An AI digital transformation evaluation method and system based on multi-source data fusion

Abstract

The application discloses an AI digital transformation evaluation method and system based on multi-source data fusion, relates to the technical field of data fusion technology, and comprises the following steps: splicing a first business system positive yield record and a second business system negative cost record according to a preset common feature field to obtain a paired data set; determining a delay offset interval based on a business type attribute, performing an offset alignment operation on the occurrence time of the negative cost record to obtain a time sequence alignment data set; performing a dimensionless normalization mapping operation on the numerical value in the time sequence alignment data set to obtain a first standard yield value and a second standard cost value; performing difference calculation on the first standard yield value and the second standard cost value to obtain a net yield value, and determining an AI digital transformation evaluation grade according to a target numerical value interval in which the net yield value is located. The application correlates, aligns and offsets the cross-system yield and cost, and solves the problem of one-way distortion of the evaluation result.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, and in particular to an AI digital transformation assessment method and system based on multi-source data fusion. Background Technology

[0002] In the process of enterprise digital transformation, evaluating the actual benefits brought by AI applications is a core aspect of measuring the effectiveness of the transformation. Currently, the enterprise has deployed multiple independent business systems, such as a first business system responsible for sales records and a second business system responsible for after-sales records. The first business system records the positive benefits brought by AI recommendations, while the second business system records the negative costs caused by inaccurate recommendations.

[0003] However, the data between the first and second business systems is not shared, causing the evaluation system to only obtain data from one side. For example, if a company uses AI recommendation algorithms to push products, the first business system records an increase in customer clicks and purchases, and the evaluation system concludes that AI has brought about performance growth. However, records of returns and complaints due to inaccurate recommendations are stored in the second business system, and the evaluation system cannot access the negative cost records from the second business system.

[0004] The evaluation system calculated the AI's impact solely based on the positive revenue records of the first business system, ignoring the surge in complaints in the second business system. This one-sided calculation led the system to conclude that AI significantly increased revenue. Based on this conclusion, the company deemed its digital transformation successful, but in reality, hidden negative costs increased after-sales expenses and customer dissatisfaction.

[0005] Therefore, existing evaluation methods are limited by the fragmentation of multi-source systems and cannot correlate, align and offset positive benefits with negative costs, resulting in one-way distortion of AI digital transformation evaluation results. Summary of the Invention

[0006] In view of the aforementioned problems, this application is hereby filed.

[0007] Therefore, this application provides an AI digital transformation evaluation method and system based on multi-source data fusion, which can solve the problem that existing evaluation methods cannot correlate, align and offset positive benefits and negative costs across systems, resulting in one-way distortion of evaluation results.

[0008] To solve the above-mentioned technical problems, this application provides the following technical solution: In the first aspect, this application provides an AI digital transformation evaluation method based on multi-source data fusion, including: in response to obtaining positive revenue records of a first business system and negative cost records of a second business system for AI applications, performing concatenation on the positive revenue records of the first business system and the negative cost records of the second business system according to preset common feature fields to obtain a paired dataset; The delay offset interval is determined based on the business type attribute in the paired dataset. The offset alignment operation is then performed on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. Perform a dimensional normalization mapping operation on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. The difference between the first standard benefit value and the second standard cost value is calculated to obtain the net benefit value. Based on the target value range in which the net benefit value falls, the AI ​​digital transformation assessment level is determined.

[0009] Preferably, the positive revenue records of the first business system and the negative cost records of the second business system are concatenated according to preset common feature fields to obtain a paired dataset, including: Obtain the set of first item identifiers from the positive revenue record of the first business system and the set of second item identifiers from the negative cost record of the second business system; Perform an intersection extraction operation on the first set of item identifiers and the second set of item identifiers to obtain the target set of item identifiers; The positive revenue record of the first business system and the negative cost record of the second business system corresponding to the item identifier in the target item identifier set are identified as target pairing records; Data aggregation is performed based on the target paired record combination to generate a paired dataset containing positive revenue records of the first business system and negative cost records of the second business system.

[0010] Preferably, the delay offset interval is determined based on the business type attribute in the paired dataset, including: Obtain the set of historical business flow durations corresponding to the business type attributes in the paired dataset, and extract the value of each flow duration in the set of historical business flow durations; By comparing the various processing times in the historical business processing time set, the maximum and minimum processing times are extracted; The minimum turnover time is determined as the offset start point, and the maximum turnover time is determined as the offset end point; Perform a combined operation on the offset start point and offset end point to generate a delay offset interval covering the maximum and minimum turnaround times.

[0011] Preferably, an offset alignment operation is performed on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain a time-series aligned dataset, including: The original timestamp of the negative cost record of the second business system is superimposed with each time offset step in the delay offset interval to obtain multiple candidate aligned timestamps; Extract the target revenue timestamp that matches the candidate alignment timestamp from the positive revenue records of the first business system; The time offset step corresponding to the target revenue timestamp is determined as the target offset, and the time value corresponding to the target offset is extracted. Based on the time value corresponding to the target offset, the occurrence time of the negative cost record of the second business system is updated to obtain the time-series aligned dataset.

[0012] Preferably, a dimensional normalization mapping operation is performed on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value, including: Extract the revenue value distribution range of the positive revenue records of the first business system and the cost value distribution range of the negative cost records of the second business system from the time-series aligned dataset; Perform an alignment mapping operation between the upper limit of the profit distribution interval and the upper limit of the cost distribution interval, and perform an alignment mapping operation between the lower limit of the profit distribution interval and the lower limit of the cost distribution interval to generate a mapping function. Based on the mapping function, a conversion operation is performed on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system to obtain the first standard revenue value and the second standard cost value.

[0013] As a preferred method, a difference calculation is performed based on the first standard return value and the second standard cost value to obtain the net return value, including: Perform a subtraction operation on the first standard benefit value and the second standard cost value corresponding to the same association identifier in the time-series aligned dataset to obtain the net benefit value of a single record; The net profit values ​​of all individual records within the same evaluation period are summed to obtain the periodic net profit value. The periodic net income value is defined as the net income value, and the net income value is used as the basic data output for AI digital transformation assessment.

[0014] As a preferred approach, the AI ​​digital transformation assessment level is determined based on the target value range in which the net profit falls, including: The periodic net income value is compared with multiple preset income threshold ranges to determine the inclusion relationship between the periodic net income value and each income threshold range. Based on the inclusion relationship between the periodic net income value and each income threshold interval, determine the target income threshold interval into which the periodic net income value falls. Extract the transformation status labels associated with the target return threshold range, and output the transformation status labels in text form; The text-based transformation status labels are used to determine the AI ​​digital transformation assessment level, completing the assessment process based on multi-source data fusion.

[0015] Preferably, in response to obtaining the positive revenue record of the first business system and the negative cost record of the second business system for AI applications, the following are also included prior to this: Obtain the first user's identity identifier from the positive revenue record of the first business system and the second user's identity identifier from the negative cost record of the second business system; Perform a hash comparison operation on the first user's identity identifier and the second user's identity identifier to obtain the identity comparison result; In response to the identity comparison result being consistent, the first user identity identifier and the second user identity identifier are identified as preset common feature fields; In response to an inconsistency in identity verification results, the system obtains the first item attribute from the positive revenue record of the first business system and the second item attribute from the negative cost record of the second business system, performs a matching operation on the first item attribute and the second item attribute, and determines the successfully matched item attribute as a preset common feature field.

[0016] Preferably, in response to obtaining the positive revenue record of the first business system and the negative cost record of the second business system for AI applications, the following are included: In response to the order completion signal generated by the first business system, the transaction amount and recommendation source identifier associated with the order completion signal are obtained, and a positive revenue record for the first business system is generated based on the transaction amount and recommendation source identifier. In response to the customer complaint registration signal generated by the second business system, the system obtains the refund amount and customer complaint reason identifier associated with the customer complaint registration signal, and generates a negative cost record for the second business system based on the refund amount and customer complaint reason identifier.

[0017] Secondly, this application also provides an AI digital transformation evaluation system based on multi-source data fusion, including: an association splicing module, configured to respond to obtaining positive benefit records of a first business system and negative cost records of a second business system for AI applications, and to splice the positive benefit records of the first business system and the negative cost records of the second business system according to preset common feature fields to obtain a paired dataset; The time alignment module is configured to determine the delay offset interval based on the business type attribute in the paired dataset, and perform an offset alignment operation on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. The dimension mapping module is configured to perform dimension normalization mapping operations on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. The assessment module is configured to perform a difference calculation operation based on the first standard benefit value and the second standard cost value to obtain a net benefit value, and determine the AI ​​digital transformation assessment level based on the target value range in which the net benefit value falls.

[0018] Implementing this application will have the following beneficial effects: This application provides an AI digital transformation evaluation method and system based on multi-source data fusion. The method involves concatenating positive revenue records from a first business system and negative cost records from a second business system using preset common feature fields to obtain a paired dataset. A delay offset interval is determined based on business type attributes, and an offset alignment operation is performed on the occurrence time of the negative cost records to obtain a time-series aligned dataset. A dimensional normalization mapping operation is performed on the values ​​in the time-series aligned dataset to obtain a first standard revenue value and a second standard cost value. A net revenue value is obtained by calculating the difference between the first standard revenue value and the second standard cost value. The AI ​​digital transformation evaluation level is determined based on the target value range in which the net revenue value falls. This invention correlates, aligns, and offsets cross-system revenue and cost, solving the problem of one-way distortion in evaluation results. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an overall flowchart of an AI digital transformation assessment method based on multi-source data fusion, which is the subject of this application. Figure 2 This is a logic diagram of identity comparison and feature determination for an AI digital transformation assessment method based on multi-source data fusion, which is involved in this application. Figure 3 This application relates to a business benefit cost offsetting ecosystem diagram of an AI digital transformation assessment method based on multi-source data fusion. Figure 4This application relates to an assessment data flow and alignment diagram for an AI digital transformation assessment method based on multi-source data fusion. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] Example 1, referring to Figures 1-4 This is the first embodiment of the present invention, which provides an AI digital transformation assessment method based on multi-source data fusion, including: This invention provides a method that can effectively solve the problems mentioned above. The following will describe in detail how to implement this AI digital transformation assessment method based on multi-source data fusion, using multiple embodiments. Figure 1 A flowchart illustrating an AI-based digital transformation assessment method based on multi-source data fusion is shown, including: Step S1: In response to obtaining the positive revenue record of the first business system and the negative cost record of the second business system for AI applications, the positive revenue record of the first business system and the negative cost record of the second business system are concatenated according to the preset common feature fields to obtain a paired dataset. In the process of enterprise digital transformation, the positive benefits and negative costs generated by AI applications are often scattered in different business systems. Data silos lead to one-way distortion in evaluation. Therefore, this invention stitches together the benefit and cost records into a paired dataset based on common features, providing a complete data foundation for subsequent time-series alignment and offsetting calculations.

[0023] In a preferred embodiment, in response to obtaining a record of positive revenue from a first business system and a record of negative costs from a second business system for AI applications, the following steps are included: Step S1.1: In response to the first business system generating an order completion signal, obtain the transaction amount value and recommendation source identifier associated with the order completion signal, and generate a positive revenue record for the first business system based on the transaction amount value and recommendation source identifier; In this application, the system first listens for the order completion signal sent by the first business system, which represents the end of the life cycle of a transaction; it then parses the data packet of the order completion signal and extracts the transaction amount value and the recommendation source identifier carried in the data packet; it uses the transaction amount value as a quantitative indicator of revenue and the recommendation source identifier as a traceability pointer for related AI applications, and combines the transaction amount value and the recommendation source identifier to form a structured positive revenue record of the first business system.

[0024] Furthermore, when extracting the transaction amount, regular expressions are used to filter out non-numeric characters in the data packet, retaining only pure numbers and decimal point format to ensure the compatibility of numerical calculations. When extracting the recommendation source identifier, identify the algorithm version number suffix in the identifier string, extract the valid segments, and prevent matching failures caused by version number updates.

[0025] Accordingly, the generated positive revenue record of the first business system is assigned a unique primary key identifier. An index mapping relationship is established between the primary key identifier and the recommendation source identifier, which facilitates rapid retrieval based on the recommendation source identifier in subsequent steps.

[0026] Step S1.2: In response to the second business system generating a customer complaint registration signal, obtain the refund amount value and customer complaint reason identifier associated with the customer complaint registration signal, and generate a negative cost record for the second business system based on the refund amount value and customer complaint reason identifier; It should be noted that collecting data on after-sales issues caused by inaccurate AI recommendations is to quantify hidden losses. Specifically, it involves monitoring customer complaint registration signals generated by the second business system, which indicate that customers have initiated rights protection requests. Furthermore, the refund amount and the reason for the complaint are read from the customer complaint registration signal; Furthermore, it checks whether the refund amount exceeds the preset upper limit. If it does, it is truncated to the upper limit; otherwise, the original value is retained. Furthermore, identify the category code corresponding to the customer complaint reason identifier, determine whether the category code belongs to the category of AI recommendation error, and retain the customer complaint reason identifier if so, otherwise discard the current record. Furthermore, the verified refund amount and the filtered customer complaint reason identifier are encapsulated into a negative cost record for the second business system.

[0027] Furthermore, a timestamp is added to the negative cost records of the second business system. The timestamp records the precise number of milliseconds that the customer complaint registration signal was generated, which serves as the reference anchor point for subsequent time offset alignment, ensuring that the cost records have temporal traceability.

[0028] In a preferred embodiment, the positive revenue records of the first business system and the negative cost records of the second business system are concatenated according to preset common feature fields to obtain a paired dataset, including: Step S1.3: Obtain the first item identifier set in the positive revenue record of the first business system and the second item identifier set in the negative cost record of the second business system; When constructing the pairing dataset, the item identifier set is deconstructed into a set of the smallest logical units. If the positive revenue record of the first business system contains product SKU codes, then the first item identifier set is a discrete set composed of multiple independent SKU codes. Each SKU code in the set is a logical unit, carrying product category and specification attributes. Similarly, the set of second item identifiers in the negative cost record of the second business system consists of the defective product SKU codes involved in customer complaints.

[0029] Specifically, iterate through the positive revenue records of the first business system, extract the item identifier field embedded in each record, and gather the extracted item identifier fields to form the first item identifier set after deduplication; If a record contains a null value, it is skipped to ensure the validity of the collection elements.

[0030] Similarly, by traversing the negative cost records of the second business system, the item identifier fields are extracted, aggregated, and deduplicated to form the second item identifier set. The first item identifier set and the second item identifier set are independent at the logical level, and both are stored in a hash table structure at the physical level, which improves the comparison speed of subsequent intersection extraction operations.

[0031] Step S1.4: Perform an intersection extraction operation on the first item identifier set and the second item identifier set to obtain the target item identifier set; In detail, the intersection extraction operation is the convergence process of filtering out common elements from two independent universal sets. It uses the first set of item identifiers as the source for traversal and the second set of item identifiers as the target source for querying. Furthermore, each identifier element in the first item identifier set is extracted sequentially, and an existence query is performed in the hash table of the second item identifier set; If the query matches, the identifier element is determined to be a common element, and the common element is added to the target item identifier set; If the query does not find a match, the element is discarded. Furthermore, the traversal and query operations continue until all elements in the first item identifier set have been processed.

[0032] It should be noted that after the intersection extraction operation is completed, the set of target item identifiers is validated to remove dirty data identifiers with incorrect format or abnormal length, ensuring that each identifier in the set of target item identifiers is a valid object that exists in both business systems at the same time, thereby achieving the convergence of the data range from the broad whole set to the precise target subset.

[0033] Step S1.5: Determine the positive revenue record of the first business system and the negative cost record of the second business system corresponding to the item identifier in the target item identifier set as the target pairing record; Specifically, the target pairing record is a data combination of positive benefits and negative costs linked by the same item identifier.

[0034] It should be noted that the process of determining the target pairing record is a key step in establishing causal relationships. For any target identifier in the target item identifier set, all positive revenue records containing the target identifier are retrieved from the positive revenue record database of the first business system to form a subset of positive records. Retrieve all negative cost records containing the target identifier from the negative cost record database of the second business system to form a subset of negative records; perform a Cartesian product operation on the subset of positive records and the subset of negative records to generate all possible pairing combinations under the target identifier; Furthermore, by traversing all target identifiers in the target item identifier set and summarizing all generated pairing combinations, a complete set of target pairing records is formed. Each record in the complete set simultaneously possesses data characteristics from both a benefit perspective and a cost perspective, providing a closed data link for subsequent offsetting calculations.

[0035] Step S1.6: Perform data aggregation based on the target paired record combination to generate a paired dataset containing positive revenue records of the first business system and negative cost records of the second business system; In this embodiment, the pairing dataset is a two-dimensional data structure encoded by the revenue information dimension, cost information dimension, and association identifier dimension. In detail, the data aggregation operation arranges and combines the scattered target pairing records according to the preset table structure.

[0036] This application constructs a blank dataset containing a revenue field, a cost field, an item identifier field, and a timestamp field; and sequentially reads the target pairing records, writes the transaction amount value from the positive revenue record of the first business system into the revenue field, writes the refund amount value from the negative cost record of the second business system into the cost field, and writes the shared item identifier into the item identifier field; During the writing process, the consistency of field types is checked. If a type conflict occurs, the character type is forcibly converted to a floating-point type to ensure that the numeric field does not contain character data.

[0037] In one implementation, data aggregation can be achieved by concatenating memory arrays, for example, by loading target paired records into a memory array and merging them into a matrix structure using an array concatenation function.

[0038] In another implementation, data aggregation can also be achieved through database merge statements. For example, the JOIN operation in SQL can be used to join the positive revenue record table and the negative cost record table based on item identifiers to generate a paired dataset view.

[0039] For example, in an e-commerce scenario for smart home appliance sales, an AI recommendation system pushes a robot vacuum cleaner (model X1) and an air purifier (model Y2) to a user. When a user purchases the X1 robot vacuum cleaner, the first business system generates an order completion signal, extracts the transaction amount of 3000 yuan and the recommendation source identifier AI-REC-01, and generates a positive revenue record for the first business system. Simultaneously, the user discovers a defect in the X1 robot vacuum cleaner and files a complaint. The second business system generates a customer complaint registration signal, extracts the refund amount of 1000 yuan and the customer complaint reason identifier QUALITY-ERR, and generates a negative cost record for the second business system. The system obtains a first item identifier set containing X1 and Y2, and a second item identifier set containing X1. An intersection extraction operation is performed to obtain the target item identifier set X1. The revenue record and cost record corresponding to X1 are identified as the target paired record. Data aggregation is performed to generate a paired dataset, completing the initial fusion of cross-system data.

[0040] Step S2: Determine the delay offset interval based on the business type attribute in the paired dataset, and perform offset alignment operation on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. It should be noted that, due to the objective time lag in business processes, the time of generation of positive revenue records and the time of generation of negative cost records are usually misaligned. Direct calculation would lead to causal mismatch. Therefore, this invention restores the true temporal correspondence between revenue and cost by dynamically determining the delay offset interval and implementing time offset alignment, thereby eliminating the evaluation bias caused by the time difference.

[0041] In a preferred embodiment, determining the delay offset interval based on the business type attribute in the paired dataset includes: Step S2.1: Obtain the set of historical business flow durations corresponding to the business type attributes in the paired dataset, and extract the value of each flow duration in the set of historical business flow durations; Specifically, the essential logic of the historical business flow duration set reflects the objective delay pattern of a specific business type from the positive revenue node to the negative cost node.

[0042] In one alternative implementation, for each record in the paired dataset, the business type attribute is read, which represents the category or service category to which the product belongs. Using the business type attribute as the query key, retrieve multiple completed historical circulation cycles under the business type attribute from the historical business database; Furthermore, calculate the time difference between the revenue point and the cost point in each historical cycle, and aggregate all time differences into a set of historical business cycle durations; Furthermore, extract the individual transaction duration values ​​from the historical business transaction duration set and convert them into a unified time unit such as day or hour. This eliminates the incomparability of values ​​caused by inconsistent units, thereby revealing the essential characteristics of business transactions hidden beneath the timeline.

[0043] Step S2.2: Compare the various processing times in the historical business processing time set to extract the maximum and minimum processing times; Understandably, the comparison operation is essentially a traversal and extreme value determination of the numerical sequence of circulation time, with the initial value of the maximum circulation time set to 0 and the initial value of the minimum circulation time set to positive infinity. Read the processing time values ​​from the historical business processing time set sequentially. The circulation time value Compare with the current maximum turnover time; if the turnover time value... If the current maximum circulation time is greater than the current maximum circulation time, then update the current maximum circulation time to the circulation time value. ; At the same time, the circulation time value Compare with the current minimum turnaround time; if the turnaround time value... If the current minimum processing time is less than the current minimum processing time, then update the current minimum processing time to the current processing time value. .

[0044] In some instances, the maximum and minimum processing times obtained after traversal define the fluctuation range of time delay under a business type. In this embodiment, the aforementioned set of historical business processing times is obtained by querying the processing time records of all similar orders in the previous quarter in the business system database.

[0045] Step S2.3: Determine the minimum turnover time as the offset start point and the maximum turnover time as the offset end point; According to embodiments of the present invention, unidirectional time flow is an implicit premise, and the determination of the offset start point and offset end point must follow the principle of time irreversibility. That is, the offset start point represents the lower limit of the earliest possible time delay of the negative cost, and the offset end point represents the upper limit of the latest possible time delay of the negative cost.

[0046] It is easy to understand that assigning the minimum flow duration to the offset start point variable and the maximum flow duration to the offset end point variable can ensure that the offset interval covers all possible time spans of the business flow, preventing the omission of the actual cost occurrence time due to interval truncation, thereby providing a reasonable search boundary for subsequent time alignment.

[0047] Step S2.4: Perform a combined operation on the offset start point and offset end point to generate a delay offset interval covering the maximum and minimum turnover time. In a preferred implementation, the interval generation follows the closure principle, meaning that the delayed offset interval contains two boundary values: the offset start point and the offset end point.

[0048] It should be noted that the combination operation transforms discrete endpoint values ​​into a continuous interval range. Specifically, it requires constructing an interval data structure, using the offset start point as the left endpoint of the interval and the offset end point as the right endpoint; thus generating the delayed offset interval. ,in As the starting point of the offset, This is the endpoint of the offset.

[0049] In a specific business scenario, if the set of historical business processing times contains only one value, i.e., the maximum processing time equals the minimum processing time, then the delay offset interval degenerates into a single-point interval. In this case, the value is directly used as a fixed offset for alignment, thus demonstrating the applicability and flexibility of the rule.

[0050] In a preferred embodiment, an offset alignment operation is performed on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain a time-series aligned dataset, including: Step S2.5: Perform an overlay operation on the original timestamp of the negative cost record of the second business system and each time offset step in the delay offset interval to obtain multiple candidate aligned timestamps; It should be noted that the logic for deriving candidate alignment timestamps is based on a reverse backtracking search of the time axis. Within the delay offset interval, a discrete time offset step sequence is generated according to a preset step size. Since the cost lags behind the benefit, the time offset step size is set to a negative value sequence. The negative time offset step size is added to the original timestamp to generate a series of candidate alignment timestamps located on the historical time axis.

[0051] In some instances, the aforementioned candidate alignment timestamps can be obtained through the following specific calculation method, wherein: ; In the formula, Indicates the first Each candidate alignment timestamp has a value range of positive integers and its physical meaning is the possible matching time point obtained by superimposing negative offsets. This represents the original timestamp of the negative cost record in the second business system. The value range is positive integer, and the physical meaning is the precise time when the customer complaint registration signal was generated. It represents the time offset step size, with a value range of negative integers, and its physical meaning is the step granularity of searching back to history; This represents the step size index, which is a positive integer and physically represents the current search step index.

[0052] It should be noted that, according to the embodiments of this application, by adjusting... The value of makes such that It covers the entire delay offset interval, enabling step-by-step probing from the original timestamp to the historical time direction, ensuring the completeness of time-series backtracking.

[0053] Step S2.6: Extract the target revenue timestamp that matches the candidate alignment timestamp from the positive revenue records of the first business system; Preferably, filtering and selecting the uniquely matching target benefit timestamp from multiple candidate aligned timestamps is a filtering and optimization process.

[0054] In some optional implementations, the absolute value of the time distance between each candidate alignment timestamp and the revenue timestamp in the positive revenue record of the first business system is calculated; all candidate alignment timestamps are traversed to find the candidate alignment timestamp with the smallest absolute value of time distance. Determine if the absolute value of the minimum time distance is less than the preset time tolerance threshold. If it is, the match is successful, and the revenue timestamp corresponding to the candidate alignment timestamp is extracted as the target revenue timestamp. Otherwise, it is determined that there is no match.

[0055] Understandably, the time tolerance threshold ensures that small time deviations are allowed, improving the robustness of the matching. In other words, the extraction operation ensures that the cost and benefit are precisely linked in time.

[0056] Step S2.7: Determine the time offset step size corresponding to the target revenue timestamp as the target offset, and extract the time value corresponding to the target offset; It should be noted that the target offset is different from the time offset step during the traversal process. The target offset is the final offset determined after matching verification.

[0057] In detail, read the step size sequence number corresponding to the successful matching of the target revenue timestamp. Calculate the target offset Extract target offset The corresponding time value represents the actual lag time between the negative cost and the positive gain in the current pairing record.

[0058] In one possible implementation, if multiple target revenue timestamps match successfully, the time offset step corresponding to the one with the smallest absolute time distance is selected as the target offset, ensuring that the extracted time value is the optimal time series correction value.

[0059] Step S2.8: Based on the time value corresponding to the target offset, perform an update operation on the occurrence time of the negative cost record of the second business system to obtain the time-series aligned dataset; According to the embodiments of this application, the update operation ensures the consistency of the timing reference and ensures that the results of multiple executions under the same input are completely consistent. The present invention replaces the original timestamp in the negative cost record of the second business system with the sum of the time values ​​corresponding to the original timestamp and the target offset, and generates the corrected aligned timestamp. Furthermore, the original timestamp field is overwritten using the aligned timestamp to complete the time alignment of a single record; Furthermore, by iterating through all records in the paired dataset and performing the same update operation on each negative cost record, a time-aligned dataset is finally generated.

[0060] Preferably, the benefit time and cost time in the time-aligned dataset are logically on the same causal time axis, eliminating the net benefit calculation deviation caused by time misalignment.

[0061] For example, in an e-commerce scenario for smart home appliance sales, for the pairing record of an X1 robot vacuum cleaner, the business type attribute is large home appliance. A query of the historical database reveals that the customer complaint processing time for this type of product is typically between 7 and 15 days. Therefore, the offset start point is determined to be 7 days, and the offset end point to be 15 days, generating a delay offset interval of [7, 15] days. The original timestamp of a negative cost record is January 20th. A series of candidate aligned timestamps from January 27th to February 4th are generated with a step size of 1 day. Searching the positive revenue records of the first business system, a purchase record of X1 on January 15th is found, matching the candidate timestamp offset 5 days from January 20th, with a target offset of 5 days. The occurrence time of the negative cost record is updated to January 15th, consistent with the time-series aligned dataset records, restoring the causal relationship that the customer complaint originated from the purchase on that day.

[0062] Step S3: Perform a dimensional normalization mapping operation on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. Understandably, since the values ​​of benefits and costs often have different dimensions and orders of magnitude, directly calculating the difference will result in the larger value term dominating the result due to the difference in numerical scale. Therefore, the difference in numerical scale and dimension can be eliminated by the dimensional normalization mapping operation, which maps the values ​​of different dimensions to a unified standard space, thus ensuring the fairness and rationality of the evaluation calculation.

[0063] In a preferred embodiment, a dimensional normalization mapping operation is performed on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain a first standard revenue value and a second standard cost value, including: Step S3.1: Extract the distribution range of revenue values ​​for the positive revenue records of the first business system and the distribution range of cost values ​​for the negative cost records of the second business system from the time-series aligned dataset. In this application, it is necessary to first traverse the time-series aligned dataset, collect the values ​​of all positive revenue records of the first business system, and form a revenue value sequence. Furthermore, collect the values ​​of all negative cost records from the second business system to form a cost value sequence; Calculate the maximum and minimum values ​​of the profit value sequence to form the profit value distribution interval; Calculate the maximum and minimum values ​​of the cost value sequence to form the cost value distribution interval.

[0064] Correspondingly, when the profit or cost sequence is empty (i.e., no record is recorded within that period), the corresponding distribution interval is set to a preset default interval [0,1] to prevent division by zero errors or interval collapse caused by an empty set. In practice, the default interval setting in the case of an empty set ensures the stability and availability of subsequent mapping functions.

[0065] Step S3.2: Perform an alignment mapping operation on the upper limit of the profit value distribution interval and the upper limit of the cost value distribution interval, and perform an alignment mapping operation on the lower limit of the profit value distribution interval and the lower limit of the cost value distribution interval to generate a mapping function; It should be noted that the alignment mapping operation establishes a linear algebraic relationship between the source data space and the target standard space. The mapping function requires that the extreme points of the source space be accurately mapped to the extreme points of the target space to ensure the order preservation of normalization.

[0066] According to an embodiment of the present invention, the uniform dimension interval of the target standard space is set as [0,1]. The lower limit of the distribution interval of the revenue value is mapped to 0 and the upper limit is mapped to 1 to construct a revenue mapping function; the lower limit of the distribution interval of the cost value is mapped to 0 and the upper limit is mapped to 1 to construct a cost mapping function.

[0067] In this embodiment, the upper and lower limits of the above-mentioned profit distribution range are obtained by extracting extreme values ​​from the numerical fields of all positive profit records in the current period.

[0068] Step S3.3: Based on the mapping function, perform a conversion operation on the value of the positive revenue record of the first business system and the value of the negative cost record of the second business system to obtain the first standard revenue value and the second standard cost value. Preferably, a linear mapping function is selected to perform the transformation operation to ensure that the linear characteristics of the data distribution are not distorted.

[0069] In some optional specific implementations, the value of the positive revenue record of the first business system is substituted into the revenue mapping function to calculate the difference between the value and the lower limit of the revenue value distribution interval, and the ratio of the difference to the span of the revenue value distribution interval is calculated. The ratio result is determined as the first standard revenue value. Similarly, the negative cost record of the second business system is substituted into the cost mapping function to calculate the difference between the value and the lower limit of the cost value distribution interval. The ratio of the difference to the span of the cost value distribution interval is then calculated, and the ratio result is determined as the second standard cost value.

[0070] Understandably, the transformation operation eliminates the difference in dimensions and magnitude of the original values. Both the first standard benefit value and the second standard cost value are in the dimensionless space of [0,1], and are directly subtractable.

[0071] For example, in an e-commerce scenario for smart home appliance sales, the positive revenue record for the X1 robot vacuum cleaner is 3000 yuan, and the positive revenue record for the Y2 air purifier is 6000 yuan, with the revenue value distribution range [3000, 6000]. The negative cost record is 1000 yuan, and another cost record is 300 yuan, with the cost value distribution range [300, 1000]. An alignment mapping operation is performed, mapping 3000 yuan and 300 yuan to 0, and mapping 6000 yuan and 1000 yuan to 1. A conversion operation is performed on the 3000 yuan revenue, with a difference of 0, and the first standard revenue value is 0; a conversion operation is performed on the 6000 yuan revenue, with a difference of 3000, and the ratio to the span of 3000 is 1, so the first standard revenue value is 1; a conversion operation is performed on the 1000 yuan cost value, with a difference of 700, and the ratio to the span of 700 is 1, so the second standard cost value is 1, thus achieving dimensional unification.

[0072] Step S4: Perform a difference calculation operation based on the first standard benefit value and the second standard cost value to obtain the net benefit value, and determine the AI ​​digital transformation assessment level based on the target value range in which the net benefit value is located. Understandably, calculating net income is the core step in the evaluation. Simply adding up income cannot reflect the true effectiveness. This step achieves hedging through difference calculation and classifies grades based on the net results after hedging, thereby objectively and accurately reflecting the actual incremental value brought by AI applications and solving the problem of evaluation distortion.

[0073] In a preferred embodiment, a difference calculation operation is performed based on a first standard return value and a second standard cost value to obtain a net return value, including: Step S4.1: Perform a subtraction operation on the first standard benefit value and the second standard cost value corresponding to the same association identifier in the time-series aligned dataset to obtain the net benefit value of a single record; When calculating the net benefit value of a single record, if the first standard benefit value is greater than the second standard cost value, the net benefit value of the single record is determined to be positive, indicating that the AI ​​application has a positive empowering effect on the single record. If the first standard return value is less than the second standard cost value, then the net return value of a single record is determined to be negative, indicating that the AI ​​application generates negative losses on a single record. If the first standard benefit value equals the second standard cost value, then the net benefit value of a single record is determined to be zero, indicating that the benefit and cost of AI application on a single record are equal.

[0074] Specifically, the net profit for a single record is calculated as follows: ; In the formula, This represents the net profit value for a single record, with a value range of [-1, 1]. Its physical meaning is the standardized net profit level in a single business interaction. This represents the first standard return value, with a value range of [0,1]. Its physical meaning is the normalized positive return indicator. It represents the second standard cost value, with a value range of [0,1], and its physical meaning is the normalized negative cost index.

[0075] It should be noted that the large-scale customer complaints and refunds caused by non-AI factors (such as logistics strikes) have resulted in an abnormally high cost. Directly calculating this cost would lead to a misjudgment of the AI ​​transformation. Therefore: When calculating the net profit of a single record, if Greater than 0.9 and Greater than 0.5 and If it is less than 0, then Forced to be set to 0; like Greater than 0.9 and Less than 0.5 and If the difference is less than 0, the original difference is retained as... ; like Greater than 0.9 and Greater than 0.9 and If the value is 0, it is considered a normal fluctuation; like If the difference is less than 0.1, the difference will be calculated using normal logic.

[0076] when When it is in the range of 0.2 to 0.8, if In the range of 0 to 0.2 and If it is greater than 0, then for Apply a reward coefficient of 0.1; like In the range of 0.8 to 1 and If less than 0, then for Apply a penalty coefficient of 0.9; If the difference calculation result overflows the interval [-1, 1], it is truncated to the boundary value.

[0077] Step S4.2: Perform an accumulation calculation on the net profit values ​​of all individual records within the same evaluation period to obtain the periodic net profit value; In simple terms, cumulative calculation is a transformation of the data flow path that aggregates the effects of individual micro-level records into macro-level cyclical indicators.

[0078] This invention actively retrieves the net profit values ​​of all individual records within the same evaluation period to construct a net profit value sequence; it then iterates through the net profit value sequence and adds the net profit value of each individual record to the period summary variable. Furthermore, the initial value of the periodic summary variable is 0. For each individual record's net profit value read, the sum of the periodic summary variable and the individual record's net profit value is calculated, and the result is written back to the periodic summary variable. This process continues until the sequence is traversed, and the final value of the periodic summary variable is the periodic net profit value.

[0079] Step S4.3: Determine the periodic net profit value as the net profit value, and use the net profit value as the basic data output for AI digital transformation assessment. In one alternative implementation, the normal output path is to write the net profit value into the corresponding field of the evaluation result database. When the write operation encounters a database connection exception, an exception branch is triggered, the net profit value is temporarily stored in the local message queue, and a retry mechanism is started. It should be noted that the net profit figure, as the basic data output, ensures that the downstream links of the evaluation process can obtain a stable and reliable data source, avoiding the interruption of the entire evaluation chain due to a single point of failure.

[0080] In a preferred embodiment, the AI ​​digital transformation assessment level is determined based on the target value range in which the net revenue value falls, including: Step S4.4: Perform a numerical comparison operation between the periodic net income value and multiple preset income threshold intervals to determine the inclusion relationship between the periodic net income value and each income threshold interval. It should be noted that the net return value of the period belongs to the continuous numerical domain, while the return threshold range belongs to the discrete classification domain. The numerical comparison operation is a mapping process that crosses the logical domain.

[0081] Specifically, it reads multiple preset profit threshold ranges, each of which is defined by a lower limit and an upper limit; it then determines whether the net profit value for the period is greater than or equal to the lower limit of the range and less than or equal to the upper limit of the range. In one optional implementation, if the periodic net income value is exactly equal to the boundary value of the adjacent interval, the periodic net income value is determined to belong to the interval with the larger value according to the left-closed and right-open principle, thereby eliminating the problem of ambiguity in the boundary value attribution. In this application, the above-mentioned preset multiple income threshold intervals are obtained by analyzing the percentile of the distribution of historical net income data in the industry.

[0082] Step S4.5: Determine the target return threshold range into which the periodic net return value falls, based on the inclusion relationship between the periodic net return value and each return threshold range. Furthermore, the absolute value of the periodic net income determines the relative position of the periodic net income in the threshold system. An inclusion relationship means that the periodic net income falls within a certain income threshold range.

[0083] In one implementation, if the periodic net income value has only one inclusion relationship with an income threshold range, then the income threshold range is the target income threshold range. If the inclusion relationship is not unique due to the overlap of intervals, the interval with the smallest revenue threshold range is selected as the target revenue threshold interval to ensure the highest positioning accuracy. In other words, the comparison of absolute values ​​is ultimately transformed into the locking of relative intervals.

[0084] Step S4.6: Extract the transformation status labels associated with the target revenue threshold range and output the transformation status labels in text form; If the target return threshold range is successfully matched and the associated transformation status label exists in the label dictionary, then the transformation status label is extracted. When the transition status label is in numeric encoding format, the encoding conversion table is invoked to map the numeric encoding into a text form of natural language description; If the associated transformation state label does not exist, output the unknown state text; In some embodiments, the text output follows a uniform encoding format such as UTF-8 to prevent garbled characters and ensure the accurate transmission of tag semantics. Tag extraction and conversion achieve the final leap from the numerical domain to the semantic domain.

[0085] Step S4.7: Determine the transformation status label in text form as the AI ​​digital transformation assessment level, and complete the assessment process based on multi-source data fusion; This invention provides a method for determining the evaluation level of AI digital transformation, which solves the problem of one-way distortion of evaluation results caused by judging the transformation effectiveness based solely on one-sided benefits in the existing technology. Furthermore, by integrating multi-source data and aligning the net benefit values ​​obtained from offset calculations, it truly reflects the comprehensive effectiveness of AI applications. Furthermore, the transformation status label is used as an assessment level to intuitively present the health of digital transformation; It is important to understand that, compared to the inflated assessments that relied solely on positive returns before the improvement, the improved assessment ratings, by incorporating a negative cost hedging mechanism, can effectively identify pseudo-successful transformations that involve high returns accompanied by high costs. This avoids strategic misjudgments by enterprises based on erroneous assessment conclusions, thereby ensuring that digital transformation proceeds along a path of substantial efficiency gains.

[0086] For example, in an e-commerce scenario for smart home appliance sales, for a record of an X1 robot vacuum cleaner, the first standard benefit value is 0.5, and the second standard cost value is 0.8. Subtraction is performed to obtain a net benefit value of -0.3 for a single record. The net benefit values ​​of all records within this period are summed to obtain a periodic net benefit value of 0.6. The preset benefit threshold range includes [0, 0.3] corresponding to inefficient transformation, [0.3, 0.7] corresponding to stable transformation, and [0.7, 1] corresponding to excellent transformation. The periodic net benefit value of 0.6 falls within the [0.3, 0.7] range, so the target benefit threshold range is determined to be [0.3, 0.7]. The associated transformation status label "stable transformation" is extracted and converted into text output, which is determined as the AI ​​digital transformation evaluation level. This objectively reflects the current situation where AI recommendations, while increasing sales, have resulted in net benefits not reaching the excellent level due to increased customer complaints.

[0087] In a preferred embodiment, in response to obtaining the positive revenue record of the first business system and the negative cost record of the second business system for the AI ​​application, the process further includes: Step S5: Obtain the first user identity identifier of the positive revenue record of the first business system and the second user identity identifier of the negative cost record of the second business system; It is also important to understand that in the operation of a business system, user identity is a key link connecting different business processes. The S5 designed in this invention extracts identity identifiers belonging to different systems, providing a prerequisite for subsequent cross-system identity verification and data splicing, thereby achieving accurate association based on the user dimension.

[0088] Step S5.1: Perform a hash comparison operation on the first user identity identifier and the second user identity identifier to obtain the identity comparison result; In one implementation, the hash comparison operation is designed based on the principle of symmetry to ensure the fairness and consistency of the two identifiers during the comparison process. It should be noted that the first hash digest of the first user identifier and the second hash digest of the second user identifier are calculated separately; and each character bit of the first hash digest and the second hash digest are compared. If all characters are the same, the identity comparison result is consistent; If at least one character is different, the identity comparison result is inconsistent.

[0089] In other words, hash comparison avoids the risks of plaintext transmission and comparison, while improving comparison efficiency through fixed-length digests, ensuring that even if there are superficial differences in the identifier format, as long as the substantive content is the same, it can be determined that they are consistent.

[0090] Step S5.2: In response to the identity comparison result being consistent, the first user identity identifier and the second user identity identifier are determined as preset common feature fields; In one optional implementation, when the identity comparison results are consistent, it indicates that the same user has left a mapping trace in two business systems. Optionally, for the extreme case caused by hash collision, that is, different identifiers generate the same hash digest, it is necessary to introduce an auxiliary verification field, such as the last four digits of the mobile phone number, for secondary confirmation. For example, if the second confirmation is successful, the first user identity identifier and the second user identity identifier are set as preset common feature fields. The user identity becomes a bridge connecting positive benefits and negative costs. In this embodiment, the first user identity identifier and the second user identity identifier are obtained through the user primary key of the encryption and decryption business system.

[0091] Step S5.3: In response to the identity comparison result being inconsistent, obtain the first item attribute of the positive revenue record of the first business system and the second item attribute of the negative cost record of the second business system, perform a matching operation on the first item attribute and the second item attribute, and determine the successfully matched item attribute as a preset common feature field; Furthermore, when the identity comparison results are inconsistent, it means that a connection cannot be established at the user level. In this case, it is necessary to map downwards, reducing the dimension from the user level to the item level.

[0092] It is understandable that user identity is a macro-level summary of the business entity, while item attributes are a micro-level description of the business object. In one implementation, it is necessary to parse the positive revenue records of the first business system to extract the first item attributes, and parse the negative cost records of the second business system to extract the second item attributes. Furthermore, the semantic similarity score between the first item attribute and the second item attribute is calculated. If the semantic similarity score is greater than the set similarity threshold, the item attribute is determined to be successfully matched. The successfully matched item attribute is then promoted to a preset common feature field to ensure that an association alignment relationship can still be established based on the item dimension even when the user's identity is invalid.

[0093] Figure 2 The logic for determining shared feature fields across systems is as follows: For positive benefit records and negative cost records of AI applications, the user identity identifiers of both ends are first extracted and a hash comparison operation is performed. If the identity comparison results match, the association feature of the user identity dimension is directly established; if the comparison results do not match, a degradation processing mechanism is triggered, and the first item attribute and the second item attribute are obtained and semantic matching is performed. If the match is successful, the association feature of the item attribute dimension is established; if it fails, the unrelated records are removed. This mechanism ensures the accurate anchoring of the association link between heterogeneous data sources, provides a reliable field basis for the subsequent association and splicing of benefits and costs, and avoids evaluation data mismatch caused by association failure.

[0094] Figure 3 This represents an abstract mapping relationship for offsetting benefits and costs within the business ecosystem. When a customer's business role triggers an AI recommendation application, positive benefit records and negative cost records are generated. This dual-source data is aggregated through shared features and enters an asynchronous alignment space of time series and dimensions: one side determines the delay offset interval and generates a mapping function, while the other side performs time series offset alignment and dimension normalization mapping. The standardized data, after cross-domain calibration, is finally fed into the difference offset calculation node. Through hedging, the true net benefit is derived, thereby determining the target numerical range and outputting an objective transformation status label. This achieves a dimensional upgrade in business evaluation logic, moving from fragmented and distorted data to integrated and truthful assessment.

[0095] Figure 4 To evaluate the flow and alignment trajectory of data during multi-source fusion, revenue and cost records from dual business systems were extracted and aggregated using target item identifiers to generate paired datasets. The data flow was then processed in parallel on two paths: one path addressed temporal misalignment by overlaying time offset steps and extracting target revenue timestamps to generate a time-aligned dataset; the other path addressed dimensional differences by aligning mapping extreme points and performing dimensional transformation operations to achieve numerical normalization. The final data, after time-alignment and dimensional unification, was incorporated into a subtraction operation, accumulated to obtain the periodic net revenue value, and finally, a threshold comparison was used to determine the transformation status label, fully presenting the complete evolution of data from fragmented and heterogeneous to unified alignment and offsetting.

[0096] Example 2, this example also provides an AI digital transformation evaluation system based on multi-source data fusion, including: an association splicing module, configured to respond to obtaining positive revenue records of a first business system and negative cost records of a second business system for AI applications, and to splice the positive revenue records of the first business system and the negative cost records of the second business system according to preset common feature fields to obtain a paired dataset; The time alignment module is configured to determine the delay offset interval based on the business type attribute in the paired dataset, and perform an offset alignment operation on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. The dimension mapping module is configured to perform dimension normalization mapping operations on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. The assessment module is configured to perform a difference calculation operation based on the first standard benefit value and the second standard cost value to obtain a net benefit value, and determine the AI ​​digital transformation assessment level based on the target value range in which the net benefit value falls.

[0097] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0098] 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 application.

[0099] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An AI-based digital transformation evaluation method based on multi-source data fusion, characterized in that, include: In response to obtaining the positive revenue records of the first business system and the negative cost records of the second business system for AI applications, the positive revenue records of the first business system and the negative cost records of the second business system are concatenated according to preset common feature fields to obtain a paired dataset; The delay offset interval is determined based on the business type attribute in the paired dataset. The offset alignment operation is then performed on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. Perform a dimensional normalization mapping operation on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. The difference between the first standard benefit value and the second standard cost value is calculated to obtain the net benefit value. Based on the target value range in which the net benefit value falls, the AI ​​digital transformation assessment level is determined.

2. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 1, characterized in that: Based on preset common feature fields, the positive revenue records of the first business system and the negative cost records of the second business system are concatenated to obtain a paired dataset, including: Obtain the set of first item identifiers from the positive revenue record of the first business system and the set of second item identifiers from the negative cost record of the second business system; Perform an intersection extraction operation on the first set of item identifiers and the second set of item identifiers to obtain the target set of item identifiers; The positive revenue record of the first business system and the negative cost record of the second business system corresponding to the item identifier in the target item identifier set are identified as target pairing records; Data aggregation is performed based on the target paired record combination to generate a paired dataset containing positive revenue records of the first business system and negative cost records of the second business system.

3. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 1, characterized in that: The delay offset interval is determined based on the business type attribute in the paired dataset, including: Obtain the set of historical business flow durations corresponding to the business type attributes in the paired dataset, and extract the value of each flow duration in the set of historical business flow durations; By comparing the various processing times in the historical business processing time set, the maximum and minimum processing times are extracted; The minimum turnover time is determined as the offset start point, and the maximum turnover time is determined as the offset end point; Perform a combined operation on the offset start point and offset end point to generate a delay offset interval covering the maximum and minimum turnaround times.

4. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 3, characterized in that: Based on the delay offset interval, perform offset alignment operations on the occurrence time of the negative cost records of the second business system to obtain a time-series aligned dataset, including: The original timestamp of the negative cost record of the second business system is superimposed with each time offset step in the delay offset interval to obtain multiple candidate aligned timestamps; Extract the target revenue timestamp that matches the candidate alignment timestamp from the positive revenue records of the first business system; The time offset step corresponding to the target revenue timestamp is determined as the target offset, and the time value corresponding to the target offset is extracted. Based on the time value corresponding to the target offset, the occurrence time of the negative cost record of the second business system is updated to obtain the time-series aligned dataset.

5. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 1, characterized in that: Perform a dimensional normalization mapping operation on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value, including: Extract the revenue value distribution range of the positive revenue records of the first business system and the cost value distribution range of the negative cost records of the second business system from the time-series aligned dataset; Perform an alignment mapping operation between the upper limit of the profit distribution interval and the upper limit of the cost distribution interval, and perform an alignment mapping operation between the lower limit of the profit distribution interval and the lower limit of the cost distribution interval to generate a mapping function; Based on the mapping function, a conversion operation is performed on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system to obtain the first standard revenue value and the second standard cost value.

6. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 5, characterized in that: The net profit is calculated by performing a difference calculation between the first standard return value and the second standard cost value, including: Perform a subtraction operation on the first standard benefit value and the second standard cost value corresponding to the same association identifier in the time-series aligned dataset to obtain the net benefit value of a single record; The net profit values ​​of all individual records within the same evaluation period are summed to obtain the periodic net profit value. The periodic net income value is defined as the net income value, and the net income value is used as the basic data output for AI digital transformation assessment.

7. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 6, characterized in that: Based on the target value range in which the net income value falls, the AI ​​digital transformation assessment level is determined, including: The periodic net income value is compared with multiple preset income threshold ranges to determine the inclusion relationship between the periodic net income value and each income threshold range. Based on the inclusion relationship between the periodic net income value and each income threshold interval, determine the target income threshold interval into which the periodic net income value falls. Extract the transformation status labels associated with the target return threshold range, and output the transformation status labels in text form; The text-based transformation status labels are used to determine the AI ​​digital transformation assessment level, completing the assessment process based on multi-source data fusion.

8. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 1, characterized in that: In response to obtaining records of positive revenue from the first business system and negative costs from the second business system for AI applications, previous steps included: Obtain the first user's identity identifier from the positive revenue record of the first business system and the second user's identity identifier from the negative cost record of the second business system; Perform a hash comparison operation on the first user's identity identifier and the second user's identity identifier to obtain the identity comparison result; In response to the identity comparison result being consistent, the first user identity identifier and the second user identity identifier are identified as preset common feature fields; In response to an inconsistency in identity verification results, the system obtains the first item attribute from the positive revenue record of the first business system and the second item attribute from the negative cost record of the second business system, performs a matching operation on the first item attribute and the second item attribute, and determines the successfully matched item attribute as a preset common feature field.

9. The AI ​​digital transformation evaluation method based on multi-source data fusion as described in claim 1, characterized in that: In response to obtaining records of positive benefits from the first business system and negative costs from the second business system for AI applications, including: In response to the order completion signal generated by the first business system, the transaction amount and recommendation source identifier associated with the order completion signal are obtained, and a positive revenue record for the first business system is generated based on the transaction amount and recommendation source identifier. In response to the customer complaint registration signal generated by the second business system, the system obtains the refund amount and customer complaint reason identifier associated with the customer complaint registration signal, and generates a negative cost record for the second business system based on the refund amount and customer complaint reason identifier.

10. An AI digital transformation evaluation system based on multi-source data fusion, employing the AI ​​digital transformation evaluation method based on multi-source data fusion as described in any one of claims 1 to 9, characterized in that, include: The association and splicing module is configured to, in response to obtaining the positive revenue records of the first business system and the negative cost records of the second business system for AI applications, splice the positive revenue records of the first business system and the negative cost records of the second business system according to preset common feature fields to obtain a paired dataset; The time alignment module is configured to determine the delay offset interval based on the business type attribute in the paired dataset, and perform an offset alignment operation on the occurrence time of the negative cost record of the second business system according to the delay offset interval to obtain the time-series aligned dataset. The dimension mapping module is configured to perform dimension normalization mapping operations on the values ​​of the positive revenue records of the first business system and the negative cost records of the second business system in the time-series aligned dataset to obtain the first standard revenue value and the second standard cost value. The assessment module is configured to perform a difference calculation operation based on the first standard benefit value and the second standard cost value to obtain a net benefit value, and determine the AI ​​digital transformation assessment level based on the target value range in which the net benefit value falls.

Images