42 to ensure reproducibility. The experiment used descriptive, severity, affected scope, repair information, and reference fields as the main evaluation fields, with weights set to 0.15, 0.25, 0.25, 0.25, and 0.10, respectively. The structural divergence weight α was set to 0.55, the lag weight β to 0.45, and the low / high risk thresholds to 0.17 and 0.32, respectively. The remaining parameters used the default experimental configuration. The above parameters are used to illustrate one possible embodiment of the method of this invention and do not constitute a limitation on the scope of protection of this invention.

[0374] I. Controllable simulation data experiment.

[0375] To evaluate the risk identification and interpretation capabilities of the method of this invention under known annotation conditions, a dataset containing 300 simulated vulnerability clusters was constructed. The simulated data was used to construct typical differences in different fields of multi-source vulnerability information, and risk type annotations were assigned to each vulnerability cluster. The simulated scenarios included four types: stable, time-lag, structural divergence, and hybrid. Specifically, stable indicates that the fields from multiple sources are basically consistent; time-lag indicates that there is a time difference between the first availability of fields from different sources; structural divergence indicates that the final field expressions from different sources differ; and hybrid indicates the simultaneous existence of time lag and structural divergence.

[0376] The experiment compared the method of this invention with four types of baseline methods. The first type is the missing rate baseline method, which scores based solely on the field's missing rate; the second type is the final value conflict baseline method, which compares only the conflicts between final field values ​​from different sources; the third type is the timeliness baseline method, which considers only the lag difference between the first available times of a field; and the fourth type is the structure baseline method, which considers only the structural discrepancies between final field expressions from different sources. The method of this invention integrates timeliness lag, structural discrepancies, and field weights. Evaluation metrics include macro-average F1 score, accuracy, Spearman correlation coefficient, mean absolute error, root mean square error, hit rate of the first high-risk field, hit rate of the top three high-risk fields, accuracy of locating the latest source, and hit rate of the source pair with the greatest discrepancy. Table 1 shows the main evaluation results of the method of this invention in controlled simulation data experiments.

[0377] Table 1 Main Results of Controllable Simulation Data Experiment

[0378]

[0379] In terms of interpretation and localization, the method of this invention achieved a hit rate of 0.9502 for the first high-risk field, 1.0000 for the first three high-risk fields, 0.9936 for the latest source localization, and 0.4940 for the largest source discrepancy pair. These results demonstrate that the method of this invention can produce relatively stable outputs in risk level identification, risk ranking, and interpretation and localization. It should be noted that the time-based baseline method may exhibit a lower mean absolute error on some error indicators due to its narrower risk score range, but it cannot simultaneously characterize structural discrepancies, nor can it output complete field contribution, largest source discrepancy pair, and source responsibility score; therefore, it lacks the comprehensive interpretive capability of the method of this invention.

[0380] II. Real-world three-source vulnerability data experiment.

[0381] To verify the feasibility of the method of this invention on real heterogeneous vulnerability data sources, experiments were further conducted using three real vulnerability data sources collected locally: OSV, GHSA, and NVD. The experiments used local static snapshot files, without internet access or external API calls. Since real data sources do not provide complete field-level historical versions, the experiments did not use manually labeled ground truth values, nor did they calculate macro average F1 scores or accuracy metrics; field timeliness drift was approximated based on record-level time fields such as release time and update time in the source records. In practical application scenarios with field-level historical versions or incremental update logs, the first availability time of the real field can be directly used to replace this approximate estimate.

[0382] After field standardization, the real three-source data are aligned primarily based on CVE identifiers. When a CVE identifier is missing or cannot be directly matched, auxiliary alignment is performed by combining GHSA / OSV identifiers, alias information, and the intersection of reference links. The actual data size and cross-source clustering are shown in Table 2.

[0383] Table 2 Statistical Analysis of Real Three-Source Data

[0384]

[0385] Among the 200 three-source vulnerability clusters analyzed in detail, the information stability risk levels were distributed as follows: 138 high-risk, 61 medium-risk, and 1 low-risk, with an average vulnerability cluster risk score of 0.3913. The highest-risk vulnerability cluster was REAL-00005, with a representative vulnerability identifier of CVE-2000-0062 and a risk score of 0.6237. The main contributing field was the affected scope field. The average risk ranking of the fields showed that the average risk for the affected scope field was 0.1378, for the referenced field it was 0.0966, for the severity field it was 0.0795, for the description field it was 0.0774, and for the remediation information field it was 0.0000. This result indicates that in real multi-source vulnerability data, the affected scope, referenced information, severity, and description fields exhibit varying degrees of expression differences across different sources. The method of this invention can quantify these differences and pinpoint the main contributing fields.

[0386] For example, in a real three-source experiment, the vulnerability cluster REAL-00005 corresponds to the representative vulnerability identifier CVE-2000-0062, with an information stability risk score of 0.6237, a risk level of high risk, and the main contributing field being the affected scope field. This result indicates that there are significant differences in the expression of the affected scope of this vulnerability across different sources, and the method of this invention can further pinpoint the key fields causing information instability.

[0387] Regarding source responsibility scoring, the total source responsibility scores for GHSA, OSV, and NVD in this experiment were 39.4071, 24.5500, and 14.3067, respectively. This result does not indicate that the data quality of any particular source is absolutely poor, but rather that, under the field mapping, field weighting, and risk fusion settings of this experiment, different sources contribute differently to the instability of multi-source fields. This result can be used for source quality monitoring, conflict field review, and priority ranking of manual audits.

[0388] Furthermore, real-world data experiments revealed differences in the availability of the repair information field across different sources. Specifically, 332,778 records in OSV contained identifiable repair information fields, representing approximately 42.29% of all OSV records; however, this field was essentially unavailable in GHSA and NVD under the current mapping rules. These results reflect differences in the granularity of repair information disclosure and the organization of fields across different data sources, demonstrating that the method of this invention can identify and quantify differences in field coverage across different sources.

[0389] III. Verification of Structured Interpretation Output.

[0390] To verify the completeness of the structured assessment results described in S8 of this invention, the experiment further verified the fields of the final structured output. The experimental results already include vulnerability cluster-level stability risk score, risk level, field-level risk score, field-level contribution, earliest valid source set for field validity, latest valid source set for field validity, field with the greatest divergence source pair, and source responsibility score. Simultaneously, based on the existing results, stability score, set of lagging responsibility source pairs, and source responsibility ranking results were supplemented and generated. Specifically, the stability score is calculated by normalizing the vulnerability cluster-level risk score; the set of lagging responsibility source pairs is generated from the earliest valid source set and the latest valid source set; and the source responsibility ranking results are obtained by sorting the source responsibility scores in descending order.

[0391] The above outputs are consistent with the structured evaluation results of this invention. The components are consistent, enabling the vulnerability intelligence fusion system to automatically mark conflicting fields, reduce their weight, select priority sources, and generate a queue for review. It also supports source quality monitoring and manual review assistance.

[0392] IV. Experimental Conclusions.

[0393] The experiments described above demonstrate that the method of this invention can standardize fields and construct cross-source vulnerability clusters for multi-source vulnerability information entries without relying on a completely unified schema, and can jointly quantify field time lag drift and structural divergence. Compared with baseline methods that only consider field missingness, final field conflict, single time lag, or single structural divergence, the method of this invention can simultaneously output field-level risk scores, vulnerability cluster-level stability risk scores, stability scores, risk levels, field contribution, earliest / latest valid source set, lag responsibility source pair set, field maximum divergence source pair, source responsibility score, and source responsibility ranking results, thereby improving the interpretability of multi-source vulnerability information fusion and manual review processes.

[0394] Real-world three-source experiments further demonstrate that the method of this invention can be applied to real heterogeneous vulnerability data sources such as OSV, GHSA, and NVD. It completes standardization, cross-source alignment, risk calculation, and interpretation output on millions of locally collected records, exhibiting good engineering feasibility and scalability. This method can be used in scenarios such as vulnerability intelligence fusion, source quality monitoring, conflict field location, automated early warning, and manual review assistance.

[0395] The method of the present invention can be implemented by a device, an electronic device, and a computer-readable storage medium.

[0396] like Figure 2 As shown, the device may include:

[0397] The data acquisition and standardization module is used to perform step S1 above;

[0398] The vulnerability alignment and clustering module is used to perform step S2 above;

[0399] The unified timeline construction module is used to perform step S3 above;

[0400] The field validity determination and time lag calculation module is used to perform the above steps S4 and S5;

[0401] The structural divergence calculation module is used to perform the above step S6;

[0402] The risk fusion module is used to perform step S7 above;

[0403] The output module is used to perform step S8 above.

[0404] The electronic device may include a processor, a memory, a communication interface, and a bus, wherein the memory stores a computer program, which, when executed by the processor, is used to implement the steps S1 to S8 described above.

[0405] The computer-readable storage medium stores a computer program, which, when executed by a processor, is used to implement the steps S1 to S8 described above.

[0406] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

[0407] The above embodiments merely illustrate several implementation methods of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A method for assessing vulnerability information field drift and stability based on multiple data sources, characterized in that, include: Obtain raw vulnerability information entries from multiple data sources, and perform field unification on each raw vulnerability information entry to form a standardized field representation on a unified field set; Based on the vulnerability identifier field in the standardized field representation, original vulnerability information entries pointing to the same vulnerability from different data sources are merged into the same vulnerability cluster; for the remaining original vulnerability information entries, the alignment confidence between any two original vulnerability information entries is calculated, and vulnerability clusters are merged based on the alignment confidence; For each vulnerability cluster, the release time and update time of the original vulnerability information entries from each source are collected to construct a unified timeline and obtain an ordered unified timeline. Based on the unified timeline, a field status time function for each source is established for any field in the unified field set. Using the time in the ordered unified timeline, and based on the field status time function of each source, determine the first available time of the field in each source, and obtain the normalized cross-source time lag drift, the earliest valid source set, and the latest valid source set of the field; Aggregate identical fields from the same data source in the vulnerability cluster, perform pairwise comparisons on aggregated fields from different sources, calculate the structural divergence degree, and take the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair. By integrating the time lag drift and structural divergence according to field weights and dimension weights, a field-level risk score and a vulnerability cluster-level stability risk score are obtained. The output includes a structured assessment report containing field-level risk scores, vulnerability cluster-level stability risk scores, the earliest valid source set, the latest valid source set, and the largest divergence source pair.

2. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The step of calculating the alignment confidence between any two original vulnerability information entries for the remaining original vulnerability information entries, and merging vulnerability clusters based on the alignment confidence, includes: Based on the vulnerability identifier field, calculate the vulnerability identifier similarity between two original vulnerability information entries; Calculate the description similarity between two original vulnerability information entries based on the description field; Calculate the temporal proximity between two original vulnerability information entries based on the release time field and the update time field. Calculate the similarity of affected scope between two original vulnerability information entries based on the affected scope field; Calculate the citation similarity between two original vulnerability information entries based on the citation field; The weighted sum of vulnerability identifier similarity, description similarity, temporal proximity, affected scope similarity, and reference similarity is used as the alignment confidence between two original vulnerability information entries; Two original vulnerability information entries with an alignment confidence score greater than the alignment threshold are grouped into the same vulnerability cluster.

3. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The step of establishing a field status time function from various sources for any field in a unified field set based on a unified timeline includes: Prior to the publication time of the original vulnerability information entry, the fields in the original vulnerability information entry were in an unknown state; Between the publication time of the original vulnerability information entry and the first update time, the fields in the original vulnerability information entry are in their initial state values; In the original vulnerability information entry, at the... The time of the next update and the first Between the update times, the fields in the original vulnerability information entry are the first... The updated state value , Indicates the number of times the original vulnerability information entry has been updated.

4. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The process of determining the first available time of a field in each source, based on the field status time function in the ordered unified timeline, includes: Take the time from the ordered unified timeline, substitute it into the field status time function, and get the field status value at the current time; Determine the validity of the field's status value at the current time based on the set of valid value ranges for the field; The current time when the state value first becomes valid is taken as the first available time; if there is no time in the entire ordered unified timeline that makes the state value valid, then the preset upper bound of time is taken as the first available time.

5. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The normalized cross-source time lag drift, the earliest valid source set, and the latest valid source set for the field include: The difference between the latest first-access time and the earliest first-access time of a field on a vulnerability cluster is used as the cross-source time lag drift of the field on the vulnerability cluster. The cross-source time lag drift is normalized to obtain the normalized cross-source time lag drift. Combine one or more data sources that reach their first availability time earliest from all data sources to form the earliest valid source set; Combine one or more data sources that reached their first availability time latest from all data sources to form the latest valid source set.

6. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, It also includes the calculation of standardized lag responsibility components, including: Construct a set of lagging responsible source pairs based on the Cartesian product of the earliest and latest valid source sets; In the formula, Indicate source In vulnerability cluster Middle field The standardization lag responsibility weight, For data source In vulnerability cluster Middle field The first available time, For data source In vulnerability cluster Middle field The first available time, For data source collection, Representation field In vulnerability cluster Cross-source time lag drift.

7. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The step of performing pairwise comparisons on the aggregated representations of fields from different sources, calculating the structural divergence degree, and taking the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair includes: For the same vulnerability cluster, calculate the aggregated distance value between pairs of different data sources with the same field, and take the mean of all distance values ​​as the structural divergence degree; The data sources of the two fields with the highest structural divergence are taken as the source pair with the highest structural divergence in the vulnerability cluster. For a field within the same vulnerability cluster, calculate the distance between the aggregated representation under one data source and the aggregated representation under other data sources, and take the average of all distance values ​​as the source outlier of the field under the current data source.

8. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, The field-level risk score is a fusion value that integrates time lag drift and structural divergence according to field weight and dimension weight; The vulnerability cluster-level stability risk score is the sum of the field-level risk scores of all fields in the same vulnerability cluster.

9. The method for assessing vulnerability information field drift and stability based on multiple data sources according to claim 1, characterized in that, This also includes ranking the source responsibility within the vulnerability cluster, including: The weighted sum of the standardized lag responsibility component and the source outlier is taken as the comprehensive responsibility score of the data source on the field in the vulnerability cluster; The overall responsibility score of the data source in the vulnerability cluster is obtained by summing the comprehensive responsibility scores of all fields of the same data source in the vulnerability cluster. The overall responsibility score is normalized to obtain the normalized source responsibility score of the data source on the vulnerability cluster; All data sources are ranked from highest to lowest according to the normalized source responsibility score, resulting in the source responsibility ranking results for data sources within the vulnerability cluster.

10. A device for assessing vulnerability information field drift and stability based on multiple data sources, characterized in that, include: The data acquisition and standardization module is used to acquire raw vulnerability information entries from multiple data sources and perform field unification on each raw vulnerability information entry to form a standardized field representation on a unified field set. The vulnerability alignment and clustering module is used to merge original vulnerability information entries pointing to the same vulnerability from different data sources into the same vulnerability cluster based on the vulnerability identifier field in the standardized field representation; for the remaining original vulnerability information entries, it calculates the alignment confidence between any two original vulnerability information entries and merges vulnerability clusters based on the alignment confidence; The unified timeline construction module is used to collect the release and update times of the original vulnerability information entries from various sources for each vulnerability cluster, construct a unified timeline, and obtain an ordered unified timeline. Based on the unified timeline, a field status time function for each source is established for any field in the unified field set. The field validity determination and time lag calculation module is used to determine the first available time of a field in each source based on the field status time function of each source, using the time in the ordered unified timeline, and to obtain the normalized cross-source time lag drift of the field, the earliest valid source set, and the latest valid source set; The structural divergence calculation module is used to aggregate identical fields from the same data source in the vulnerability cluster, compare the aggregated fields from different sources pairwise, calculate the structural divergence degree, and take the two sources corresponding to the maximum structural divergence degree as the maximum divergence source pair. The risk fusion module is used to fuse time lag drift and structural divergence according to field weights and dimension weights to obtain field-level risk scores and vulnerability cluster-level stability risk scores. The explanation output module is used to output a structured assessment report that includes field-level risk scores, vulnerability cluster-level stability risk scores, the earliest valid source set, the latest valid source set, and the largest divergence source pair.