User feature evaluation method and device based on cross-domain data fusion and electronic equipment

Abstract

The application discloses a user feature evaluation method and device based on cross-domain data fusion and electronic equipment, and relates to the field of data processing. The method comprises the following steps: receiving an evaluation request about user features from a business system; extracting first domain feature labels of a user in a first data domain according to a desensitized user identifier in the evaluation request, wherein the first data domain is an aviation travel data domain; initiating a collaborative computing request to a second data domain through an identity matching protocol based on the desensitized user identifier, obtaining second domain feature labels of the user in the second data domain, and wherein the second data domain is a communication behavior data domain; determining a comprehensive evaluation value according to the first domain feature labels and the second domain feature labels; and generating a decision label according to the comparison result of the comprehensive evaluation value and a preset threshold. The application solves the problem that the existing passenger evaluation method cannot fuse aviation behavior and communication behavior features across industries without exchanging detailed data, resulting in data blind spots in the evaluation result and poor information security.

Description

Technical Field

[0001] This application relates to the field of data processing, and more specifically, to a user feature evaluation method, apparatus, and electronic device based on cross-domain data fusion. Background Technology

[0002] In user characteristic assessment, traditional methods typically rely solely on historical behavioral data from a single data source. For example, in the air transport industry, passenger value assessments are often based only on frequent flyer data accumulated by the airline itself or historical reservation data from its global distribution system. When using such methods, the accuracy of the assessment results is limited because the data sources are confined to the aviation sector, resulting in a limited range of user characteristic dimensions.

[0003] For example, relying solely on internal data such as passengers' historical flight frequency and cabin class for value assessment cannot reflect passengers' spending power and travel intentions in non-aviation scenarios. Furthermore, user behavior data held by external data sources such as telecommunications operators is detailed data, and acquiring or exchanging such data requires meeting strict privacy protection compliance requirements, resulting in poor information security. Therefore, under current technological conditions, airlines can only conduct passenger value assessments based on limited internal data and cannot overcome the limitations of data sources.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a user feature assessment method, apparatus, and electronic device based on cross-domain data fusion, which at least solves the technical problem that existing passenger assessment methods cannot integrate aviation behavior and communication behavior features across industries without exchanging detailed data, resulting in data blind spots and poor information security in the assessment results.

[0006] According to one aspect of the embodiments of this application, a user feature evaluation method based on cross-domain data fusion is provided, comprising: receiving an evaluation request for user features from a business system; extracting a first domain feature label corresponding to the user in a first data domain according to a de-identified user identifier in the evaluation request, wherein the first data domain is an air travel data domain, and the first domain feature label includes at least one of the user's flight activity, cabin class preference, and route preference; initiating a collaborative computing request to a second data domain based on the de-identified user identifier through an identity matching protocol to obtain a second domain feature label of the user in the second data domain, wherein the second data domain is a communication behavior data domain, and the second domain feature label includes at least one of the user's spending power level and activity level of travel-related applications; determining a comprehensive evaluation value according to the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label; generating a decision label according to the comparison result of the comprehensive evaluation value and a preset threshold, wherein the decision label is a discrete business execution instruction, and the decision label does not contain the original information of the first domain feature label or the second domain feature label.

[0007] According to another aspect of the embodiments of this application, a user feature evaluation apparatus based on cross-domain data fusion is also provided, comprising: a request receiving unit, configured to receive an evaluation request for user features from a business system; a first tag acquisition unit, configured to extract a first domain feature tag corresponding to the user in a first data domain based on a de-identified user identifier in the evaluation request, wherein the first data domain is an air travel data domain, and the first domain feature tag includes at least one of the user's flight activity, cabin class preference, and route preference; and a second tag acquisition unit, configured to initiate a collaborative computing request to the second data domain based on the de-identified user identifier through an identity matching protocol, and acquire the user's second domain feature tag. According to the second domain feature label within the domain, wherein the second data domain is a communication behavior data domain, the second domain feature label includes at least one of the user's consumption capacity level and the activity level of travel-related applications; the evaluation processing unit is used to determine a comprehensive evaluation value based on the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label; the decision processing unit is used to generate a decision label based on the comparison result of the comprehensive evaluation value and a preset threshold, wherein the decision label is a discrete business execution instruction, and the decision label does not contain the original information of the first domain feature label or the second domain feature label.

[0008] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the above-described user feature evaluation method based on cross-domain data fusion.

[0009] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described user feature evaluation method based on cross-domain data fusion.

[0010] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions, which, when executed by a processor, implement the above-described user feature evaluation method based on cross-domain data fusion.

[0011] In this embodiment, the cross-domain fusion computing system can utilize anonymized user identifiers as the unique credentials for cross-domain association and initiate collaborative computing requests to the second data domain via an identity matching protocol, rather than directly requesting the original detailed data of the second data domain. Since the identity matching protocol can avoid exposing the non-shared user identifiers and detailed data of either party during the interaction process, the cross-domain fusion computing system can securely obtain the feature tags of the second domain without acquiring the underlying detailed data of the second data domain. This allows the introduction of features from the communication behavior domain, such as spending power levels and the activity level of travel-related applications, which helps to fill the information blind spots of a single aviation data domain. It also ensures that no detailed data is exchanged during the entire acquisition process, thus protecting the security of cross-domain data interaction. Furthermore, the cross-domain fusion computing system merges the feature tags of the first and second data domains into a dimensionless comprehensive evaluation value, then compares this comprehensive evaluation value with a preset threshold, and outputs a discrete business execution instruction as a decision tag. Because the decision label does not contain the original information of the first or second domain feature labels, the business system can only obtain the final level or action instruction, but cannot deduce sensitive original data such as the user's specific flight activity, cabin class preference, and spending power level. This helps ensure the security of the cross-domain data fusion results and avoids the leakage of original feature information in the output stage. Thus, the cross-domain fusion computing system, by combining desensitized user identifiers with identity matching protocols, can securely obtain cross-domain feature labels without exchanging detailed data, improving information security. Furthermore, by introducing second-domain feature labels and fusing them with first-domain feature labels for evaluation, it helps reduce data blind spots. The final generated decision label, which does not contain original information, can both meet the evaluation needs of the business system and reduce the risk of leakage of original feature information. This solves the technical problem of existing passenger evaluation methods, which cannot integrate aviation behavior and communication behavior features across industries without exchanging detailed data, resulting in data blind spots and poor information security in the evaluation results. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0013] Figure 1 This is a schematic diagram of an optional user feature evaluation method based on cross-domain data fusion according to an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of the architecture of an optional cross-domain fusion computing system according to an embodiment of this application;

[0015] Figure 3 This is a flowchart of an optional user feature evaluation method based on cross-domain data fusion according to an embodiment of this application;

[0016] Figure 4 This is a schematic diagram of an optional user feature evaluation device based on cross-domain data fusion according to an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] According to an embodiment of this application, a method embodiment for user feature evaluation based on cross-domain data fusion is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0020] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0021] According to the embodiments of this application, a cross-domain fusion computing system (hereinafter referred to as "the system") can be used as the execution subject of the user feature evaluation method based on cross-domain data fusion in the embodiments of this application. The system can be a software system or an embedded system combining software and hardware. Of course, the execution subject of the method in the embodiments of this application can also be other forms of execution subject, such as devices, equipment, etc. It should be known by those skilled in the art that this application does not particularly limit the specific form of the execution subject.

[0022] Figure 1 This is a schematic diagram of a user feature evaluation method based on cross-domain data fusion according to an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps:

[0023] Step S101: Receive an evaluation request regarding user characteristics from the business system.

[0024] For example, receiving an evaluation request regarding user characteristics from a business system can refer to the cross-domain fusion computing system receiving a call instruction initiated by the business system to trigger cross-domain feature fusion computing for a target user. The evaluation request may include a de-identified unique user identifier, used to achieve secure matching of the same user across different data domains without exposing the original user information. The cross-domain fusion computing system receives evaluation requests through a unified interface, aggregating and standardizing dispersed business-side evaluation requirements. This decouples subsequent feature extraction and fusion computing processes from the business system, improving the system's ability to handle multi-source business requests and reducing the computational complexity of the business system.

[0025] In some embodiments, the cross-domain fusion computing system can receive evaluation requests sent in batches by the business system through a preset application programming interface. The batch evaluation requests can carry multiple anonymized user identifiers in list form, with each anonymized user identifier corresponding to an independent user feature evaluation task. After receiving the batch evaluation requests, the cross-domain fusion computing system assigns task batch identifiers to the batch evaluation requests for association with subsequent batch feature extraction and fusion computing processes. This embodiment, through a batch receiving mechanism, helps improve the system's processing efficiency for large-scale user evaluation tasks.

[0026] In other embodiments, the cross-domain fusion computing system can receive evaluation requests pushed in real-time by the business system through a message queue middleware. The real-time evaluation request is generated by the business system when it detects a user triggering a preset interactive behavior. The evaluation request includes a de-identified user identifier and a scenario type parameter to identify the current business scenario. The cross-domain fusion computing system, acting as a message consumer, pulls the evaluation request from the message queue and parses the de-identified user identifier and scenario type parameter. This embodiment, through an asynchronous message receiving mechanism, facilitates immediate response to real-time user behavior and smoothly handles instantaneous high-concurrency evaluation request traffic.

[0027] Step S102: Based on the de-identified user identifier in the evaluation request, extract the first domain feature label corresponding to the user in the first data domain. The first data domain is the air travel data domain, and the first domain feature label includes at least one of the user's flight activity, cabin class preference, and route preference.

[0028] For example, based on the anonymized user identifier in the evaluation request, the first domain feature label corresponding to the user is extracted within the first data domain. This can refer to the cross-domain fusion computing system querying and obtaining air travel behavior feature data associated with the anonymized user identifier within the air travel data domain. Here, the first data domain is the air travel data domain, which can represent a pre-constructed data set space for storing passenger air travel behavior feature labels. The first domain feature labels include at least one of the user's flight activity, cabin class preference, and route preference. Flight activity can represent a quantitative indicator of the user's flight frequency or flight mileage within a certain time period; cabin class preference can represent the distribution tendency of the cabin class selected by the user during historical bookings, such as first class, business class, and economy class; route preference can represent the pattern of routes between the user's frequently used departure and destination points. The cross-domain fusion computing system performs localized feature extraction within the first data domain based on the anonymized user identifier, which can reduce the need to expose the internal raw data of the air travel data domain to external domains. This is beneficial for providing structured local feature inputs for subsequent cross-domain fusion computing while protecting the security of air domain data assets.

[0029] In some embodiments, the cross-domain fusion computing system can use the received anonymized user identifier as a query key to perform matching retrieval in a locally stored aviation domain feature tag library. The aviation domain feature tag library pre-stores mapping relationships between anonymized user identifiers and at least one corresponding first domain feature tag. When a match is successful, the cross-domain fusion computing system reads at least one of the flight activity value, cabin class preference code, and route preference code corresponding to the anonymized user identifier, and uses the read data as the extracted first domain feature tag. This embodiment achieves feature extraction through key-value matching, which helps improve the query efficiency of the domain feature tag.

[0030] In other embodiments, the cross-domain fusion computing system sends the anonymized user identifier to the feature service interface of the first data domain. The feature service interface then calculates and returns the corresponding first domain feature tag in real time based on the anonymized user identifier. The real-time calculation process includes: retrieving historical behavior records associated with the anonymized user identifier from the raw log database of the first data domain; performing statistical analysis and feature encoding on the historical behavior records; and generating at least one tag among flight activity, cabin class preference, and route preference. This embodiment generates feature tags through real-time calculation, which helps ensure that the extracted first domain feature tags reflect the user's latest air travel behavior status.

[0031] Step S103: Based on the de-identified user identifier, initiate a collaborative computing request to the second data domain through an identity matching protocol to obtain the user's second domain feature label within the second data domain. The second data domain is a communication behavior data domain, and the second domain feature label includes at least one of the user's spending power level and the activity level of using travel-related applications.

[0032] For example, based on anonymized user identifiers, a collaborative computing request is initiated to the second data domain via an identity matching protocol to obtain the user's second-domain feature tags within the second data domain. This can refer to a cross-domain fusion computing system using anonymized user identifiers as the matching basis, employing an identity matching protocol that does not expose non-shared user identifiers to initiate joint computing instructions to the communication behavior data domain, and receiving the user's second-domain feature tags returned by the communication behavior data domain. Here, the identity matching protocol can represent a cryptographic security protocol that confirms the shared identity of users across multiple data domains without disclosing non-shared user information within each domain. That is, the identity matching protocol can refer to a privacy protection protocol executed in a trusted data space. A trusted data space can be an environment that enables cross-domain data usage in a way that is invisible, controllable, and measurable, while ensuring the data provider's data ownership and privacy security, such as a trusted computing environment. The second data domain is the communication behavior data domain, which can represent a data set space constructed by a communication operator and formed after anonymization and irreversible processing to store user communication behavior feature tags. The second-domain feature tags include at least one of the user's spending power level and activity level in travel-related applications. Consumption capacity level can represent a tiered identifier based on users' communication consumption amount; the activity level of travel-related applications can represent a quantitative indicator of the frequency or duration of users' use of navigation, ticketing, and hotel applications. The cross-domain fusion computing system initiates collaborative computing requests through an identity matching protocol. This allows the second data domain to provide second-domain feature tags only to successfully matched shared users, without returning non-shared user information or exposing original detailed data. This approach helps to ensure data privacy and security in the communication behavior data domain while obtaining cross-domain feature information.

[0033] In some embodiments, the cross-domain fusion computing system can use anonymized user identifiers as input and invoke a privacy intersection protocol to perform privacy set intersection calculations with the second data domain. The result of the privacy set intersection calculation only indicates whether the anonymized user identifier exists in the user identifier set of the second data domain. When the anonymized user identifier exists in the user identifier set of the second data domain, the cross-domain fusion computing system sends a feature acquisition request to the second data domain; the second data domain queries the consumption capacity level and travel-related application usage activity associated with the anonymized user identifier based on the feature acquisition request, and returns the obtained second-domain feature tags to the cross-domain fusion computing system. This embodiment achieves shared user confirmation through a privacy intersection protocol, which is beneficial for protecting the privacy of user identifiers in both data domains during the matching stage.

[0034] In other embodiments, the cross-domain fusion computing system encapsulates the anonymized user identifier within a collaborative computing request and sends the request to a second data domain feature service interface within a trusted computing environment. The trusted computing environment ensures that data is not spied on by external entities during computation. Within the trusted computing environment, the second data domain feature service interface queries the corresponding second domain feature tag based on the anonymized user identifier and returns the retrieved tag to the cross-domain fusion computing system via the trusted computing environment. Throughout the query process, the second data domain does not expose its internal raw user behavior details to the cross-domain fusion computing system. This embodiment, by performing feature queries through a trusted computing environment, facilitates the acquisition of second domain feature tags and enhances the security of the data computation process.

[0035] Step S104: Determine the comprehensive evaluation value based on the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label.

[0036] For example, determining the comprehensive evaluation value based on the first domain feature label and the second domain feature label can refer to the cross-domain fusion computing system performing numerical conversion and weighted combination calculation on the first domain feature label from the aviation travel data domain and the second domain feature label from the communication behavior data domain to obtain a dimensionless value representing the overall value level of the user. The comprehensive evaluation value can represent a single numerical result obtained after weighted fusion of the first and second domain feature labels. This numerical result has no physical dimension and is only used for value comparison between different users or for comparison with a preset threshold. Weighted fusion can refer to the calculation process of assigning different weight coefficients to the first contribution value corresponding to the first domain feature label and the second contribution value corresponding to the second domain feature label, and then summing them. By fusing feature labels from two different data domains into a dimensionless comprehensive evaluation value, the cross-domain fusion computing system maps heterogeneous features of different dimensions and types to the same evaluation scale. This helps reduce the impact of dimensional differences between cross-domain features on the evaluation results and provides a unified and comparable numerical basis for subsequent threshold comparison and decision label generation.

[0037] In some embodiments, the cross-domain fusion computing system can numerically map the first domain feature label and the second domain feature label respectively to obtain a first numerical feature and a second numerical feature. The cross-domain fusion computing system obtains preset first weight coefficients and second weight coefficients, where the first weight coefficient corresponds to the first numerical feature and the second weight coefficient corresponds to the second numerical feature. The cross-domain fusion computing system calculates a first product of the first numerical feature and the first weight coefficient, calculates a second product of the second numerical feature and the second weight coefficient, and uses the sum of the first and second products as the comprehensive evaluation value. This embodiment fuses cross-domain features through a linear weighting method, which simplifies the calculation process and improves evaluation efficiency.

[0038] In other embodiments, the cross-domain fusion computing system can input the first domain feature label into a first nonlinear normalization function to obtain a first normalized value with a uniform value range; and input the second domain feature label into a second nonlinear normalization function to obtain a second normalized value with a uniform value range. The cross-domain fusion computing system obtains a dynamic weight allocation strategy, and determines the first dynamic weight corresponding to the first normalized value and the second dynamic weight corresponding to the second normalized value according to the dynamic weight allocation strategy. The cross-domain fusion computing system calculates a first weighted value of the first normalized value and the first dynamic weight, calculates a second weighted value of the second normalized value and the second dynamic weight, and uses the sum of the first weighted value and the second weighted value as the comprehensive evaluation value. This embodiment, through nonlinear normalization processing, helps to reduce the distribution differences of feature labels from different domains, and combined with the dynamic weight allocation strategy, helps to improve the adaptability of the comprehensive evaluation value under different business scenarios.

[0039] Step S105: Generate a decision label based on the comparison result between the comprehensive evaluation value and the preset threshold. The decision label is a discrete business execution instruction and does not contain the original information of the first domain feature label or the second domain feature label.

[0040] For example, generating a decision label based on the comparison result of a comprehensive evaluation value and a preset threshold can refer to the cross-domain fusion computing system comparing the calculated comprehensive evaluation value with one or more preset thresholds, and outputting a discrete business execution instruction as the decision label based on the comparison result. Here, the decision label is a discrete business execution instruction output after dimensionality reduction mapping, indicating that the value of the decision label comes from a finite, predefined set of business instructions, rather than a continuous numerical range or original feature data. The decision label does not contain the original information of the first domain feature label or the second domain feature label, indicating that there is no reversible mapping relationship between the decision label and the first and second domain feature labels, and the specific numerical or attribute content of the first or second domain feature labels cannot be deduced from the decision label. By mapping continuous comprehensive evaluation values ​​to discrete decision labels and outputting them in the form of business execution instructions, the cross-domain fusion computing system allows the business-side system to only obtain the final decision result without knowing the original cross-domain feature information used to generate the decision. This helps to meet the business-side's need for accurate reach while preventing the risk of reverse reconstruction of the underlying cross-domain feature data, thus contributing to privacy protection in cross-domain data fusion computing.

[0041] In some embodiments, the cross-domain fusion computing system can obtain a preset evaluation value threshold. The cross-domain fusion computing system determines whether the comprehensive evaluation value is greater than or equal to the preset evaluation value threshold. If the comprehensive evaluation value is greater than or equal to the preset evaluation value threshold, the cross-domain fusion computing system outputs a first decision label, which corresponds to the recommended service execution instruction. If the comprehensive evaluation value is less than the preset evaluation value threshold, the cross-domain fusion computing system outputs a second decision label, which corresponds to the service execution instruction that is not to be executed temporarily. This embodiment divides users into two levels and outputs corresponding discrete instructions through a single threshold comparison, which helps to simplify the decision logic and improve the efficiency of decision label generation.

[0042] In other embodiments, the cross-domain fusion computing system acquires multiple preset evaluation value thresholds. These thresholds divide the range of the comprehensive evaluation value into multiple consecutive numerical sub-intervals, each corresponding to a preset decision label. The cross-domain fusion computing system detects the numerical sub-interval into which the comprehensive evaluation value falls and determines the corresponding decision label based on the detected sub-interval. The decision label is a structured business execution instruction containing a policy level identifier and an action type identifier. This embodiment, by using multiple thresholds to segment users into multiple levels and output corresponding structured instructions, facilitates providing more refined hierarchical decision-making basis for the business-side system.

[0043] Figure 2 This diagram illustrates an optional cross-domain fusion computing system architecture according to an embodiment of this application. The user feature evaluation method based on cross-domain data fusion provided in this embodiment can be applied to a cross-domain fusion computing system. This cross-domain fusion computing system includes a first data domain, a second data domain, a trusted data space, and a business system. The first data domain can be an air travel data domain. A de-identification processing unit and a first domain feature tag library are deployed within the first data domain. The de-identification processing unit performs irreversible encryption on the original user identifier to generate a de-identified user identifier. The first domain feature tag library stores first domain feature tags, which include at least one of the user's flight activity level, cabin class preference, and route preference. The second data domain can be a communication behavior data domain. A de-identification processing unit and a second domain feature tag library are deployed within the second data domain. The de-identification processing unit performs irreversible encryption on the original user identifier to generate a de-identified user identifier. The second domain feature tag library stores second domain feature tags, which include at least one of the user's spending power level and activity level in using travel-related applications.

[0044] The trusted data space communicates with the first data domain, the second data domain, and the business system. Within the trusted data space, an identity matching module, a fusion computing engine, a policy dimensionality reduction output module, and a billing and compliance audit module are deployed. The billing and compliance audit module, indicated by a dashed line, is an optional configuration module. If the billing and compliance audit module is not configured, the policy dimensionality reduction output module can directly output decision tags to the business system. The identity matching module is used to confirm the de-identified user identifiers shared by the first and second data domains through an identity matching protocol. Without exposing unmatched user identifiers, it obtains second-domain feature tags from the second data domain for the shared de-identified user identifiers. The fusion computing engine is used to determine a comprehensive evaluation value based on the first-domain feature tags provided by the first data domain and the second-domain feature tags provided by the second data domain. The comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first-domain and second-domain feature tags. The policy dimensionality reduction output module is used to generate decision tags based on the comparison result of the comprehensive evaluation value and a preset threshold. The decision tags are discrete business execution instructions and do not contain the original information of the first-domain or second-domain feature tags. The billing and compliance audit module records the number of times business systems call the Trusted Data Space service and the tracking identifier for each call, enabling metering and compliance auditing of cross-domain data fusion services. This module supports the controllable and measurable data flow characteristics of the Trusted Data Space, allowing cross-domain fusion computing systems to provide traceable service call records for business systems while ensuring data security.

[0045] The business system, as the caller, can send an evaluation request to the trusted data space. The evaluation request carries an anonymized user identifier. In response to the evaluation request, the trusted data space performs identity matching, feature acquisition, fusion calculation, and decision label generation operations, and returns the generated decision labels to the business system. The business system then executes the corresponding business instructions based on the decision labels.

[0046] In some optional embodiments, determining the comprehensive evaluation value based on the first domain feature label and the second domain feature label includes: the cross-domain fusion computing system can convert the first domain feature label into a first numerical feature and the second domain feature label into a second numerical feature according to a preset numerical mapping rule; input the first numerical feature into a first normalization function to obtain a first normalized value; input the second numerical feature into a second normalization function to obtain a second normalized value; wherein the first normalization function and the second normalization function are both nonlinear normalization functions; the first normalization function and the second normalization function are the same function, or are different functions adapted to the feature distributions of different data domains respectively; obtain a first weight corresponding to the first normalized value and a second weight corresponding to the second normalized value; and perform a weighted summation calculation on the first normalized value and the second normalized value according to the first weight and the second weight to obtain the comprehensive evaluation value.

[0047] For example, according to a preset numerical mapping rule, the first domain feature label is converted into a first numerical feature, and the second domain feature label is converted into a second numerical feature. This can refer to the cross-domain fusion computing system converting discrete or non-numerical first domain feature labels and second domain feature labels into first and second numerical features that can participate in numerical operations, respectively, based on a predefined mapping relationship. Both the first and second normalization functions are nonlinear normalization functions. A nonlinear normalization function can represent a function that maps input numerical features to a predetermined numerical interval (such as the interval between 0 and 1) through a nonlinear transformation (such as logarithmic transformation, sigmoid transformation, or tangent transformation). This nonlinear normalization function can unify the scale of numerical features with different dimensions or distribution patterns. The first and second normalization functions can be the same, or different nonlinear functions can be set according to the actual distribution of features in different data domains. The cross-domain fusion computing system transforms the first and second numerical features through a nonlinear normalization function, which helps to reduce the impact of differences in feature dimensions and distributions between the aviation travel data domain and the communication behavior data domain on the fusion results. This allows features from different data domains to be weighted and combined under the same evaluation scale, thereby improving the scientific nature and cross-domain comparability of the comprehensive evaluation value.

[0048] For example, a cross-domain converged computing system can use the following formula to calculate the comprehensive evaluation value:

[0049] ;

[0050] in, Indicates the first domain feature label, This represents the first normalization function. Indicates the first weight; Indicates the feature label of the second domain. This represents the second normalization function. Indicates the second weight. This represents the comprehensive evaluation value. Features from two data domains are combined into a dimensionless comprehensive evaluation value through a weighted summation method.

[0051] In some embodiments, the cross-domain fusion computing system can obtain a preset category and numerical mapping table. Based on this table, it maps cabin class preference tags (e.g., first-class, business-class, economy-class) in the first domain feature labels to values ​​of 0.9, 0.6, and 0.3, respectively, as part of the first numerical feature. Similarly, it maps consumption capacity level tags (e.g., high-level, medium-level, low-level) in the second domain feature labels to values ​​of 0.8, 0.5, and 0.2, respectively, as part of the second numerical feature. The system can input the first numerical feature into a first normalization function based on a sigmoid function to obtain a first normalized value. It can also input the second numerical feature into a second normalization function, also based on a sigmoid function, to obtain a second normalized value. The cross-domain fusion computing system obtains preset first and second weights. For example, the first weight can be preset to 0.4 and the second weight to 0.6. It calculates the first product of the first weight and the first normalized value, and the second product of the second weight and the second normalized value. The sum of the first and second products is used as the comprehensive evaluation value. This embodiment uses the same nonlinear normalization function to process cross-domain features, which helps simplify the configuration of normalization parameters and ensures the uniformity of transformation rules.

[0052] In other embodiments, the cross-domain fusion computing system directly uses the flight frequency value corresponding to the flight activity tag in the first domain feature label as the first numerical feature, and directly uses the weekly usage duration value corresponding to the travel-related application usage activity tag in the second domain feature label as the second numerical feature. The cross-domain fusion computing system inputs the first numerical feature into a first normalization function based on logarithmic transformation to obtain a first normalized value with compressed value range; it inputs the second numerical feature into a second normalization function based on tangent transformation to obtain a second normalized value with saturated value range, wherein the first normalization function and the second normalization function are different. The cross-domain fusion computing system obtains a first weight corresponding to the first normalized value and a second weight corresponding to the second normalized value. The first weight is dynamically determined based on the magnitude of the first normalized value; the larger the first normalized value, the smaller the first weight. The second weight is also dynamically determined based on the magnitude of the second normalized value; the larger the second normalized value, the larger the second weight. The first product of the first weight and the first normalized value, and the second product of the second weight and the second normalized value are calculated. The sum of the first and second products is used as the comprehensive evaluation value. This embodiment selects different nonlinear normalization functions and dynamic weights for different distributions of features in different data domains, which helps to more precisely adapt to the data characteristics of each domain and improve the accuracy of the comprehensive evaluation value.

[0053] In another embodiment, the first normalization function and the second normalization function can be replaced by a first local machine learning model deployed in the first data domain and a second local machine learning model deployed in the second data domain, respectively. The cross-domain fusion computing system controls the first and second local machine learning models to interact only with intermediate parameters within a trusted data space without exchanging underlying first-domain and second-domain feature labels. It dynamically allocates first and second weights according to the current business scenario and obtains a comprehensive evaluation value using a weighted summation formula. This embodiment extends the weighted fusion mechanism of this application to federated learning architectures, which is beneficial for achieving cross-domain joint inference with a higher level of privacy protection.

[0054] In some optional embodiments, obtaining a first weight corresponding to a first normalized value and a second weight corresponding to a second normalized value includes: the cross-domain fusion computing system can detect the business scenario identifier of the business system, wherein the business scenario identifier includes at least a first type of scenario identifier, a second type of scenario identifier, and a third type of scenario identifier. The first type of scenario identifier corresponds to a wake-up scenario where the user's flight activity is lower than a preset activity threshold and the communication behavior activity is higher than a preset communication activity threshold; the second type of scenario identifier corresponds to a scenario where the user's historical cabin class consumption level is higher than a preset cabin class level; the third type of scenario identifier corresponds to a scenario where potential passengers are screened for newly opened routes; if the business scenario identifier of the business system is detected to be a first type of scenario identifier, the second weight is determined to be greater than the first weight; if the business scenario identifier of the business system is detected to be a second type of scenario identifier, the first weight is determined to be greater than the second weight; if the business scenario identifier of the business system is detected to be a third type of scenario identifier, the first weight is determined to be equal to the second weight.

[0055] For example, detecting the business scenario identifier of the business system can refer to the cross-domain fusion computing system parsing an identifier information from the received evaluation request to identify the current business scenario type. The business scenario identifier includes at least a first type of scenario identifier, a second type of scenario identifier, and a third type of scenario identifier. The first type of scenario identifier corresponds to a wake-up scenario where the user's flight activity is below a preset activity threshold but their communication behavior activity is above a preset communication activity threshold. This first type of scenario indicates that the user has recently been in a low-activity state in the air travel domain but is in a high-activity state in the communication behavior domain, requiring wake-up. The second type of scenario identifier corresponds to a scenario where the user's historical cabin class consumption level is higher than a preset cabin class level. This second type of scenario indicates that the user has historically had high cabin class consumption capacity. The third type of scenario identifier corresponds to a scenario for screening potential passengers for newly opened routes. This third type of scenario indicates that it is necessary to balance the user's potential demand for new routes from cross-domain features. The cross-domain fusion computing system dynamically adjusts the relationship between the first and second weights based on the detected different business scenario identifiers, making the calculation of the comprehensive evaluation value biased towards data domain features that contribute more to the current scenario. This is beneficial for improving the adaptability and accuracy of the fusion evaluation results under different business objectives.

[0056] In some embodiments, the cross-domain fusion computing system directly reads a pre-set scenario type field from the evaluation request sent by the business system as a business scenario identifier. When the business scenario identifier is detected as a first-type scenario identifier, the cross-domain fusion computing system determines that the second weight is greater than the first weight to enhance the contribution ratio of communication behavior domain features in the comprehensive evaluation value. When the business scenario identifier is detected as a second-type scenario identifier, the cross-domain fusion computing system determines that the first weight is greater than the second weight to enhance the contribution ratio of aviation travel domain features in the comprehensive evaluation value. When the business scenario identifier is detected as a third-type scenario identifier, the cross-domain fusion computing system determines that the first weight is equal to the second weight, so that the features of the first data domain and the second data domain have the same degree of contribution in the comprehensive evaluation value. This embodiment, by directly parsing the scenario identifier from the evaluation request and executing the corresponding weight comparison strategy, helps to simplify the scenario identification process and quickly adapt to different business objectives.

[0057] In other embodiments, the cross-domain fusion computing system automatically determines the scenario identifier type corresponding to the current business scenario based on the flight activity value in the first domain feature label and the communication behavior activity value in the second domain feature label. For example, when the flight activity value is lower than a preset activity threshold and the communication behavior activity value is higher than a preset communication activity threshold, the cross-domain fusion computing system identifies the current scenario identifier as a first-type scenario identifier; when the historical cabin class consumption level value in the first domain feature label is higher than a preset cabin class level threshold, the cross-domain fusion computing system identifies the current scenario identifier as a second-type scenario identifier; when there is a preset similarity matching relationship between the route preference in the first domain feature label and the new route to be evaluated, the cross-domain fusion computing system identifies the current scenario identifier as a third-type scenario identifier. The cross-domain fusion computing system sets the magnitude relationship between the first weight and the second weight according to the automatically determined scenario identifier type. This embodiment, by automatically identifying scenario types and dynamically configuring weight relationships through the cross-domain fusion computing system, helps to reduce the dependence on pre-marked scenario identifiers by the business system and improves the system's intelligence level.

[0058] In some optional embodiments, a decision label is generated based on the comparison result of the comprehensive evaluation value and a preset threshold, including: the cross-domain fusion computing system can obtain a preset first evaluation value threshold and a second evaluation value threshold, wherein the first evaluation value threshold is greater than the second evaluation value threshold; detect whether the comprehensive evaluation value is greater than or equal to the first evaluation value threshold; if the comprehensive evaluation value is detected to be greater than or equal to the first evaluation value threshold, output a first decision label; if the comprehensive evaluation value is detected to be less than the first evaluation value threshold, detect whether the comprehensive evaluation value is greater than or equal to the second evaluation value threshold; if the comprehensive evaluation value is detected to be greater than or equal to the second evaluation value threshold, output a second decision label; if the comprehensive evaluation value is detected to be less than the second evaluation value threshold, output a third decision label; the recommendation level of the business execution instruction corresponding to the first decision label is higher than the recommendation level of the business execution instruction corresponding to the second decision label, and the recommendation level of the business execution instruction corresponding to the second decision label is higher than the recommendation level of the business execution instruction corresponding to the third decision label; the business execution instruction contains a strategy level and recommended action in the form of a structured instruction.

[0059] For example, obtaining the preset first and second evaluation value thresholds can refer to the cross-domain fusion computing system reading two comparison boundary values ​​used to divide the comprehensive evaluation value levels from local configuration or business request parameters, wherein the first evaluation value threshold is greater than the second evaluation value threshold. The first, second, and third decision labels are three discrete business execution instructions. The recommendation level of the business execution instruction corresponding to the first decision label is higher than that of the business execution instruction corresponding to the second decision label, and vice versa. The business execution instructions contain policy levels and recommended actions in the form of structured instructions. The structured instructions can represent a data format containing predefined fields; for example, the predefined fields may include a policy level field and a recommended action field. By comparing the comprehensive evaluation value with the first and second evaluation value thresholds in a hierarchical manner, the cross-domain fusion computing system discretizes the continuous comprehensive evaluation value into three decision labels with different recommendation levels and outputs them in the form of structured instructions containing policy levels and recommended actions. This facilitates providing hierarchical and directly executable business guidance to the business-side system and avoids exposing the comprehensive evaluation value itself and any original domain feature information.

[0060] In some embodiments, the cross-domain fusion computing system acquires a pre-stored first evaluation value threshold and a second evaluation value threshold. For example, the first evaluation value threshold can be set to 85, and the second evaluation value threshold can be set to 60. The cross-domain fusion computing system first detects whether the overall evaluation value is greater than or equal to the first evaluation value threshold. If the overall evaluation value is detected to be greater than or equal to the first evaluation value threshold, the cross-domain fusion computing system outputs a first decision label. In the structured instruction corresponding to the first decision label, the strategy level field is set to high level, and the recommended action field is set to recommend high-value benefits. If the overall evaluation value is detected to be less than the first evaluation value threshold, the cross-domain fusion computing system continues to detect whether the overall evaluation value is greater than or equal to the second evaluation value threshold. If the overall evaluation value is detected to be greater than or equal to the second evaluation value threshold and less than the first evaluation value threshold, the cross-domain fusion computing system outputs a second decision label. In the structured instruction corresponding to the second decision label, the strategy level field is set to medium level, and the recommended action field is set to recommend regular benefits. If the overall evaluation value is detected to be less than the second evaluation value threshold, the cross-domain fusion computing system outputs a third decision label. In the structured instruction corresponding to the third decision label, the strategy level field is set to low level, and the recommended action field is set to not recommend. This embodiment divides users into three recommendation levels using two thresholds, which helps to provide differentiated hierarchical decision-making basis for business systems.

[0061] In other embodiments, the cross-domain fusion computing system parses a first evaluation threshold and a second evaluation threshold specified by the business system from the received evaluation request. These two thresholds dynamically change with different evaluation requests. The cross-domain fusion computing system sequentially performs comparison steps with the first and second evaluation thresholds according to their numerical relationship. When the comprehensive evaluation value meets different threshold ranges, the cross-domain fusion computing system outputs a corresponding decision label. The structured instructions of the decision label also include a batch identifier for identifying the current evaluation task. This embodiment, by allowing the business system to dynamically specify evaluation thresholds, improves the flexibility of the decision label generation strategy, enabling the same evaluation method to adapt to the differentiated grading standards of different business systems.

[0062] By introducing the real-time status and consumption capacity characteristics of the second data domain, the limitations of historical data from a single air travel data domain can be reduced, facilitating a shift from inferring user value based on historical data to predicting user value based on current status. Furthermore, by generating decision labels that do not contain the original information of the first or second domain feature labels, and replacing the original multi-dimensional data output with discrete business execution instructions, it is beneficial to meet the precise reach requirements of the business side while preventing the risk of reverse engineering or overall leakage of underlying cross-domain feature data, thus meeting the data security requirement of usability without visibility in cross-domain data fusion. The embodiments of this application, through the dual-domain dynamic weight scoring mechanism in the fusion computing engine, facilitate the configuration of differentiated contribution levels for features of different data domains. The strategy dimensionality reduction output module facilitates the isolated output of original feature data by masking it with business execution instructions.

[0063] In some optional embodiments, after converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: the cross-domain fusion computing system can obtain the timestamp corresponding to the first numerical feature; determine a time decay coefficient based on the time difference between the current system time and the timestamp, wherein the larger the time difference, the smaller the time decay coefficient; multiply the first numerical feature by the time decay coefficient to obtain the updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0064] For example, obtaining the timestamp corresponding to the first numerical feature can refer to the cross-domain fusion computing system acquiring time information associated with the first domain feature label. This time information can record the occurrence time or statistical period of the user behavior corresponding to the first domain feature label. The time decay coefficient can represent a numerical factor that monotonically decreases as the time difference increases, used to reduce the contribution of outdated features to the comprehensive evaluation value. The cross-domain fusion computing system determines the time decay coefficient based on the time difference between the current system time and the timestamp, where the larger the time difference, the smaller the time decay coefficient. Multiplying the first numerical feature by the time decay coefficient yields the updated first numerical feature, which is used to replace the original first numerical feature to redetermine the comprehensive evaluation value. By introducing a time decay mechanism, the cross-domain fusion computing system weakens the weight of older historical behavior data in the comprehensive evaluation value, while maintaining a higher contribution of more recent behavior data. This helps improve the comprehensive evaluation value's ability to represent the user's current state and reduces the interference of outdated data on the evaluation results.

[0065] In some embodiments, the cross-domain fusion computing system reads the generation timestamp associated with the first domain feature label from the feature label library of the first data domain. This generation timestamp can record the date of the most recent update of the first domain feature label. The cross-domain fusion computing system obtains the current system date and calculates the difference in days between the current system date and the generation timestamp as the time difference. The cross-domain fusion computing system uses an exponential decay function to calculate the time decay coefficient with the time difference as the independent variable. The base of the exponential decay function is a positive number less than 1, so that the time decay coefficient decreases exponentially as the time difference increases. The cross-domain fusion computing system multiplies the first numerical feature by the time decay coefficient to obtain the updated first numerical feature. This embodiment uses an exponential decay method to handle time decay, which helps to make the decay rate of old features smooth and accelerated, which is more in line with the objective law that the timeliness of behavioral data gradually decreases.

[0066] In other embodiments, the cross-domain fusion computing system acquires multiple timestamps corresponding to the first numerical feature, each timestamp corresponding to a statistical time period of different sub-item labels in the first domain feature label. The cross-domain fusion computing system calculates the sub-item time difference between the current system time and each timestamp, and determines the corresponding sub-item time decay coefficient based on each sub-item time difference. The cross-domain fusion computing system multiplies each sub-item value in the first numerical feature by its corresponding sub-item time decay coefficient, and then recombines the updated values ​​of each sub-item into the updated first numerical feature. This embodiment, by applying time decay coefficients to different sub-items within the first domain feature label, facilitates more precise handling of the timeliness differences of different types of air travel behavior data, improving the accuracy of the first numerical feature.

[0067] For example, before conducting an activity, an airline's business system sends an assessment request to a cross-domain fusion computing system for a batch of dormant users who have had no flight records for more than 6 months. The anonymized user identifiers in these requests are phone numbers encrypted using an irreversible encryption algorithm. The cross-domain fusion computing system extracts first-domain feature tags from the anonymized user identifiers within the first data domain, discovering that these users have near-zero flight activity but high historical cabin class preferences. The cross-domain fusion computing system then initiates a collaborative computing request to the second data domain within a trusted data space using an identity matching protocol. Without returning specific amounts or details, the second data domain provides feedback that some users in this batch have high spending power levels and high activity levels in travel-related applications. The cross-domain fusion computing system determines a comprehensive assessment value based on the first and second domain feature tags. Due to the higher contribution value of the second domain feature tags, the comprehensive assessment value can reach a high score, for example, 85 points. The cross-domain fusion computing system generates decision tags based on the comparison between the comprehensive assessment value and a preset threshold. These decision tags are structured business execution instructions containing strategy levels and recommended actions.

[0068] In some optional embodiments, after converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: the cross-domain fusion computing system can obtain the user's itinerary information, wherein the itinerary information includes flight number, planned departure time, and planned arrival time; obtain dynamic status information of the flight number, wherein the dynamic status information includes at least on-time status, delay duration, and cancellation flag; determine the itinerary impact coefficient based on the dynamic status information, including: when the on-time status is on-time, determining the itinerary impact coefficient as a first coefficient; when the on-time status is delayed and the delay duration exceeds a preset duration threshold, determining the itinerary impact coefficient as a second coefficient, wherein the second coefficient is less than the first coefficient; when the cancellation flag indicates flight cancellation, determining the itinerary impact coefficient as a third coefficient, wherein the third coefficient is less than the second coefficient; multiplying the first numerical feature by the itinerary impact coefficient to obtain an updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0069] For example, obtaining a user's travel information can refer to the cross-domain fusion computing system reading travel records associated with the de-identified user identifier from the first data domain. The travel information includes the flight number, scheduled departure time, and scheduled arrival time. Dynamic status information can represent real-time status data generated during the operation of the flight corresponding to the flight number, including at least on-time status, delay duration, and cancellation flag. The travel impact coefficient can represent a numerical factor determined based on the flight's dynamic status, used to adjust the first numerical feature to reflect the impact of travel anomalies on the value of the user's air travel behavior. The cross-domain fusion computing system determines the travel impact coefficient based on the dynamic status information: when the on-time status is "on-time," the travel impact coefficient is determined as the first coefficient; when the on-time status is "delayed" and the delay duration exceeds a preset duration threshold, the travel impact coefficient is determined as the second coefficient, where the second coefficient is less than the first coefficient; when the cancellation flag indicates flight cancellation, the travel impact coefficient is determined as the third coefficient, where the third coefficient is less than the second coefficient. The cross-domain fusion computing system multiplies the first numerical feature by the travel impact coefficient to obtain an updated first numerical feature, which is used to replace the original first numerical feature to redetermine the comprehensive evaluation value. By introducing a trip impact coefficient to dynamically adjust the first numerical feature, abnormal states such as flight delays or cancellations can reduce the overall contribution of users' air travel characteristics, which helps the comprehensive evaluation value to more accurately reflect the negative impact of trip abnormalities on users' actual travel value and willingness.

[0070] In some embodiments, the cross-domain fusion computing system retrieves the dynamic status information of a flight number by calling a third-party flight data service interface, using the flight number as the query key. This dynamic status information includes on-time status, delay duration, and cancellation flag. The cross-domain fusion computing system first detects the cancellation flag: if the cancellation flag indicates flight cancellation, the system determines the trip impact coefficient as the third coefficient. If the cancellation flag indicates the flight is not cancelled, the system further detects the on-time status: if the on-time status is on-time, the system determines the trip impact coefficient as the first coefficient; if the on-time status is delayed, the system obtains the delay duration and compares it with a preset duration threshold. If the delay duration exceeds the preset duration threshold, the system determines the trip impact coefficient as the second coefficient; if the delay duration does not exceed the preset duration threshold, the system determines the trip impact coefficient as the first coefficient. This embodiment determines the trip impact coefficient through a hierarchical judgment mechanism, which is beneficial for applying differentiated numerical adjustments based on the degree of flight anomaly, allowing the updated first numerical feature to accurately reflect the severity of trip interference.

[0071] In other embodiments, the cross-domain fusion computing system obtains multiple trip information entries from the first data domain, with each trip information entry corresponding to a trip record. The system then obtains the dynamic status information of the flight number in each trip record and determines the sub-trip impact coefficient for each trip record based on this information. The system performs a weighted average or minimum value processing on the multiple sub-trip impact coefficients to obtain the final trip impact coefficient. Finally, it multiplies the first numerical feature by the final trip impact coefficient to obtain the updated first numerical feature. This embodiment determines the trip impact coefficient by comprehensively considering the dynamic status of multiple user trips, which helps avoid the excessive influence of a single trip anomaly on the evaluation results, making the updated first numerical feature more reflective of the stability of the user's overall trip status.

[0072] In some optional embodiments, after converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: the cross-domain fusion computing system can obtain the user's historical flight records, wherein the historical flight records include the takeoff time, landing time, departure airport, and arrival airport for each flight; converting the flight activity in the first domain feature label into a flight activity value; determining a flight activity attenuation coefficient based on the time difference between the current time and the most recent flight in the historical flight records, wherein the flight activity attenuation coefficient decreases as the time difference increases; and multiplying the flight activity value by the attenuation coefficient to obtain the time-attenuated value. Flight activity value; obtain flight plan information within a future preset time window, and predict the user's travel probability within the future preset time window based on the user's route preferences and flight plan information. The flight plan information includes flight time, route, and remaining cabin class. The travel probability is calculated based on the similarity between the user's past flight routes and future flight routes, as well as the cabin class matching degree. Using preset activity weights and travel probability weights, the time-decayed flight activity value and travel probability are weighted and summed to obtain the updated first numerical feature. The updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0073] For example, obtaining a user's historical flight records can refer to the cross-domain fusion computing system reading historical flight records associated with the de-identified user identifier from the first data domain. These historical flight records include the takeoff time, landing time, departure airport, and arrival airport for each flight. The flight activity decay coefficient can represent a decreasing factor determined based on the time difference between the current time and the most recent flight, used to reduce the contribution of outdated flight records to the flight activity value. The travel probability can represent the probability of a user traveling within a preset future time window, predicted based on the matching degree between the user's route preferences and future flight plans. The cross-domain fusion computing system converts the flight activity in the first domain feature label into a flight activity value, determines the flight activity decay coefficient based on the time difference between the current time and the most recent flight in the historical flight records, and multiplies the flight activity value by the decay coefficient to obtain the time-decayed flight activity value. The cross-domain fusion computing system obtains flight plan information within the preset future time window, predicts the travel probability based on the user's route preferences and flight plan information, and uses preset activity weights and travel probability weights to weighted sum the time-decayed flight activity value and the travel probability to obtain the updated first numerical feature. By updating the first numerical feature by decaying historical flight activity over time and combining it with predictions of future travel probabilities, the comprehensive evaluation value can reflect both the timeliness of users' past flight behavior and proactively capture users' potential travel needs, thereby improving the dynamism and predictive ability of the evaluation results.

[0074] In some embodiments, the cross-domain fusion computing system extracts all historical flight records of a user over the past two years from a first data domain. The system converts the flight activity level in the first domain's feature tags into a numerical value, such as the number of flights per year or the cumulative flight mileage per year. The system calculates the difference in days between the current date and the date of the most recent flight, and uses a linear decay function to determine the flight activity decay coefficient, where the decay coefficient equals 1 minus the ratio of the difference in days to the maximum decay period, and the minimum value of the decay coefficient is 0. The system multiplies the flight activity value by the decay coefficient to obtain the time-decreased flight activity value. This embodiment handles the timeliness of flight activity through a linear decay method, which helps to reduce the activity estimation of users who have not flown for a long time in a simple and intuitive way.

[0075] In other embodiments, the cross-domain fusion computing system obtains flight schedule information for the next thirty days, including flight times, routes, and remaining cabin class. The system extracts the user's route preferences, which include the most frequent departure and arrival airport pairs from the user's historical flight history. The system calculates route similarity based on the matching degree between route preferences and the departure and arrival airports of each route in the future flight schedule, and calculates cabin class matching degree based on the user's historical cabin class preferences and the remaining cabin class of each route in the future flight schedule. The weighted sum of route similarity and cabin class matching degree is used as the travel probability. The system uses preset activity weights and travel probability weights to weight and sum the time-decayed flight activity value with the travel probability to obtain the updated first numerical feature. This embodiment, by combining route preferences and cabin class matching degree to predict travel probability, helps to make the updated first numerical feature more accurately reflect the user's true travel intentions in the near future.

[0076] In some optional embodiments, after calculating the weighted sum of the first normalized value and the second normalized value based on the first weight and the second weight to obtain the comprehensive evaluation value, the method further includes: the cross-domain fusion computing system can record the feedback results after the business system executes the business instruction according to the decision tag, wherein the feedback results include at least one of whether the user responded to the business instruction and the actual consumption behavior after responding to the business instruction; based on the feedback results, calculate the actual conversion rate in the current business scenario, wherein the actual conversion rate is the ratio of the number of users responding to the business instruction to the total number of users receiving the decision tag; compare the actual conversion rate with the preset target conversion rate to obtain a conversion rate deviation value; based on the conversion rate deviation value, use a PID algorithm or a gradient descent algorithm to calculate the adjustment amount of the first weight and the second weight respectively; increase the first weight and the second weight by the corresponding adjustment amount respectively to obtain the updated first weight and the updated second weight.

[0077] For example, recording the feedback results after a business system executes a business instruction based on a decision tag can refer to the cross-domain fusion computing system receiving and storing user behavior response data generated after the business system executes the business instruction corresponding to the decision tag. The feedback results include at least one of whether the user responded to the business instruction and the actual consumption behavior after responding to the business instruction. The actual conversion rate can be expressed as the ratio of the number of users responding to the business instruction to the total number of users who received the decision tag; this ratio can be a value between 0 and 1. The conversion rate deviation can be expressed as the difference between the actual conversion rate and the preset target conversion rate; this difference can be positive or negative. A PID algorithm (Proportional-Integral-Derivative Control Algorithm) or gradient descent algorithm can be used to iteratively calculate the weight adjustment amount based on the conversion rate deviation. The cross-domain fusion computing system calculates the actual conversion rate in the current business scenario based on feedback results. It compares the actual conversion rate with the preset target conversion rate to obtain a conversion rate deviation value. Based on this deviation value, it uses either a PID algorithm or a gradient descent algorithm to calculate the adjustment amounts for the first and second weights, respectively. The first and second weights are then increased by the corresponding adjustment amounts to obtain updated first and second weights. By introducing a closed-loop weight adjustment mechanism based on feedback results, the first and second weights can be adaptively optimized according to the actual business conversion effect. This helps to gradually improve the accuracy of the comprehensive evaluation value in predicting business conversion behavior and tilts the weight allocation towards data domain features that contribute more to business.

[0078] In some embodiments, the cross-domain fusion computing system receives feedback records from the business system within a preset time window after outputting a decision label. These feedback records contain a Boolean flag indicating whether each user who received the decision label responded to the business instruction. The cross-domain fusion computing system counts the number of users who responded to the business instruction within the preset time window and the total number of users who received the decision label, calculating the ratio of these two as the actual conversion rate. The cross-domain fusion computing system obtains a pre-configured target conversion rate and calculates the difference between the actual conversion rate and the target conversion rate as the conversion rate deviation value. The cross-domain fusion computing system employs a proportional-integral-differential algorithm to calculate the proportional, integral, and differential terms respectively, and uses the sum of these three terms as a common adjustment amount for the first and second weights. Then, the first and second weights are increased by this common adjustment amount to obtain updated first and second weights. This embodiment uses a PID algorithm for weight adjustment, which is beneficial for achieving a smooth and stable weight convergence process and reducing drastic weight changes caused by single feedback fluctuations.

[0079] In other embodiments, the cross-domain fusion computing system statistically analyzes the actual conversion rates under the first, second, and third scenario identifiers based on feedback results, and compares them with the target conversion rates for each scenario to obtain the conversion rate deviation value for each scenario. The cross-domain fusion computing system employs a gradient descent algorithm, with the optimization objective of minimizing the sum of squares of the conversion rate deviation values ​​under each scenario. It calculates the gradients of the first and second weights under each scenario, performs a weighted average of the gradients for each scenario to obtain the comprehensive gradient, determines the adjustment amount for the first and second weights based on the comprehensive gradient, and increases the first and second weights by the corresponding adjustment amounts to obtain the updated first and second weights. This embodiment, by statistically analyzing feedback results for each scenario and using the gradient descent algorithm to optimize weights, makes weight updates more targeted and data-driven, improving the adaptability of weight configuration under different business scenarios.

[0080] In some optional embodiments, the identity matching protocol is used to identify the de-identified user identifier shared by the first data domain and the second data domain without exposing the unmatched user identifier, and to return the second domain feature label for the shared de-identified user identifier.

[0081] For example, the identity matching protocol can represent a cryptographic security protocol that allows participating parties to determine a shared set of user identifiers without disclosing unmatched user identifiers within their respective domains. Unmatched user identifiers can refer to anonymized user identifiers existing only in the first data domain or only in the second data domain. The cross-domain fusion computing system initiates a collaborative computing request through the identity matching protocol, causing the second data domain to return second domain feature labels only for the shared anonymized user identifiers, while being unable to obtain anonymized user identifier information not shared in the first data domain. Furthermore, the cross-domain fusion computing system is also unable to obtain user identifiers and their feature information not shared in the second data domain. The method provided in this application embodiment can be executed using a trusted data space, in which the cross-domain fusion computing system establishes secure connections with both the first and second data domains, completing joint scoring and decision output of cross-domain features without interacting with underlying detailed data. By using this identity matching protocol, the cross-domain fusion computing system can securely complete the matching and feature collaboration of cross-domain user identities while ensuring that the data assets of the first and second data domains are not excessively exposed. This is conducive to privacy protection in the process of cross-domain data fusion and meets the security and compliance requirements that data is usable but not visible.

[0082] In some embodiments, the cross-domain fusion computing system and the second data domain employ a privacy-preserving intersection protocol based on hash functions as the identity matching protocol. For example, the cross-domain fusion computing system maps de-identified user identifiers using a preset hash function to obtain a first set of hash values, and sends the first set of hash values ​​to the second data domain. The second data domain maps the de-identified user identifiers stored within its domain using the same hash function to obtain a second set of hash values. The second data domain performs an intersection calculation on the first and second sets of hash values ​​to obtain a hash value intersection, and identifies the de-identified user identifiers corresponding to the hash value intersection as shared de-identified user identifiers. The second data domain only queries the corresponding second-domain feature tags for the shared de-identified user identifiers and returns the query results to the cross-domain fusion computing system. This embodiment achieves identity matching through a hash privacy intersection protocol, which is beneficial for confirming shared users without exposing the original de-identified user identifiers and unmatched user identifiers, resulting in high computational efficiency.

[0083] In other embodiments, the cross-domain fusion computing system and the second data domain employ a privacy-preserving intersection protocol based on unintentional transmission as the identity matching protocol. The cross-domain fusion computing system encodes de-identified user identifiers as elements of a Bloom filter and sends the Bloom filter to the second data domain. The second data domain uses the Bloom filter to perform an existence query on each de-identified user identifier within the second data domain, obtaining a list of user identifiers present in the Bloom filter as a candidate set of shared de-identified user identifiers. The second data domain returns the candidate set to the cross-domain fusion computing system, which confirms the final shared de-identified user identifiers through two hash mappings. The second data domain returns the corresponding second-domain feature label for the confirmed shared de-identified user identifiers. This embodiment, through the identity matching protocol based on Bloom filters and unintentional transmission, facilitates cross-domain identity matching while ensuring a very high level of privacy and security, and reduces communication overhead.

[0084] Figure 3 The flowchart illustrates a user feature evaluation method based on cross-domain data fusion provided in this application embodiment. After the cross-domain fusion computing system initiates the evaluation process, it first receives an evaluation request from the business system, which carries a de-identified user identifier. The cross-domain fusion computing system parses the de-identified user identifier in the evaluation request and performs a cross-domain feature acquisition operation based on the de-identified user identifier. Specifically, it extracts the first domain feature tag corresponding to the user in the first data domain and initiates a collaborative computing request to the second data domain based on an identity matching protocol to obtain the second domain feature tag of the user in the second data domain.

[0085] The cross-domain fusion computing system inputs the acquired first-domain feature labels and second-domain feature labels into the fusion computing engine, which calculates a comprehensive evaluation value. The comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first-domain and second-domain feature labels. Subsequently, the cross-domain fusion computing system performs a strategy-based dimensionality reduction judgment: it acquires a preset first evaluation value threshold and a second evaluation value threshold, where the first evaluation value threshold is greater than the second evaluation value threshold; it then detects the relationship between the comprehensive evaluation value and both the first and second evaluation value thresholds. If the comprehensive evaluation value is greater than or equal to the first evaluation value threshold, a first decision label is output, corresponding to a high-level business execution instruction; if the comprehensive evaluation value is less than the first evaluation value threshold but greater than or equal to the second evaluation value threshold, a second decision label is output, corresponding to a medium-level business execution instruction; if the comprehensive evaluation value is less than the second evaluation value threshold, a third decision label is output, corresponding to a low-level business execution instruction.

[0086] After outputting the decision label, the cross-domain converged computing system can optionally generate a billing audit code. This code records the service call information corresponding to this evaluation request, enabling metering and compliance auditing of the cross-domain converged computing service. This facilitates support for the controllable and measurable data flow characteristics of the trusted data space. Generating the billing audit code is optional. If this step is not performed, the cross-domain converged computing system can directly generate a decision label without the original information of the first or second domain feature label and return it to the business system, ending the evaluation process. The business system then executes the corresponding business instructions based on the returned decision label.

[0087] See Figure 4According to another aspect of the embodiments of this application, a user feature evaluation device based on cross-domain data fusion is also provided, comprising: a request receiving unit, configured to receive an evaluation request for user features from a business system; a first tag acquisition unit, configured to extract a first domain feature tag corresponding to the user in a first data domain based on a de-identified user identifier in the evaluation request, wherein the first data domain is an air travel data domain, and the first domain feature tag includes at least one of the user's flight activity, cabin class preference, and route preference; and a second tag acquisition unit, configured to initiate a collaborative computing request to the second data domain based on the de-identified user identifier through an identity matching protocol, and acquire the user's second tag in the second domain. The data domain includes a second domain feature label, wherein the second data domain is a communication behavior data domain, and the second domain feature label includes at least one of the user's consumption capacity level and the activity level of travel-related applications; an evaluation processing unit is used to determine a comprehensive evaluation value based on the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label; and a decision processing unit is used to generate a decision label based on the comparison result of the comprehensive evaluation value and a preset threshold, wherein the decision label is a discrete business execution instruction, and the decision label does not contain the original information of the first domain feature label or the second domain feature label.

[0088] Optionally, the evaluation processing unit includes: a numerical conversion subunit, used to convert the first domain feature label into a first numerical feature and the second domain feature label into a second numerical feature according to a preset numerical mapping rule; a normalization processing subunit, used to input the first numerical feature into a first normalization function to obtain a first normalized value; and input the second numerical feature into a second normalization function to obtain a second normalized value; wherein the first normalization function and the second normalization function are both nonlinear normalization functions; the first normalization function and the second normalization function are the same function, or are different functions adapted to the feature distributions of different data domains respectively; a weight acquisition subunit, used to acquire a first weight corresponding to the first normalized value and a second weight corresponding to the second normalized value; and a weighted summation subunit, used to perform a weighted summation calculation on the first normalized value and the second normalized value according to the first weight and the second weight to obtain a comprehensive evaluation value.

[0089] Optionally, the weight acquisition subunit includes: a scenario detection module, used to detect the business scenario identifier of the business system, wherein the business scenario identifier includes at least a first type of scenario identifier, a second type of scenario identifier, and a third type of scenario identifier. The first type of scenario identifier corresponds to a wake-up scenario where the user's flight activity is lower than a preset activity threshold and the communication behavior activity is higher than a preset communication activity threshold; the second type of scenario identifier corresponds to a scenario where the user's historical cabin class consumption level is higher than a preset cabin class level; and the third type of scenario identifier corresponds to a scenario for screening potential passengers for newly opened routes. A first weight configuration module is used to determine that a second weight is greater than a first weight when the business scenario identifier of the business system is detected to be a first type of scenario identifier. A second weight configuration module is used to determine that a first weight is greater than a second weight when the business scenario identifier of the business system is detected to be a second type of scenario identifier. A third weight configuration module is used to determine that a first weight is equal to a second weight when the business scenario identifier of the business system is detected to be a third type of scenario identifier.

[0090] Optionally, the decision processing unit includes: a threshold acquisition subunit, used to acquire a preset first evaluation value threshold and a second evaluation value threshold, wherein the first evaluation value threshold is greater than the second evaluation value threshold; a first comparison subunit, used to detect whether the comprehensive evaluation value is greater than or equal to the first evaluation value threshold; a first output subunit, used to output a first decision label when the comprehensive evaluation value is detected to be greater than or equal to the first evaluation value threshold; a second comparison subunit, used to detect whether the comprehensive evaluation value is greater than or equal to the second evaluation value threshold when the comprehensive evaluation value is detected to be less than the first evaluation value threshold; a second output subunit, used to output a second decision label when the comprehensive evaluation value is detected to be greater than or equal to the second evaluation value threshold; and a third output subunit, used to output a third decision label when the comprehensive evaluation value is detected to be less than the second evaluation value threshold; wherein the recommendation level of the business execution instruction corresponding to the first decision label is higher than the recommendation level of the business execution instruction corresponding to the second decision label, and the recommendation level of the business execution instruction corresponding to the second decision label is higher than the recommendation level of the business execution instruction corresponding to the third decision label; the business execution instruction contains a strategy level and a recommended action in the form of a structured instruction.

[0091] Optionally, the user feature evaluation device based on cross-domain data fusion further includes: a timestamp acquisition unit, used to acquire the timestamp corresponding to the first numerical feature after the numerical conversion subunit converts the first domain feature label into the first numerical feature according to the preset numerical mapping rule; an attenuation coefficient determination unit, used to determine the time attenuation coefficient according to the time difference between the current system time and the timestamp, wherein the larger the time difference, the smaller the time attenuation coefficient; and a first numerical feature update unit, used to multiply the first numerical feature by the time attenuation coefficient to obtain the updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0092] Optionally, the user feature evaluation device based on cross-domain data fusion further includes: a trip information acquisition unit, used to acquire the user's trip information after the numerical conversion subunit converts the first domain feature label into a first numerical feature according to a preset numerical mapping rule, wherein the trip information includes flight number, scheduled departure time, and scheduled arrival time; a dynamic status acquisition unit, used to acquire the dynamic status information of the flight number, wherein the dynamic status information includes at least punctuality status, delay duration, and cancellation flag; a trip impact coefficient determination unit, used to determine the trip impact coefficient based on the dynamic status information, including: determining the trip impact coefficient as a first coefficient when the punctuality status is punctual; determining the trip impact coefficient as a second coefficient when the punctuality status is delayed and the delay duration exceeds a preset duration threshold, wherein the second coefficient is less than the first coefficient; determining the trip impact coefficient as a third coefficient when the cancellation flag indicates flight cancellation, wherein the third coefficient is less than the second coefficient; and a second numerical feature update unit, used to multiply the first numerical feature by the trip impact coefficient to obtain an updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0093] Optionally, the user feature evaluation device based on cross-domain data fusion further includes: a historical flight record acquisition unit, used to acquire the user's historical flight records after the numerical conversion subunit converts the first domain feature labels into first numerical features according to a preset numerical mapping rule, wherein the historical flight records include the takeoff time, landing time, departure airport, and arrival airport for each flight; an activity conversion unit, used to convert the flight activity in the first domain feature labels into flight activity values; an activity decay coefficient determination unit, used to determine the flight activity decay coefficient based on the time difference between the current time and the most recent flight in the historical flight records, wherein the flight activity decay coefficient decays as the time difference increases; and a time-lapse processing unit, used to multiply the flight activity value by... The attenuation coefficient yields the flight activity value after time-lapse attenuation; the travel probability prediction unit obtains flight schedule information within a future preset time window and predicts the user's travel probability within the future preset time window based on the user's route preferences and flight schedule information. The flight schedule information includes flight time, route, and remaining cabin capacity. The travel probability is calculated based on the similarity between the user's past flight routes and future flight routes, as well as cabin capacity matching. The third numerical feature update unit uses preset activity weights and travel probability weights to perform a weighted summation of the flight activity value after time-lapse attenuation and the travel probability to obtain the updated first numerical feature. The updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

[0094] Optionally, the user feature evaluation device based on cross-domain data fusion further includes: a feedback recording unit, used to record the feedback result after the business system executes the business instruction according to the decision tag after the weighted summation subunit calculates the comprehensive evaluation value by weighted summation of the first normalized value and the second normalized value according to the first weight and the second weight, wherein the feedback result includes at least one of whether the user responded to the business instruction and the actual consumption behavior after responding to the business instruction; a conversion rate calculation unit, used to calculate the actual conversion rate in the current business scenario according to the feedback result, wherein the actual conversion rate is the ratio of the number of users who responded to the business instruction to the total number of users who received the decision tag; a deviation comparison unit, used to compare the actual conversion rate with the preset target conversion rate to obtain a conversion rate deviation value; a weight adjustment calculation unit, used to calculate the adjustment amount of the first weight and the second weight respectively according to the conversion rate deviation value by using a PID algorithm or a gradient descent algorithm; and a weight update unit, used to increase the first weight and the second weight respectively by the corresponding adjustment amount to obtain the updated first weight and the updated second weight.

[0095] Optionally, the identity matching protocol is used to identify the de-identified user identifier shared by the first data domain and the second data domain without exposing the unmatched user identifier, and to return the second domain feature label for the shared de-identified user identifier.

[0096] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the above-described user feature evaluation method based on cross-domain data fusion.

[0097] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described user feature evaluation method based on cross-domain data fusion.

[0098] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions, which, when executed by a processor, implement the above-described user feature evaluation method based on cross-domain data fusion.

[0099] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0100] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0102] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0105] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A user feature evaluation method based on cross-domain data fusion, characterized in that, include: Receive evaluation requests regarding user characteristics from business systems; Based on the de-identified user identifier in the evaluation request, the first domain feature label corresponding to the user is extracted in the first data domain, wherein the first data domain is the air travel data domain, and the first domain feature label includes at least one of the user's flight activity, cabin class preference, and route preference; Based on the de-identified user identifier, a collaborative computing request is initiated to the second data domain through an identity matching protocol to obtain the user's second domain feature tag in the second data domain. The second data domain is a communication behavior data domain, and the second domain feature tag includes at least one of the user's consumption capacity level and the activity level of travel-related applications. A comprehensive evaluation value is determined based on the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label; A decision label is generated based on the comparison result between the comprehensive evaluation value and the preset threshold. The decision label is a discrete business execution instruction, and the decision label does not contain the original information of the first domain feature label or the second domain feature label.

2. The method according to claim 1, characterized in that, The comprehensive evaluation value is determined based on the first domain feature label and the second domain feature label, including: According to the preset numerical mapping rules, the first domain feature label is converted into a first numerical feature, and the second domain feature label is converted into a second numerical feature; The first numerical feature is input into the first normalization function to obtain the first normalized value; the second numerical feature is input into the second normalization function to obtain the second normalized value; wherein, the first normalization function and the second normalization function are both nonlinear normalization functions; the first normalization function and the second normalization function are the same function, or are different functions adapted to the feature distribution of different data domains respectively; Obtain the first weight corresponding to the first normalized value and the second weight corresponding to the second normalized value; The first normalized value and the second normalized value are weighted and summed according to the first weight and the second weight to obtain the comprehensive evaluation value.

3. The method according to claim 2, characterized in that, Obtaining the first weight corresponding to the first normalized value and the second weight corresponding to the second normalized value includes: The system detects business scenario identifiers, which include at least a first type of scenario identifier, a second type of scenario identifier, and a third type of scenario identifier. The first type of scenario identifier corresponds to a wake-up scenario where the user's flight activity level is lower than a preset activity threshold and their communication behavior activity level is higher than a preset communication activity threshold. The second type of scenario identifier corresponds to a scenario where the user's historical cabin class consumption level is higher than a preset cabin class level. The third type of scenario identifier corresponds to a scenario where potential passengers are screened for newly opened routes. If the business scenario identifier of the business system is detected to be a first type of scenario identifier, it is determined that the second weight is greater than the first weight; If the business scenario identifier of the business system is detected to be a second type of scenario identifier, it is determined that the first weight is greater than the second weight; If the business scenario identifier of the business system is detected to be a third-class scenario identifier, the first weight is determined to be equal to the second weight.

4. The method according to claim 1, characterized in that, Based on the comparison between the comprehensive evaluation value and the preset threshold, a decision label is generated, including: Obtain a preset first evaluation value threshold and a second evaluation value threshold, wherein the first evaluation value threshold is greater than the second evaluation value threshold; Detect whether the comprehensive evaluation value is greater than or equal to the first evaluation value threshold; If the comprehensive evaluation value is detected to be greater than or equal to the first evaluation value threshold, a first decision label is output; If the overall evaluation value is detected to be less than the first evaluation value threshold, it is detected whether the overall evaluation value is greater than or equal to the second evaluation value threshold. If the overall evaluation value is detected to be greater than or equal to the second evaluation value threshold, a second decision label is output; If the overall evaluation value is detected to be less than the second evaluation value threshold, a third decision label is output; The recommendation level of the business execution instruction corresponding to the first decision tag is higher than that of the business execution instruction corresponding to the second decision tag, and the recommendation level of the business execution instruction corresponding to the second decision tag is higher than that of the business execution instruction corresponding to the third decision tag. The business execution instruction includes a strategy level and a recommended action in the form of a structured instruction.

5. The method according to claim 2, characterized in that, After converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: Obtain the timestamp corresponding to the first numerical feature; The time decay coefficient is determined based on the time difference between the current system time and the timestamp, wherein the larger the time difference, the smaller the time decay coefficient. The first numerical feature is multiplied by the time decay coefficient to obtain the updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

6. The method according to claim 2, characterized in that, After converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: Obtain the user's travel information, wherein the travel information includes flight number, scheduled departure time, and scheduled arrival time; Obtain the dynamic status information of the flight number, wherein the dynamic status information includes at least the on-time status, delay duration, and cancellation flag; Determining the trip impact coefficient based on the dynamic status information includes: determining the trip impact coefficient as a first coefficient when the on-time status is on-time; determining the trip impact coefficient as a second coefficient when the on-time status is delayed and the delay duration exceeds a preset duration threshold, wherein the second coefficient is less than the first coefficient; and determining the trip impact coefficient as a third coefficient when the cancellation flag indicates flight cancellation, wherein the third coefficient is less than the second coefficient. The first numerical feature is multiplied by the travel influence coefficient to obtain the updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

7. The method according to claim 2, characterized in that, After converting the first domain feature label into a first numerical feature according to a preset numerical mapping rule, the method further includes: Obtain the user's historical flight records, wherein the historical flight records include the takeoff time, landing time, departure airport, and arrival airport for each flight; Convert the flight activity level in the first domain feature label into a flight activity value; The flight activity attenuation coefficient is determined based on the time difference between the current time and the most recent flight in the historical flight record, wherein the flight activity attenuation coefficient decreases as the time difference increases; Multiply the flight activity value by the attenuation coefficient to obtain the flight activity value after time attenuation; Obtain flight schedule information within a future preset time window, and predict the user's travel probability within the future preset time window based on the user's route preferences and the flight schedule information. The flight schedule information includes flight time, route, and remaining cabin capacity. The travel probability is calculated based on the similarity between the user's past flight routes and future flight routes, as well as the cabin capacity matching degree. Using preset activity weights and travel probability weights, the flight activity value after time decay is weighted and summed with the travel probability to obtain an updated first numerical feature, wherein the updated first numerical feature is used to replace the original first numerical feature to redetermine the comprehensive evaluation value.

8. The method according to claim 2, characterized in that, After calculating the comprehensive evaluation value by weighted summation of the first normalized value and the second normalized value based on the first weight and the second weight, the method further includes: Record the feedback results after the business system executes the business instruction according to the decision tag, wherein the feedback results include at least one of whether the user responded to the business instruction and the actual consumption behavior after responding to the business instruction; Based on the feedback results, the actual conversion rate in the current business scenario is calculated, wherein the actual conversion rate is the ratio of the number of users responding to the business instructions to the total number of users receiving the decision tags; The actual conversion rate is compared with the preset target conversion rate to obtain the conversion rate deviation value; Based on the conversion rate deviation value, the adjustment amounts of the first weight and the second weight are calculated using either the PID algorithm or the gradient descent algorithm. The first weight and the second weight are increased by the corresponding adjustment amount to obtain the updated first weight and the updated second weight.

9. The method according to claim 1, characterized in that, The identity matching protocol is used to identify the de-identified user identifier shared by the first data domain and the second data domain without exposing the unmatched user identifier, and to return the second domain feature label for the shared de-identified user identifier.

10. A user feature evaluation device based on cross-domain data fusion, characterized in that, include: The request receiving unit is used to receive evaluation requests regarding user characteristics from the business system. The first tag acquisition unit is used to extract the first domain feature tag corresponding to the user in the first data domain according to the de-identified user identifier in the evaluation request. The first data domain is the air travel data domain, and the first domain feature tag includes at least one of the user's flight activity, cabin class preference, and route preference. The second tag acquisition unit is used to initiate a collaborative computing request to the second data domain based on the de-identified user identifier through an identity matching protocol to obtain the user's second domain feature tag in the second data domain, wherein the second data domain is a communication behavior data domain, and the second domain feature tag includes at least one of the user's consumption capacity level and the activity level of travel-related applications. An evaluation processing unit is configured to determine a comprehensive evaluation value based on the first domain feature label and the second domain feature label, wherein the comprehensive evaluation value is a dimensionless value obtained by weighted fusion of the first domain feature label and the second domain feature label; The decision processing unit is used to generate a decision label based on the comparison result between the comprehensive evaluation value and the preset threshold, wherein the decision label is a discrete business execution instruction, and the decision label does not contain the original information of the first domain feature label or the second domain feature label.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device on which the computer-readable storage medium is located performs the user feature evaluation method based on cross-domain data fusion as described in any one of claims 1 to 9.

12. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the user feature evaluation method based on cross-domain data fusion as described in any one of claims 1 to 9.

13. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the user feature evaluation method based on cross-domain data fusion as described in any one of claims 1 to 9.

Images