An enterprise digital asset value evaluation method and system based on multi-source data fusion

Abstract

The application discloses a kind of based on multi-source data fusion's enterprise digital asset value evaluation method and system, it is related to evaluation analysis technical field, it includes extracting the first characteristic vector corresponding to target reconstruction alliance and the second characteristic vector corresponding to remaining participants, calculate the conditional mutual information value between the first characteristic vector and the second characteristic vector;The remaining participants filtered out according to conditional mutual information value are merged into super node with target reconstruction alliance, and the independent baseline value data of super node is cooperated and decayed correction;Super node is regenerated as single participant initial independent value data and returns traversal, until the sum of initial independent value data of each sub-alliance participant is not less than corresponding independent baseline value data, and the decoupling value data of each participant is output.The application has the effect of improving enterprise digital asset value evaluation efficiency.

Description

Technical Field

[0001] This application relates to the field of evaluation and analysis technology, and in particular to a method and system for evaluating the value of enterprise digital assets based on multi-source data fusion. Background Technology

[0002] As enterprises deepen their digital transformation, digital assets such as data, algorithms, business interfaces, and customer behavior records have become important production factors in business operations. Their value is reflected not only in the scale of individual data resources but also in the revenue contribution generated after multiple types of data participate in business processes. Meanwhile, models such as joint operation, platform collaboration, and data co-construction enable the combination of digital assets from multiple participants into joint asset packages, providing services and generating settlement revenue. Digital asset valuation is gradually becoming a key link in revenue distribution, asset accounting, and collaborative governance.

[0003] It is worth noting that the value of enterprise digital assets is not simply determined by the amount of data, the number of calls, or the construction cost. In a joint asset package, the sub-assets provided by different participants often have complementary fields, coupled scenarios, and business dependencies. The value performance of a sub-asset running alone may differ from its contribution to the joint service. Especially when multi-source data jointly supports the same business metric, the synergy between assets will affect service results and settlement revenue, making the value of digital assets exhibit non-linear, combinatorial, and dynamically allocated characteristics.

[0004] However, existing methods for assessing enterprise digital assets still have significant shortcomings. Some methods rely primarily on static estimations based on cost input, data scale, or historical revenue, failing to reflect the marginal contribution of each participant's data in actual business operations. While other methods consider the sharing of joint revenue, they typically treat participants as independent entities, ignoring potential dependencies and value obscuring that may arise from combining sub-assets, easily leading to revenue allocation deviating from actual contributions. Especially in scenarios involving multi-party data fusion, without refined processing of the relationships between service flows, business outcomes, and asset characteristics, the assessment results are insufficient to support subsequent asset pricing, collaborative settlement, and ownership management. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a method and system for evaluating the value of enterprise digital assets based on multi-source data fusion.

[0006] Firstly, this application provides a method for assessing the value of enterprise digital assets based on multi-source data fusion, including: Obtain the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs; Field alignment and format standardization are performed on the sub-asset feature datasets of each participant to generate a standard feature dataset; The corresponding standard feature datasets are removed one by one for each participant. The change in the target business indicator is calculated based on the service result field in the service call log. Initial independent value data for each participant is generated according to the proportion of the change in the target business indicator and the total revenue value data. Iterate through the combinations of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances; Compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, mark the sub-alliance with the largest difference as the target reconstruction alliance. Extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and calculate the conditional mutual information value between the first feature vector and the second feature vector; The remaining participants selected based on conditional mutual information values ​​are merged into the target reconstruction alliance to form super nodes, and the independent baseline value data of the super nodes are collaboratively decayed and corrected. Regenerate the initial independent value data for the supernode as a single participant and iterate back until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

[0007] Secondly, this application provides an enterprise digital asset valuation system based on multi-source data fusion, comprising: The data acquisition module is used to acquire the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs. The standardization processing module is used to perform field alignment and format standardization on the sub-asset feature datasets of each participant to generate a standard feature dataset; The calculation and generation module is used to remove the corresponding standard feature datasets one by one according to the participants, calculate the change of the target business indicators based on the service result fields in the service call log, and generate the initial independent value data of each participant according to the proportion of the change of the target business indicators and the total revenue value data. The traversal module is used to traverse the combination of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances. The comparison and judgment module is used to compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, the sub-alliance with the largest difference is marked as the target reconstruction alliance. The vector extraction module is used to extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and to calculate the conditional mutual information value between the first feature vector and the second feature vector. The merging module is used to merge the remaining participants selected based on conditional mutual information values ​​with the target reconstruction alliance into a super node, and to perform collaborative decay correction on the independent baseline value data of the super node. The output module is used to regenerate the initial independent value data of the super node as a single participant and return the traversal until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

[0008] In summary, this application includes at least one of the following beneficial technical effects: 1. This application provides a method for assessing the value of enterprise digital assets based on multi-source data fusion. It obtains service call logs of a joint asset package and sub-asset feature datasets of each participant. The settlement information and service result information in the service call logs are correlated with the sub-asset feature datasets of each participant. While generating total revenue value data, the data of each participant is aligned and standardized in format to form a standard feature dataset with a unified caliber. Based on this, the corresponding standard feature datasets are removed one by one for each participant, and the change in target business indicators is calculated based on the service result fields. This change is then mapped to the total revenue value data to form initial independent value data. Therefore, the value of digital assets no longer relies solely on data scale, construction costs, or static settlement amounts for estimation. Instead, it can identify the marginal contribution of each participant's data from the correspondence between actual service calls, changes in business results, and revenue data. This ensures that the assessment process simultaneously possesses traceability of revenue sources, quantifiable business contributions, and consistency of multi-source data caliber, effectively reducing assessment bias caused by heterogeneous fields, inconsistent data formats, and a single revenue allocation method. 2. By traversing the combinations of participants, sub-alliances are generated. Based on the standard feature dataset and the settlement field in the service call log within the sub-alliances, the independent baseline value data is calculated. The sum of the initial independent value data of the participants in the sub-alliances is then compared with the independent baseline value data. When there are undervalued combinations, the sub-alliance with the largest difference is identified as the target reconstruction alliance. Subsequently, the first feature vector of the target reconstruction alliance and the second feature vector of the remaining participants are extracted, and the conditional mutual information value between them is calculated. Based on this, the remaining participants with collaborative dependencies are selected and merged into super nodes. The independent baseline value data of the super nodes is corrected by collaborative decay and then re-participates in the value calculation until the allocation results of each sub-alliance satisfy the independent baseline constraints. The aforementioned features form a closed-loop processing chain from individual contribution identification, combined value verification, collaborative dependency discovery to super node reconstruction. This chain can not only discover the combined value that is obscured by individual elimination methods in multi-source data fusion scenarios, but also avoid double-counting of strongly related assets through collaborative attenuation. As a result, the decoupled value data of each participant in the final output is more in line with the real service contribution of the joint asset package, providing a more stable, detailed and interpretable data basis for digital asset pricing, revenue distribution and multi-party collaborative settlement. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram showing the connection of each step in the method of the present invention.

[0011] Figure 2 This is a schematic diagram of the connection between the modules of the system of the present invention. Detailed Implementation

[0012] The following description, in conjunction with the implementation of the present invention, is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention, and all such modifications and additions should fall within the protection scope of the present invention.

[0013] Example 1

[0014] This application applies to scenarios where multiple participants jointly provide digital assets such as data, interfaces, tags, and rule bases, and these digital assets are combined to form a joint asset package for external business services. The evaluation processor uses service call logs as a common source of revenue and business results, and the sub-asset feature datasets of each participant as the basis for calculating asset contributions. Within the same target evaluation period, it aligns service revenue, service results, and asset features to the same data caliber. Then, through process-by-process elimination, sub-alliance traversal, conditional mutual information value identification, and supernode reconstruction, it corrects the collaborative dependencies and value obscuration among participants, ultimately forming decoupled value data for each participant. This process does not simply allocate data based on volume or historical revenue, but rather processes data around its role path in actual service calls, enabling the evaluation results to reflect the marginal contribution and combinatorial dependencies in multi-source data fusion scenarios.

[0015] This application discloses a method for evaluating the value of enterprise digital assets based on multi-source data fusion.

[0016] Reference Figure 1 A method for assessing the value of enterprise digital assets based on multi-source data fusion includes the following steps: Obtain the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs; Furthermore, obtain the service call logs of the joint asset package and the sub-asset feature datasets of each participant. Generate total revenue value data based on the service call logs, including: Read the service call logs of the joint asset package during the target assessment period; Extract the number of service requests, the settlement amount per request, and the settlement completion identifier from the service call log; Remove service call records with empty settlement completion flags to obtain valid service call records; The periodic revenue value is calculated based on the number of service requests and the settlement amount of a single request in the valid service call record, and the periodic revenue value is rounded to a predetermined number of places to generate the total revenue value data. Read the original business data tables submitted by each participant, and establish a mapping relationship between the original business data tables and the corresponding participants based on the participant identifiers; Extract business fields that match the joint asset package service call fields from the original business data table to generate a sub-asset feature dataset for each participant.

[0017] In one specific implementation, the evaluation processor first acquires the service call logs of the joint asset package and the sub-asset feature datasets of each participant. The service call logs are derived from call records generated by the joint asset package within the target evaluation period, which can be a calendar month, a settlement cycle, or an evaluation period determined internally by the enterprise. Each service call log entry includes at least the service call number, service timestamp, number of service requests, settlement amount per request, settlement completion identifier, settlement field, service result field, and service call field. The service call number is used to associate a service request with the corresponding business field, settlement field, and service result field; the service timestamp is used for subsequent time slice alignment; and the settlement completion identifier is used to exclude records that have not yet been settled. The sub-asset feature datasets of each participant are derived from the original business data tables submitted by each participant, such as customer profile fields, transaction behavior fields, device status fields, tag fields, and risk rule fields. The evaluation processor generates total revenue value data based on the service call logs, ensuring that subsequent initial independent value data and independent baseline value data fall within the same revenue dimension.

[0018] Specifically, when the evaluation processor reads the service call logs of the joint asset package within the target evaluation period, it can merge them from the enterprise's service gateway log library, billing system log library, and business result log library by service call number. A single service call log entry can be stored in a structured format. For example, a service call number of S202405010001, a service timestamp of May 1, 2024 at 10:05 AM, 1 service request, a single request settlement amount of 0.80 yuan, a settlement completion flag of "completed," a service result field of "passed and fulfilled successfully," and service call fields of customer profile, transaction behavior, and risk label. The evaluation processor extracts the service request count, single request settlement amount, and settlement completion flag from the service call logs, and filters out service call records with empty settlement completion flags to obtain valid service call records. This filtering action ensures that calls that have not yet generated actual revenue are not included in the total revenue value data, preventing subsequent initial independent value data from being amplified by unsettled records. Periodic revenue values ​​can be calculated by... Calculate, where, This indicates the number of valid service call records within the target evaluation period. Indicates the first The number of service requests in each valid service call record. This indicates the settlement amount for a single request in this valid service call record. This represents the periodic revenue value generated by all valid service call records within the target evaluation period. For the number of times, The formula represents the sum of the amounts: each completed service call is multiplied by itself, and the sum remains the sum. The physical meaning of this formula is to convert each completed service call into the total revenue of the joint asset package within the target assessment period. The assessment processor then rounds the periodic revenue value to a predetermined number of decimal places to generate the total revenue value data. The predetermined number of decimal places can be set to two decimal places to suit the commonly used precision of enterprise settlement amounts.

[0019] The original business data tables submitted by each participant can be accessed by the evaluation processor via data exchange interfaces, offline files, or database views. After reading the original business data tables, the evaluation processor establishes a mapping relationship between the original business data tables and the corresponding participants based on their identifiers. For example, if participant A submits a customer basic field table, participant B submits a transaction behavior field table, and participant C submits a risk label field table, the evaluation processor records the binding relationships between participant identifiers A, B, and C and their respective original business data tables. Subsequently, the evaluation processor extracts business fields from the original business data tables that match the service call fields of the joint asset package, generating sub-asset feature datasets for each participant. If a business field has never been referenced by a service call field, that field will not be included in the sub-asset feature dataset, reducing the interference of data unrelated to service revenue on value assessment from the source.

[0020] Field alignment and format standardization are performed on the sub-asset feature datasets of each participant to generate a standard feature dataset; Furthermore, the sub-asset feature datasets of each participant are aligned and standardized in format to generate a standard feature dataset, including: Read the field names, field types, and field value ranges from the feature datasets of each sub-asset; Match the field names with the unified field directory to obtain the standard field names corresponding to each sub-asset feature dataset; Convert fields with inconsistent data types to a unified data type, perform unit conversion and fill in missing values ​​for the field value range, and generate candidate standard fields. Based on the service timestamps in the service call logs, the candidate standard fields are aligned by time slices according to the target evaluation period; The candidate standard fields, after being aligned by time slices, are recombined according to the participant identifiers to generate standard feature datasets for each participant.

[0021] In one specific implementation, after obtaining the aforementioned data, the evaluation processor performs field alignment and format standardization on the sub-asset feature datasets of each participant. Since different participants typically have inconsistent field names, field types, field units, and field value ranges—for example, a customer level field with the same meaning might be stored as a numeric level, a character label, and an enumeration code—direct calculation would result in data with the same meaning being treated as different features. Therefore, the evaluation processor converts each sub-asset feature dataset into standard field names based on a unified field catalog, converts the field types to a unified field type, performs unit conversion and fills in missing values ​​for the field value ranges, and places the data into the time slice of the target evaluation period according to the service timestamp in the service call log. This generates a standard feature dataset for each participant. The standard feature dataset retains both participant identifiers and standard fields corresponding to the service call fields, thus allowing direct association with service result fields and settlement fields in subsequent steps.

[0022] Specifically, in the field alignment and format standardization phase, the evaluation processor reads the field names, field types, and field value ranges from the feature datasets of each sub-asset. Field names are used to match against a unified field catalog, field types are used to determine conversion rules, and field value ranges are used to identify unit differences and outliers. A unified field catalog can be established by the enterprise before implementation based on the service call fields of the joint asset package, for example, by... `customer_no` is uniformly mapped to customer ID, `risk_score` and `score_risk` are uniformly mapped to risk scores, and `order_amt` and `trade_amount` are uniformly mapped to transaction amounts. The evaluation processor matches field names with a unified field directory to obtain standard field names corresponding to each sub-asset feature dataset. For fields with inconsistent data types, character numbers are converted to numeric types, enumerated levels are converted to unified enumeration codes, and date strings are converted to a unified time format. Unit conversion for field value ranges is also completed at this stage; for example, amounts stored in points are converted to yuan, and scores stored in percentages are converted to values ​​between 0 and 1. Missing value imputation can be completed based on the median, mode, or preset missing value markers for the same standard field of the same participant within the target evaluation period. For core fields used to calculate changes in target business indicators, the evaluation processor can also record missing value markers to preserve the impact of missing values ​​during subsequent elimination calculations.

[0023] After generating the candidate standard fields, the evaluation processor aligns them with time slices according to the target evaluation period based on the service timestamps in the service call logs. The length of the time slice can be determined based on the service call density, with common values ​​being 1 hour, 1 day, or 1 week. A 1-hour slice can be used for high-frequency risk control services, while a 1-day slice can be used for low-frequency enterprise services. The evaluation processor uses the time slice in which the service timestamp falls as an index to associate the candidate standard fields with service call records within the same time slice. For cases where multiple candidate standard field records exist within a time slice, numerical fields can take the mean, maximum, minimum, or most recent value, while enumerated fields can take the most frequently occurring value. The time-sliced ​​aligned candidate standard fields are then recombined according to the participant identifiers to generate standard feature datasets for each participant. This ensures that the subsequent first feature vector, second feature vector, and condition variable sequence can all be calculated on the same time axis, avoiding false associations caused by different data generation times.

[0024] The corresponding standard feature datasets are removed one by one for each participant. The change in the target business indicator is calculated based on the service result field in the service call log. Initial independent value data for each participant is generated according to the proportion of the change in the target business indicator and the total revenue value data. Furthermore, the standard feature datasets corresponding to each participant are removed one by one. The change in target business metrics is calculated based on the service result fields in the service call logs. Initial independent value data for each participant is generated according to the proportion of change in target business metrics and the total revenue value data, including: Based on the matching records between the standard feature dataset of all participants and the service result field in the service call log, calculate the benchmark value of the target business indicator of the joint asset package within the target evaluation period; Remove the standard feature dataset of each individual participant sequentially from the standard feature dataset of all participants, while retaining the service result field within the same target evaluation period, and calculate the comparison value of the target business indicator after removal. The difference between the baseline value of the target business indicator and the comparison value of the target business indicator is determined as the change in the target business indicator of the participating party; The changes in the target business indicators of all participants are normalized to generate contribution percentage data for each participant. The contribution percentage of each participant is multiplied by the total revenue value to generate the initial independent value data for each participant.

[0025] In one specific implementation, the evaluation processor removes the corresponding standard feature datasets for each participant one by one, and calculates the change in the target business indicator based on the service result field in the service call log. Taking a risk control service scenario as an example, the service result field may include whether the service was hit, whether the service passed, the result score, and the subsequent performance status. The target business indicator may be an indicator consistent with the joint asset package service target, such as the performance success rate after the service passes, the effective conversion rate, and the anomaly interception accuracy rate. In this embodiment, the evaluation processor first calculates the baseline value of the target business indicator using the matching records between the standard feature datasets of all participants and the service result fields; then, it sequentially removes the standard feature datasets of individual participants, keeping the service result fields unchanged within the same target evaluation period, and recalculates the comparison value of the target business indicator. The difference between the baseline value of the target business indicator and the comparison value of the target business indicator is the change in the target business indicator for the corresponding participant. If the baseline value of the target business indicator is denoted as... The target business metric comparison value after removing the i-th participant is denoted as . Then the first The change in the target business indicators of each participant can be expressed as: ,in, Indicates the first The amount of business metrics lost due to the removal of a participant. and All values ​​are for the same target business indicator and have the same dimensions. The physical meaning of this difference is the marginal support strength of the standard feature dataset of the participating party for the business results of the joint asset package.

[0026] The evaluation processor normalizes the changes in target business metrics for all participants, generating contribution percentage data. This contribution percentage data is then multiplied by the total revenue value data to obtain the initial independent value data for each participant. If the... The contribution percentage of each participant is recorded as follows: but You can press Divide by the sum of changes in all target business metrics to obtain the total revenue value; if the total revenue value is denoted as... Then the first The initial independent value data for each participant is ,in, The dimension of the quantity is monetary. As a dimensionless proportion, this calculation maps the contribution of business metrics to the amount of distributable revenue.

[0027] Specifically, the calculation of the change in the target business metric is performed after the aforementioned standard feature dataset is formed. The evaluation processor calculates the baseline value of the target business metric for the joint asset package within the target evaluation period based on the matching records between the standard feature datasets of all participants and the service result fields in the service call logs. Matching records can be determined by the service call number and time slice; only records where the standard fields in the standard feature dataset correspond to the service call fields in the service call logs, and the service result field is not empty, are included in the calculation. If the target business metric uses the effective conversion rate, the baseline value can be obtained by dividing the number of effective conversion records by the number of effective service call records; if it uses the anomaly interception accuracy rate, it can be obtained by dividing the number of correctly intercepted records by the number of records that should have been intercepted. The evaluation processor sequentially removes the standard feature datasets of individual participants from the standard feature datasets of all participants, retaining the service result fields within the same target evaluation period, and calculates the comparative value of the target business metric after removal. Retaining the service result fields here ensures that the baseline and comparative values ​​are calculated under the same business facts, avoiding distortion of the change in the target business metric due to sample replacement.

[0028] Once the changes in target business metrics are determined, the evaluation processor normalizes the changes in target business metrics for all participants. If any individual target business metric changes are less than zero, it indicates that the target business metric did not decrease after removing that participant. In this case, the evaluation processor can normalize the change in that target business metric as zero to avoid negative contributions occupying a share of revenue allocation. Normalization can be performed as follows: In progress, among which... Indicates the first Changes in target business metrics for each participant Indicates the first Changes in target business metrics for each participant Indicates the first Data on the contribution percentage of each participant. Indicates taking The larger of the two zeros. As a dimensionless proportion, the sum of all contribution percentage data is 1. The physical meaning of this formula is to transform the marginal impact of different participants on the same target business metric into comparable allocation shares. If the changes in all target business metrics are not greater than zero, the evaluation processor can generate contribution percentage data based on the percentage of field coverage in the effective service call records for each participant, allowing the calculation process to continue without introducing unprovided parameters. Finally, the evaluation processor multiplies each participant's contribution percentage data by the total revenue value data to generate initial independent value data for each participant. This data then enters the sub-alliance comparison phase.

[0029] Iterate through the combinations of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances; Furthermore, by iterating through the participating parties to generate sub-alliances, and calculating independent baseline value data based on the standard feature datasets and settlement fields in the service call logs within the sub-alliances, including: The participants are combined and enumerated according to their identifiers to generate multiple sub-alliances; For each sub-alliance, extract the standard feature dataset corresponding to the participants within the sub-alliance to generate the sub-alliance feature dataset; Extract candidate settlement records from the settlement field of the service call log that match the service field in the sub-alliance feature dataset; Calculate the sub-alliance revenue data based on the number of service requests and settlement amount in the candidate settlement records; Based on the number of valid service fields covered by the sub-alliance feature dataset, the field coverage of the sub-alliance revenue data is corrected to generate the independent baseline value data of the sub-alliance.

[0030] In one specific implementation, after obtaining the initial independent value data, the evaluation processor does not directly output the allocation result. Instead, it iterates through the combinations of participants to generate sub-alliances. A sub-alliance is a combination consisting of at least two participants. The evaluation processor enumerates combinations of participants according to their identifiers. For example, when participants are A, B, C, and D, it can generate combinations such as A-B, A-C, A-D, B-C, B-D, C-D, and higher-order combinations. For each sub-alliance, the evaluation processor extracts the standard feature dataset corresponding to the participants within that sub-alliance, generating a sub-alliance feature dataset. Then, it extracts candidate settlement records from the settlement fields of the service call logs that match the service fields in the sub-alliance feature dataset. The number of service requests and the settlement amount in the candidate settlement records are used to calculate the sub-alliance revenue data. Subsequently, field coverage correction is performed based on the number of valid service fields covered by the sub-alliance feature dataset, generating the independent baseline value data for that sub-alliance. The independent baseline value data represents the observable revenue basis of the sub-alliance as a combination in historical service settlements, and is used to verify whether the sum of the initial independent value data of individual participants underestimates the actual performance of the combination.

[0031] Specifically, for the calculation of sub-alliances, the evaluation processor enumerates combinations of participants based on their identifiers, generating multiple sub-alliances. In practical applications, to control the computational scale, an upper limit on the combination order can be set based on the number of participants. For example, when the number of participants does not exceed 10, second- to fourth-order sub-alliances are enumerated; when the number of participants exceeds 10, participants with higher initial independent value data rankings and overlapping service fields are selected for higher-order combinations. This selection is still based on participant identifiers, initial independent value data, and the intersection of service fields, without changing the calculation scope of the sub-alliances. For each sub-alliance, the evaluation processor extracts the standard feature dataset corresponding to the participants within the sub-alliance, generating a sub-alliance feature dataset. The service fields in the sub-alliance feature dataset are matched with the settlement fields of the service call logs to obtain candidate settlement records. Matching can be done directly using service call numbers; in historical records where direct numbers are not available, a combination of service field combinations and time slices can also be used for matching.

[0032] The number of service requests and settlement amount in the candidate settlement records are used to calculate the sub-alliance revenue data. Sub-alliance revenue data can be categorized by... Calculate, where, Indicates alliance with sub-alliance The number of matching candidate settlement records. Indicates a sub-alliance, Indicates the first The number of service requests in each candidate settlement record. Indicates the first The settlement amount in the candidate settlement record, This represents the revenue amount generated by the sub-consortium under the candidate settlement records. To avoid sub-consortiums with insufficient field coverage from being overvalued due to accidental settlement records, the evaluation processor corrects the field coverage of the sub-consortium revenue data according to the number of effective service fields covered by the sub-consortium feature dataset. Let... The number of valid service fields covered by the sub-consortium feature dataset. The field coverage factor can be expressed as the total number of valid service fields invoked within the same target valuation period for a joint asset package. , This is a dimensionless proportion, with values ​​ranging from 0 to 1. The independent baseline value data for this sub-consortium can be expressed as: ,in, and All are monetary amounts. This is used to indicate the coverage of service fields in the sub-consortium. The physical meaning of this correction is to reduce the revenue amount in the candidate settlement record according to the range of service fields that the sub-consortium can actually support, so that the coverage of the independent baseline value data matches the coverage of the sub-consortium feature dataset.

[0033] Compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, mark the sub-alliance with the largest difference as the target reconstruction alliance. Furthermore, the sum of the initial independent value data of the participants within each sub-alliance is compared with the corresponding independent baseline value data. The sub-alliance with the largest difference is marked as the target reconstruction alliance, including: For each sub-alliance, the initial independent value data of each participant within the sub-alliance is aggregated to generate sub-alliance allocated value data; Sub-allocated value data of sub-alliances is subtracted from independent baseline value data of sub-alliances to generate sub-alliance difference data; Suballians with difference data less than or equal to zero are eliminated to obtain a set of candidate reconstructed alliances. When the candidate reconstruction alliance set is not empty, sort the sub-alliance difference data in the candidate reconstruction alliance set and extract the sub-alliance with the largest sub-alliance difference data; The extracted sub-alliance and its sub-alliance difference data are bound together to generate target reconstructed alliance data.

[0034] In one specific implementation, the evaluation processor further compares the sum of the initial independent value data of the participants within each sub-alliance with the corresponding independent baseline value data. For any sub-alliance, the evaluation processor adds the initial independent value data of all participants in that sub-alliance to obtain the sub-alliance allocated value data, and then subtracts the sub-alliance allocated value data from the independent baseline value data to obtain the sub-alliance difference data. If the sub-alliance difference data is less than or equal to zero, it indicates that the combined return of that sub-alliance is not higher than the current allocation result, and the evaluation processor removes it; if there are sub-alliances with sub-alliance difference data greater than zero, it indicates that these combinations may have synergistic value masked by individual elimination calculations. The evaluation processor sorts the sub-alliance difference data of the remaining sub-alliances and marks the sub-alliance with the largest difference as the target reconstruction alliance. This target reconstruction alliance is subsequently used as the central object for synergistic relationship identification, rather than simply directly compensating based on the difference, avoiding duplicate calculations caused by combination overlap.

[0035] Specifically, in the comparison phase, the evaluation processor aggregates the initial independent value data of each participant within each sub-consortium, generating sub-consortium allocation value data. If a certain sub-consortium... Including participants and participating parties Then its sub-alliance's value allocation data can be If more participants are involved, the values ​​are accumulated in the same way. The evaluation processor subtracts the allocated value data of the sub-alliance from its independent baseline value data to generate sub-alliance difference data. The sub-alliance difference data is an amount; a positive value indicates that the independent baseline value data exceeds the current allocated value, while a negative or zero value indicates that the current allocated value has already covered the independent baseline value data of the sub-alliance. After filtering out sub-alliances with difference data less than or equal to zero, the evaluation processor obtains a set of candidate reconstructed alliances. When the set of candidate reconstructed alliances is not empty, the evaluation processor sorts the sub-alliance difference data in descending order, extracts the sub-alliance with the largest difference data, and binds the extracted sub-alliance and its difference data to generate target reconstructed alliance data. The binding relationship retains the sub-alliance member identifier, sub-alliance difference data, corresponding independent baseline value data, and the list number of participants in this round, facilitating subsequent supernode backtracking and final splitting.

[0036] Extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and calculate the conditional mutual information value between the first feature vector and the second feature vector; Furthermore, the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants are extracted, and the conditional mutual information value between the first feature vector and the second feature vector is calculated, including: Extract the standard fields of each participant in the alliance from the standard feature dataset and concatenate them according to time slices to generate the first feature vector. Extract the standard fields of the remaining participants from the standard feature dataset, and generate second feature vectors for each of the same time slices; Generate a sequence of condition variables based on the service result field in the service call log; Binning encoding is performed on the first feature vector, the second feature vector, and the condition variable sequence to generate a discrete vector sequence; The joint occurrence frequency and conditional occurrence frequency of the discrete vector sequence are statistically analyzed, and the conditional mutual information value between the first eigenvector and each second eigenvector is calculated based on the joint occurrence frequency and conditional occurrence frequency.

[0037] In one specific implementation, after the target reconstruction consortium is determined, the evaluation processor extracts a first feature vector corresponding to the target reconstruction consortium from the standard feature dataset and a second feature vector from the standard feature dataset corresponding to the remaining participants. The first feature vector is formed by concatenating the standard fields of each participant in the target reconstruction consortium according to time slices, while the second feature vector corresponds to each remaining participant and is generated according to the same time slice. The evaluation processor generates a sequence of condition variables based on the service result fields in the service call log, for example, encoding service success, service failure, and service exception as different discrete states. The first feature vector, the second feature vector, and the sequence of condition variables are binned and encoded to form a discrete vector sequence. The evaluation processor counts the joint occurrence frequency and condition occurrence frequency of the discrete vector sequence, and calculates the conditional mutual information value between the first feature vector and each second feature vector accordingly. The conditional mutual information value is used to measure the degree of information dependency that remains between the target reconstruction consortium and a certain remaining participant when the service result field is given. One calculation method is as follows: ,in, This represents the discrete state of the first eigenvector. This represents the discrete state of a certain second eigenvector. Represents the discrete state of a sequence of condition variables. express The percentage of times they appear together Indicates in In state and The proportion of conditions that co-occur. and They represent in In state and The respective frequency percentages of each condition. The conditional mutual information value obtained by this formula is dimensionless data, and its physical meaning is the strength of the remaining information overlap between the two types of asset features under the constraints of the service result field. The larger the value, the stronger the collaborative dependency between the two under the same business result.

[0038] Specifically, after the target reconstruction alliance data is formed, the evaluation processor extracts the standard fields of each participant in the target reconstruction alliance from the standard feature dataset and concatenates them according to time slices to generate the first feature vector. If the target reconstruction alliance includes participant A and participant B, and within the same time slice T, participant A's standard fields are risk score and customer level, while participant B's standard fields are transaction amount and transaction frequency, then the first feature vector corresponding to this time slice can be concatenated in a fixed order of risk score, customer level, transaction amount, and transaction frequency. The standard fields of the remaining participants are used to generate second feature vectors according to the same time slice, and the vector order remains consistent throughout the entire target evaluation period to avoid the impact of field position changes on discrete statistics. The evaluation processor generates a sequence of conditional variables based on the service result fields in the service call log, for example, encoding successful service results as 1, failed service results as 0, and abnormal service results as 2; if the service result field is a continuous score, it can be divided into three states: low, medium, and high, according to the percentile.

[0039] During binning encoding, the evaluation processor processes the first eigenvector, the second eigenvector, and the condition variable sequence separately. Numerical standard fields can be binned using equal frequency or equal interval binning, with the number of bins ranging from 3 to 10, determined by the number of valid service call records within the target evaluation period. When the number of valid service call records is small, fewer bins are used to reduce estimation bias caused by excessively low joint occurrence frequency. Enumerated standard fields are directly treated as discrete states using enumeration codes. After binning encoding, the evaluation processor counts the joint occurrence frequency and conditional occurrence frequency of the discrete vector sequence and calculates the conditional mutual information value. The joint occurrence frequency refers to the number of times a certain discrete state of the first eigenvector, a certain discrete state of the second eigenvector, and a certain discrete state of the condition variable sequence co-occur within the same time slice; the conditional occurrence frequency refers to the number of times the first eigenvector and the second eigenvector co-occur within a certain discrete state of the condition variable sequence. To avoid the logarithm calculation being impossible due to zero occurrence frequency, the evaluation processor can add a smoothing count to each occurrence frequency, with the smoothing count set to a fixed value between 0.5 and 1. The conditional mutual information values ​​obtained in this way retain the statistical dependencies between asset features while eliminating some information that is interpreted by the service result fields themselves.

[0040] The remaining participants selected based on conditional mutual information values ​​are merged into the target reconstruction alliance to form super nodes, and the independent baseline value data of the super nodes are collaboratively decayed and corrected. Furthermore, the remaining participants selected based on conditional mutual information values ​​will be merged with the target reconstruction alliance into a super node, including: The target reconstruction alliance and each remaining participant are mapped as nodes, and the conditional mutual information values ​​are mapped as edge weights between nodes to generate collaborative dependency graph data. Extract target edges with edge weights greater than a preset threshold from collaborative dependency graph data; When the target edge exists, add the remaining participant identifiers connected by the target edge to the target reconstruction alliance identifier set to generate a super node member set; When the target edge does not exist, extract the remaining participant identifier with the largest edge weight and add it to the target reconstruction alliance identifier set to generate a super node member set. Summarize the standard feature datasets of each participant in the supernode member set to generate the supernode feature dataset; Establish the binding relationship between the supernode member set, the supernode feature dataset, and the corresponding independent baseline value data to generate supernode data.

[0041] Furthermore, a collaborative attenuation correction is applied to the independent baseline value data of supernodes, including: Read the conditional mutual information values ​​among the participants in the supernode member set, and calculate the average of the conditional mutual information values; Read the independent baseline value data of each participant in the supernode member set, and calculate the independent baseline value and value; The cooperative attenuation penalty coefficient is determined based on the position of the average conditional mutual information value in the preset attenuation mapping table; Multiply the sum of independent baseline values ​​by the co-attenuation penalty coefficient to obtain the supernode corrected baseline value data; Replace the independent baseline value data of the supernode with the corrected baseline value data of the supernode, and write the replaced supernode data into the list of participants in the next round.

[0042] In one specific implementation, after the conditional mutual information value is calculated, the evaluation processor merges the remaining participants selected based on the conditional mutual information value with the target reconstruction alliance into a super node, and performs collaborative decay correction on the independent baseline value data of the super node. Specifically, the evaluation processor maps the target reconstruction alliance and each remaining participant to nodes, and maps the conditional mutual information value to the edge weights between nodes, generating collaborative dependency graph data. In the collaborative dependency graph data, the remaining participants connected by target edges with edge weights greater than a preset threshold are added to the target reconstruction alliance identifier set, forming a super node member set. If there is no target edge with an edge weight greater than the preset threshold, the evaluation processor extracts the identifier of the remaining participant with the largest edge weight and adds it to the target reconstruction alliance identifier set to ensure that the current round of reconstruction can proceed along the strongest dependency direction. Subsequently, the standard feature datasets of each participant in the super node member set are aggregated into a super node feature dataset, and a binding relationship is established with the corresponding independent baseline value data to generate super node data. During collaborative decay correction, the evaluation processor reads the conditional mutual information values ​​among all participants in the supernode member set and calculates the average of these values. It then reads the independent baseline value data for each participant and calculates the sum of their independent baseline values. The collaborative decay penalty coefficient is determined based on the position of the average conditional mutual information value in a pre-defined decay mapping table. The sum of independent baseline values ​​is multiplied by the collaborative decay penalty coefficient to obtain the corrected baseline value data for the supernode. The collaborative decay penalty coefficient is a dimensionless ratio, while both the sum of independent baseline values ​​and the corrected baseline value data for the supernode are monetary values, eliminating any inconsistency in dimensions when multiplying them. This correction subtracts the overlapping information from the simple summation, allowing the supernode to participate in the calculation in the next round of traversal in a form closer to the true combined contribution.

[0043] Specifically, when generating collaborative dependency graph data, the evaluation processor maps the target reconstruction alliance to a node and each remaining participant to a separate node. The edge weights between nodes are the conditional mutual information values ​​between the target reconstruction alliance and the remaining participant. A preset threshold can be determined by the distribution of conditional mutual information values ​​in historical evaluation periods, for example, by taking the upper quartile of the conditional mutual information values ​​in the most recent three target evaluation periods, and limiting it to between 0.05 and 0.30. The upper quartile reflects the boundaries of strong dependencies, and the value range is used to avoid abnormal samples causing the threshold to be too high or too low. The evaluation processor extracts target edges with edge weights greater than the preset threshold from the collaborative dependency graph data. When a target edge exists, the identifiers of the remaining participants connected by the target edge are added to the target reconstruction alliance identifier set, generating a super node member set; when a target edge does not exist, the identifier of the remaining participant with the largest edge weight is extracted and added to the target reconstruction alliance identifier set, generating a super node member set. This latter process ensures that even when the target reconstruction alliance with the largest difference has appeared but has not reached the threshold, a reconstruction can still be completed along the direction closest to the collaborative dependency, avoiding a loop that remains in the same difference state. The evaluation processor aggregates the standard feature datasets of each participant in the supernode member set, generates the supernode feature dataset, and establishes the binding relationship between the supernode member set, the supernode feature dataset, and the corresponding independent baseline value data to form supernode data.

[0044] Cooperative decay correction occurs after supernode data generation. The evaluation processor reads the conditional mutual information values ​​between each participant in the supernode member set and calculates the average of the conditional mutual information values. If the supernode member set contains... If there are 1 participating party, then the number of conditional mutual information values ​​between each pair of participating parties is 1. The conditional mutual information value mean is the arithmetic mean of these conditional mutual information values. This mean is dimensionless and represents the overall information overlap strength within the supernode. The evaluation processor reads the independent baseline value data corresponding to each participant in the supernode member set and calculates the sum of independent baseline values. The preset attenuation map can be segmented according to the mean of the conditional mutual information value. For example, when the mean of the conditional mutual information value is less than 0.05, the cooperative attenuation penalty coefficient is 0.95; when the mean of the conditional mutual information value is not less than 0.05 and less than 0.15, the cooperative attenuation penalty coefficient is 0.85; when the mean of the conditional mutual information value is not less than 0.15, the cooperative attenuation penalty coefficient is 0.75. The evaluation processor determines the cooperative attenuation penalty coefficient according to the position of the mean of the conditional mutual information value in the preset attenuation map, and then multiplies the sum of independent baseline values ​​by the cooperative attenuation penalty coefficient to obtain the corrected baseline value data of the supernode. If the sum of independent baseline values ​​is denoted as... The cooperative decay penalty coefficient is denoted as Then the supernode corrected baseline value data is ,in, and All are monetary amounts. This is a dimensionless ratio. The physical meaning of this formula is to reduce the duplicate value generated by information overlap within a supernode, ensuring that when a supernode enters the next round of computation as a single participant, its value base is not simply equivalent to the mechanical sum of the independent baseline value data of its members. The evaluation processor replaces the independent baseline value data of the supernode with the corrected baseline value data of the supernode, and writes the replaced supernode data into the participant list for the next round.

[0045] Regenerate the initial independent value data for the supernode as a single participant and iterate back until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

[0046] Furthermore, output the decoupling value data for each participant, including: If a supernode exists in the last round of traversal, read the initial independent value data generated by the supernode in the last round; Read the changes in the target business indicators of each participant in the super node member set before the merger, as well as the conditional mutual information values ​​between them and the super node feature dataset; The changes in target business indicators and conditional mutual information values ​​are normalized and merged to generate data on the allocation ratio of super node members. The initial independent value data of the super node is split into member decoupled value data of each member participant according to the distribution ratio of the super node members; The initial independent value data of participants who have not entered the super node are merged with the member decoupling value data according to the participant identifier to generate the decoupling value data of each participant and output it.

[0047] In one specific implementation, after completing the supernode data generation and collaborative decay correction, the evaluation processor writes the supernode as a single participant into the next round's participant list and regenerates the initial independent value data, then iterates through the participant combinations again to generate sub-alliances. Here, the single participant does not change the identification relationships among the participants within the supernode member set; it simply uses the supernode feature dataset as the overall input in this round of calculation, avoiding the repeated splitting of strongly dependent assets across multiple rounds. The loop termination condition is that the sum of the initial independent value data of all participants in each sub-alliance is not less than the corresponding independent baseline value data. When this condition is met, the evaluation processor considers that there are no undervalued combinations in the current participant list that require further reconstruction, and then outputs the decoupled value data of each participant. If a supernode exists in the final round's participant list, the evaluation processor will also split the supernode's value back into the participating parties within the supernode member set, ensuring that the final output still uses the original participant identification as the granularity.

[0048] Specifically, in the next round of calculation, the supernode participates as a single participant in the regeneration of initial independent value data. The evaluation processor uses the supernode feature dataset along with the standard feature datasets of other participants who are not part of the supernode as new input, repeating the calculation of target business indicator changes, sub-alliance generation, independent baseline value data calculation, and sub-alliance difference data comparison. Since the binding relationship between the supernode member set, the supernode feature dataset, and the corresponding independent baseline value data has been preserved, the evaluation processor can treat the supernode as a whole in each iteration while retaining the member information required for the final split. The loop stops when the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data. At this point, the candidate reconstruction alliance set is empty, indicating that there is no sub-alliance with independent baseline value data higher than the currently assigned value, and the evaluation processor enters the decoupling value data output stage.

[0049] In the final output process, when a supernode exists in the last round of traversal, the evaluation processor reads the initial independent value data generated by the supernode in the last round. This initial independent value data represents the value allocation amount of the supernode as a whole in the final participant list. To split it into original participants, the evaluation processor reads the change in target business indicators of each participant in the supernode member set before merging, as well as the conditional mutual information value between each participant and the supernode feature dataset. The change in target business indicators reflects the marginal impact of member participants on service results, and the conditional mutual information value reflects the degree of information correlation between member participants and the supernode as a whole. Both types of data come from the aforementioned elimination calculation and conditional mutual information value calculation, respectively, and both have clear sources. The evaluation processor normalizes and merges the change in target business indicators and conditional mutual information values ​​to generate supernode member allocation proportion data. This can be achieved using... Indicates the first The fusion weight of each participating member, where... Indicates the first Normalized values ​​of changes in target business metrics for each participating member. Indicates the first Normalized values ​​of conditional mutual information between each participating member and the supernode feature dataset. This represents the integration ratio, which can be 0.5 or fixed by the company before the target evaluation period begins. , and All data are dimensionless. It is also a dimensionless weight, and its physical meaning is to simultaneously consider the marginal impact of business indicators and the degree of synergistic dependence under the same allocation scale. The evaluation processor then normalizes the fusion weights of each participating member to obtain the allocation ratio data of supernode members.

[0050] After the supernode member allocation percentage data is generated, the evaluation processor splits the supernode's initial independent value data into member decoupling value data for each participating party according to the supernode member allocation percentage data. If the supernode's initial independent value data is... , No. The data on the distribution ratio of supernode members among the participating members is as follows: Then the member decoupling value data of the participating party is ,in, and All are monetary amounts. As a dimensionless proportion, the physical meaning of this calculation is to decompose the value obtained by the supernode as a whole back to the original participants based on member contributions and collaborative relationships. For participants who did not enter the supernode, the evaluation processor directly reads their initial independent value data generated in the last round. Finally, the evaluation processor merges the initial independent value data of participants who did not enter the supernode with the member decoupling value data according to the participant identifier, generating and outputting the decoupling value data for each participant. The output data records may include participant identifier, decoupling value data, target evaluation period, total source revenue value data, whether they entered the supernode, and the corresponding supernode member set identifier, facilitating subsequent asset accounting, collaborative settlement, and evaluation traceability. Example

[0051] This application also discloses an enterprise digital asset valuation system based on multi-source data fusion.

[0052] Reference Figure 2 A system for assessing the value of enterprise digital assets based on multi-source data fusion, comprising: The data acquisition module is used to acquire the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs. The standardization processing module is used to perform field alignment and format standardization on the sub-asset feature datasets of each participant to generate a standard feature dataset; The calculation and generation module is used to remove the corresponding standard feature datasets one by one according to the participants, calculate the change of the target business indicators based on the service result fields in the service call log, and generate the initial independent value data of each participant according to the proportion of the change of the target business indicators and the total revenue value data. The traversal module is used to traverse the combination of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances. The comparison and judgment module is used to compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, the sub-alliance with the largest difference is marked as the target reconstruction alliance. The vector extraction module is used to extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and to calculate the conditional mutual information value between the first feature vector and the second feature vector. The merging module is used to merge the remaining participants selected based on conditional mutual information values ​​with the target reconstruction alliance into a super node, and to perform collaborative decay correction on the independent baseline value data of the super node. The output module is used to regenerate the initial independent value data of the super node as a single participant and return the traversal until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

[0053] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0054] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0055] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for evaluating the value of enterprise digital assets based on multi-source data fusion, characterized in that, include: Obtain the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs; Field alignment and format standardization are performed on the sub-asset feature datasets of each participant to generate a standard feature dataset; The corresponding standard feature datasets are removed one by one for each participant. The change in the target business indicator is calculated based on the service result field in the service call log. Initial independent value data for each participant is generated according to the proportion of the change in the target business indicator and the total revenue value data. Iterate through the combinations of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances; Compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, mark the sub-alliance with the largest difference as the target reconstruction alliance. Extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and calculate the conditional mutual information value between the first feature vector and the second feature vector; The remaining participants selected based on conditional mutual information values ​​are merged into the target reconstruction alliance to form super nodes, and the independent baseline value data of the super nodes are collaboratively decayed and corrected. Regenerate the initial independent value data for the supernode as a single participant and iterate back until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

2. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, Obtain service call logs for the joint asset package and the sub-asset feature datasets of each participant. Generate total revenue value data based on the service call logs, including: Read the service call logs of the joint asset package during the target assessment period; Extract the number of service requests, the settlement amount per request, and the settlement completion identifier from the service call log; Remove service call records with empty settlement completion flags to obtain valid service call records; The periodic revenue value is calculated based on the number of service requests and the settlement amount of a single request in the valid service call record, and the periodic revenue value is rounded to a predetermined number of places to generate the total revenue value data. Read the original business data tables submitted by each participant, and establish a mapping relationship between the original business data tables and the corresponding participants based on the participant identifiers; Extract business fields that match the joint asset package service call fields from the original business data table to generate a sub-asset feature dataset for each participant.

3. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, Field alignment and format standardization were performed on the sub-asset feature datasets of each participant to generate a standard feature dataset, including: Read the field names, field types, and field value ranges from the feature datasets of each sub-asset; Match the field names with the unified field directory to obtain the standard field names corresponding to each sub-asset feature dataset; Convert fields with inconsistent data types to a unified data type, perform unit conversion and fill in missing values ​​for the field value range, and generate candidate standard fields. Based on the service timestamps in the service call logs, the candidate standard fields are aligned by time slices according to the target evaluation period; The candidate standard fields, after being aligned by time slices, are recombined according to the participant identifiers to generate standard feature datasets for each participant.

4. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, The standard feature datasets corresponding to each participant are removed one by one. The change in the target business metric is calculated based on the service result field in the service call log. Initial independent value data for each participant is generated according to the percentage of change in the target business metric and the total revenue value data, including: Based on the matching records between the standard feature dataset of all participants and the service result field in the service call log, calculate the benchmark value of the target business indicator of the joint asset package within the target evaluation period; Remove the standard feature dataset of each individual participant sequentially from the standard feature dataset of all participants, while retaining the service result field within the same target evaluation period, and calculate the comparison value of the target business indicator after removal. The difference between the baseline value of the target business indicator and the comparison value of the target business indicator is determined as the change in the target business indicator of the participating party; The changes in the target business indicators of all participants are normalized to generate contribution percentage data for each participant. The contribution percentage of each participant is multiplied by the total revenue value to generate the initial independent value data for each participant.

5. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, The process iterates through the participating parties to generate sub-alliances. Based on the standard feature datasets within each sub-alliance and the settlement fields in the service call logs, it calculates independent baseline value data, including: The participants are combined and enumerated according to their identifiers to generate multiple sub-alliances; For each sub-alliance, extract the standard feature dataset corresponding to the participants within the sub-alliance to generate the sub-alliance feature dataset; Extract candidate settlement records from the settlement field of the service call log that match the service field in the sub-alliance feature dataset; Calculate the sub-alliance revenue data based on the number of service requests and settlement amount in the candidate settlement records; Based on the number of valid service fields covered by the sub-alliance feature dataset, the field coverage of the sub-alliance revenue data is corrected to generate the independent baseline value data of the sub-alliance.

6. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, Compare the sum of the initial independent value data of the participants in each sub-coalition with the corresponding independent baseline value data, and mark the sub-coalition with the largest difference as the target reconstruction coalition, including: For each sub-alliance, the initial independent value data of each participant within the sub-alliance is aggregated to generate sub-alliance allocated value data; Sub-allocated value data of sub-alliances is subtracted from independent baseline value data of sub-alliances to generate sub-alliance difference data; Suballians with difference data less than or equal to zero are eliminated to obtain a set of candidate reconstructed alliances. When the candidate reconstruction alliance set is not empty, sort the sub-alliance difference data in the candidate reconstruction alliance set and extract the sub-alliance with the largest sub-alliance difference data; The extracted sub-alliance and its sub-alliance difference data are bound together to generate target reconstructed alliance data.

7. The enterprise digital asset valuation method based on multi-source data fusion according to claim 1, characterized in that, Extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and calculate the conditional mutual information value between the first feature vector and the second feature vector, including: Extract the standard fields of each participant in the alliance from the standard feature dataset and concatenate them according to time slices to generate the first feature vector. Extract the standard fields of the remaining participants from the standard feature dataset, and generate second feature vectors for each of the same time slices; Generate a sequence of condition variables based on the service result field in the service call log; Binning encoding is performed on the first feature vector, the second feature vector, and the condition variable sequence to generate a discrete vector sequence; The joint occurrence frequency and conditional occurrence frequency of the discrete vector sequence are statistically analyzed, and the conditional mutual information value between the first eigenvector and each second eigenvector is calculated based on the joint occurrence frequency and conditional occurrence frequency.

8. The enterprise digital asset valuation method based on multi-source data fusion according to claim 7, characterized in that, The remaining participants selected based on conditional mutual information values ​​will be merged with the target reconstruction alliance into a super node, including: The target reconstruction alliance and each remaining participant are mapped as nodes, and the conditional mutual information values ​​are mapped as edge weights between nodes to generate collaborative dependency graph data. Extract target edges with edge weights greater than a preset threshold from collaborative dependency graph data; When the target edge exists, add the remaining participant identifiers connected by the target edge to the target reconstruction alliance identifier set to generate a super node member set; When the target edge does not exist, extract the remaining participant identifier with the largest edge weight and add it to the target reconstruction alliance identifier set to generate a super node member set. Summarize the standard feature datasets of each participant in the supernode member set to generate the supernode feature dataset; Establish the binding relationship between the supernode member set, the supernode feature dataset, and the corresponding independent baseline value data to generate supernode data.

9. The enterprise digital asset valuation method based on multi-source data fusion according to claim 8, characterized in that, Cooperative attenuation correction is applied to the independent baseline value data of supernodes, including: Read the conditional mutual information values ​​among the participants in the supernode member set, and calculate the average of the conditional mutual information values; Read the independent baseline value data of each participant in the supernode member set, and calculate the independent baseline value and value; The cooperative attenuation penalty coefficient is determined based on the position of the average conditional mutual information value in the preset attenuation mapping table; Multiply the sum of independent baseline values ​​by the co-attenuation penalty coefficient to obtain the supernode corrected baseline value data; Replace the independent baseline value data of the supernode with the corrected baseline value data of the supernode, and write the replaced supernode data into the list of participants in the next round.

10. A system for assessing the value of enterprise digital assets based on multi-source data fusion, applied to the enterprise digital asset valuation method based on multi-source data fusion as described in any one of claims 1-9, comprising: The data acquisition module is used to acquire the service call logs of the joint asset package and the sub-asset feature datasets of each participant, and generate total revenue value data based on the service call logs. The standardization processing module is used to perform field alignment and format standardization on the sub-asset feature datasets of each participant to generate a standard feature dataset; The calculation and generation module is used to remove the corresponding standard feature datasets one by one according to the participants, calculate the change of the target business indicators based on the service result fields in the service call log, and generate the initial independent value data of each participant according to the proportion of the change of the target business indicators and the total revenue value data. The traversal module is used to traverse the combination of participants to generate sub-alliances, and calculate independent baseline value data based on the standard feature dataset and settlement fields in the service call log within the sub-alliances. The comparison and judgment module is used to compare the sum of the initial independent value data of the participants in each sub-alliance with the corresponding independent baseline value data. When there is a sub-alliance whose sum of initial independent value data is less than the independent baseline value data, the sub-alliance with the largest difference is marked as the target reconstruction alliance. The vector extraction module is used to extract the first feature vector corresponding to the target reconstruction alliance and the second feature vector corresponding to the remaining participants, and to calculate the conditional mutual information value between the first feature vector and the second feature vector. The merging module is used to merge the remaining participants selected based on conditional mutual information values ​​with the target reconstruction alliance into a super node, and to perform collaborative decay correction on the independent baseline value data of the super node. The output module is used to regenerate the initial independent value data of the super node as a single participant and return the traversal until the sum of the initial independent value data of the participants in each sub-alliance is not less than the corresponding independent baseline value data, and output the decoupled value data of each participant.

Images