Core demand and potential demand collaborative recommendation method based on behavior attribution residual and trusted exposure interval ranking

Abstract

The present application relates to the technical field of big data analysis, and particularly relates to a core demand and potential demand collaborative recommendation method based on behavior attribution residual and credible exposure interval sorting, comprising the following steps: S1: a data construction module performs multi-source data acquisition and preprocessing; S2: a demand representation learning module performs behavior credibility calculation, core demand attribution and potential demand construction; S3: a personalized recommendation module performs behavior demand matching, core demand protection type scoring and credible exposure interval sorting; S4: a feedback updating module performs recommendation output and user confirmation and model feedback updating. The present application can fully mine potential demand, and can collaboratively match based on potential demand and core demand, thereby improving demand mining accuracy and comprehensiveness.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking. Background Technology

[0002] With the rapid development of big data analytics, personalized recommendations are increasingly emerging to meet people's individual needs. However, current personalized recommendations often have the following shortcomings: they tend to extract only core needs while neglecting to uncover potential needs, resulting in limitations and poor accuracy in data mining. Summary of the Invention

[0003] This invention aims to provide a collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking, which can fully explore potential needs and improve the accuracy and comprehensiveness of demand mining by synergistically matching potential and core needs. The method includes the following steps:

[0004] S1: The data construction module performs multi-source data acquisition and preprocessing, specifically including:

[0005] S11: Obtain user behavior, core needs, auxiliary services, scenarios, and recommendation feedback data;

[0006] S12: Form a recommendation sample with uniform behavior type, auxiliary service status, time benchmark and numerical range;

[0007] S2: The requirement representation learning module performs behavior credibility calculation, core requirement attribution, and potential requirement construction;

[0008] S3: The personalized recommendation module performs behavioral demand matching, core demand protection scoring, and credible exposure range sorting.

[0009] S4: The feedback update module executes the recommended output, user confirmation, and model feedback updates;

[0010] Specifically, S3 includes:

[0011] S31: Encode the service objectives, service limitations, and effective time periods in the normalized core requirement data according to the field order to obtain the core requirement representation; read the service function, applicable conditions, service location, available time period, preparation time, service price, historical fulfillment status, and text description according to the auxiliary service identifier. Perform category encoding on the category field, normalize the time, price, and fulfillment rate, and semantically encode the text description. Then, concatenate the encoding results according to a fixed field order and map them to the same vector dimension as the core requirement representation and potential requirement representation to obtain the auxiliary service representation;

[0012] The degree of matching between auxiliary services and core requirements, and the degree of matching between auxiliary services and potential requirements are calculated separately, as shown below:

[0013] ,

[0014] ,

[0015] in, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment The core requirement is stated as follows: Indicates user At any moment Potential demand indicates, This represents the core requirement matching matrix. Represents the potential demand matching matrix. Indicates at time The auxiliary services indicate that This represents a logical function that converts the calculation result to a value between zero and one. Indicates vector transpose;

[0016] S32: Read the core requirement matching degree, potential requirement matching degree, potential requirement confidence coefficient, auxiliary service data, and scenario data. Compare the applicable conditions of auxiliary services with the service restrictions in the core requirements item by item, and compare the available time of auxiliary services with the expected service time of the current recommendation task. When all service restrictions are met and the auxiliary service is available within the expected service time, set the real-time recommendable flag to 1; when any service restriction is not met or the auxiliary service is unavailable, set the real-time recommendable flag to 0.

[0017] Personalized matching values ​​are calculated using the following formula:

[0018] ,

[0019] in, Indicates user For ancillary services At any moment Personalized matching value, Indicates user For ancillary services At any moment Recommended tags, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment Confidence coefficient of potential demand This represents the potential demand over-weighting inhibition coefficient;

[0020] S33: Read personalized matching values, user behavior sets, potential demand evidence weights, auxiliary service status sequences, and historical recommendation records. First, determine whether each user behavior is related to the auxiliary service to be ranked based on the auxiliary service identifier in the user behavior. Then, calculate the potential demand evidence weights related to the auxiliary service to be ranked. At the same time, calculate the degree of fluctuation of the auxiliary service status based on the frequency of change of the auxiliary service between the available and unavailable states within a preset historical period.

[0021] Compared with the prior art, this method has the following beneficial effects:

[0022] 1. By calculating the credibility of behavior and the degree of attribution of core needs, we can extract evidence of potential needs from credible user behaviors that cannot be fully explained by core needs, reduce the interference of factors such as accidental touches, accidental browsing and service unavailability on the identification of needs, and improve the accuracy of potential needs mining.

[0023] 2. Construct a core-needs-protected personalized scoring mechanism to increase the personalized matching value when potential needs and core needs are synergistically matched, thereby preventing potential needs from covering the user's clearly identified core needs and improving the consistency between recommendation results and the user's actual needs;

[0024] 3. Construct a reliable exposure comparison sample based on the actual display status and real-time availability status of auxiliary services to avoid incorrectly labeling unavailable and undisplayed services as negative preferences, thereby reducing exposure bias and service supply bias.

[0025] 4. Utilize the degree of matching uncertainty to construct a reliable exposure interval ranking model, and update the model by combining user selection results, actual usage results, and service fulfillment results to improve the stability, dynamic adaptability, and executability of personalized recommendation results. Attached Figure Description

[0026] Figure 1This is a flowchart illustrating the collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking, as presented in this invention. Detailed Implementation

[0027] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described below are only for explaining the present invention and do not limit the scope of protection of the present invention.

[0028] The present invention will be further described in detail below through preferred embodiments:

[0029] As attached Figure 1 As shown: A collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking includes the following steps:

[0030] S1: The data construction module performs multi-source data acquisition and preprocessing, specifically including the following steps:

[0031] S11: Obtain user behavior, core needs, auxiliary services, scenarios, and recommendation feedback data;

[0032] Users create the current recommendation task through their user terminal, inputting the service objective, service time, and service location, and confirming the user status and service restrictions related to service execution. The user terminal assigns a user identifier to the user and a task identifier to the current recommendation task. The user identifier is used to associate data of the same user across different time periods, while the task identifier is used to distinguish recommendation tasks created by the same user at different times.

[0033] The core requirement data originates from information actively input and confirmed by users when creating the current recommendation task. The system converts user status and service constraints into core requirements that can be directly compared to assistive service attributes, according to a pre-defined requirement mapping relationship. Taking the scenario of recommending mobility-assisted travel services as an example, using a wheelchair becomes the core requirement that the assistive service should support wheelchair access; disallowing stair access becomes the core requirement that the assistive service must not include stair access; requiring human assistance becomes the core requirement that the assistive service should provide human accompaniment; and the expected arrival time becomes the time constraint of the assistive service. Model inference results that have not been confirmed by the user are not included in the core requirement data.

[0034] User behavior data originates from user terminal interaction logs and auxiliary service platform business records. Interaction logs record browsing, querying, consulting, and display behaviors, while business records record appointment, cancellation, actual usage, and evaluation behaviors. Each piece of user behavior data includes a user identifier, task identifier, auxiliary service identifier, behavior type, time of occurrence, duration of occurrence, result of occurrence, and auxiliary service status at the time of occurrence.

[0035] Ancillary service data originates from ancillary service platforms and service providers, including ancillary service identifiers, service functions, applicable conditions, service locations, available time periods, preparation time, service prices, and historical fulfillment status. Scenario data comes from the current recommendation task and facility status interfaces, including the current time, estimated service period, travel node, and facility status. Recommendation feedback data comes from recommendation result display records, user selection records, service order records, and service fulfillment records.

[0036] S12: Form a recommendation sample with uniform behavior type, auxiliary service status, time benchmark and numerical range;

[0037] In step S11, the multi-source data collected is input into the data preprocessing submodule. The user identifier is used as the first association key to associate the core demand data, user behavior data, and recommendation feedback data generated by the same user. The task identifier is used as the second association key to associate the core demand data, user behavior data, and scenario data belonging to the current recommendation task. The auxiliary service identifier is used as the third association key to associate each user behavior with the corresponding auxiliary service data and the auxiliary service status when the behavior occurs.

[0038] For historical user behavior data that does not carry a task identifier, if its user identifier is the same as the current user identifier, the service function of the corresponding auxiliary service is consistent with the service goal of the current recommendation task, and the time of the behavior is within the preset historical time range, it will be associated with the current recommendation task as historical user behavior; historical user behavior data that does not meet the above conditions will not participate in the current recommendation calculation.

[0039] Unrelated user behavior data is arranged according to the time of occurrence. For similar behaviors generated consecutively by the same user for the same auxiliary service within a preset merging period, consecutive behaviors are merged into one user behavior, with the cumulative duration and the result of the last behavior used as the merging result. For user behaviors that are actually used after reservation, a subsequent confirmation relationship is established between the reservation behavior and the actual usage behavior based on the same auxiliary service identifier. For unselected records generated during periods when auxiliary services are unavailable, only the facts and auxiliary service status are retained, and they are not marked as negative preferences.

[0040] The user behavior data that has been associated, sorted, and merged will be combined into a user behavior set. .in, Indicates the user index. This indicates the current calculation time; the user behavior set only includes the time up to now. User behaviors that can be associated with the current task. Perform abnormal duration correction and missing auxiliary service status completion on user behavior data. Unify the encoding of behavior types as browsing, querying, consulting, scheduling, canceling, actual use, and evaluation; unify the encoding of auxiliary service status as available, unavailable, and unknown. Perform numerical normalization on core requirement data, user behavior data, auxiliary service data, and scenario data, ensuring that the degree values ​​used in subsequent calculations are between zero and one.

[0041] S2: The requirement representation learning module performs behavior credibility calculation, core requirement attribution, and potential requirement construction, specifically including the following steps:

[0042] S21: Read user behavior set Regarding the first For each user behavior, the basic credibility of the behavior type is obtained from the behavior type field; the degree of completion of the behavior is obtained from the behavior duration and operation completion status; the degree of confirmation of subsequent behaviors is obtained from subsequent user behaviors with the same auxiliary service identifier; and the degree of consistency of auxiliary service status is obtained from the auxiliary service status at the time of the behavior and the behavior result. All four data items are derived from the same user behavior and its associated records, avoiding the use of external data that cannot be mapped to this user behavior.

[0043] Basic credibility of behavior type The degree of behavior completion is determined based on the proportion of each behavior type that translates into actual usage behavior in historical recommendation feedback data. The degree of confirmation for subsequent actions is determined based on the ratio of the duration of the behavior to the reference duration for the corresponding behavior type and whether the operation has been completed. The degree of consistency in the status of an auxiliary service is determined based on the progressive relationship between browsing, querying, consulting, booking, and actual use of the same auxiliary service. The degree is determined based on whether the auxiliary service status is complete at the time of the behavior and whether the auxiliary service status is consistent with the behavior outcome. All four degree values ​​are normalized to between zero and one.

[0044] Cancellation is not directly considered a negative action. When cancellation is caused by the unavailability of ancillary services, changes in service hours, or failure of the service provider to fulfill its obligations, the consistency of the ancillary service status is used to retain true status information, but the cancellation does not generate a negative preference. When the ancillary service is available and the user actively cancels, the credibility of the action is determined by combining subsequent user behavior.

[0045] No. The credibility of a user's behavior is calculated using the following formula:

[0046]

[0047] in, Indicates user At any moment The The credibility of user behavior. Indicates the first The basic credibility of the behavior type to which the user behavior belongs. Indicates the first User behavior types Indicates user At any moment The The degree of completion of user behavior. Indicates user At any moment The The degree of confirmation of subsequent user behavior. Indicates user At any moment The The degree of consistency between the status of auxiliary services for user behavior. , and The three weights represent the degree of completion of the action, the degree of confirmation of subsequent actions, and the degree of consistency of the status of auxiliary services, respectively, and the sum of the three weights is greater than zero.

[0048] Basic credibility of behavior type Initialize based on the proportion of each behavior type that translates into actual usage behavior in historical recommendation feedback data. Three non-negative weights. , and The training is performed using historical user behavior samples with actual usage results, and the samples are normalized after training. When there are insufficient historical user behavior samples, the three non-negative weights use the same initial value.

[0049] S22: Read user behavior set Behavioral credibility The core requirements state that, for each user behavior, the semantic consistency, object consistency, and temporal consistency between the user behavior and each core requirement are calculated based on the auxiliary service data associated with the user behavior and the time when the behavior occurs.

[0050] Semantic consistency The service function description of the auxiliary service corresponding to user behavior and the first Semantic similarity is determined between core requirement descriptions to identify user behaviors and core requirements that are described differently but have similar meanings. Object consistency is also assessed. The service function tags of auxiliary services and the first Whether the requirement tags for the core requirements correspond and are clearly defined is used to confirm whether the two point to the same type of auxiliary service. (Time sequence consistency) The time of the behavior, the period during which ancillary services are available, and the first The degree of overlap between the effective time periods of the three core requirements is determined to exclude historical user behaviors that occurred at times unrelated to the current effective time period of the core requirement. The degree of consistency for each of the three requirements is determined by the [missing information - likely a specific factor]. User behavior, the first The core requirements and the related auxiliary service data are calculated and normalized to between zero and one.

[0051] No. User behavior can be detected by the first The degree of core requirement attribution in the explanation of core requirements is calculated using the following formula:

[0052]

[0053] in, Indicates user At any moment The User behavior can be detected by the first The degree of attribution of core requirements explained in the core requirements section Indicates user At any moment The User behavior can be detected by the first The degree of semantic consistency in the explanation of the core requirements of the item. Indicates user At any moment The User behavior can be detected by the first The degree of consistency between the objects explained in the core requirements of the item Indicates user At any moment The User behavior can be detected by the first The degree of temporal consistency in the explanation of the core requirements of the project. , and The non-negative attribution weights represent semantic consistency, object consistency, and temporal consistency, respectively, and the sum of the three attribution weights is: .

[0054] Three Attribution Weights , and The training is performed using historical samples of users who have confirmed their core needs and subsequently used the auxiliary services, and the samples are then normalized. When historical samples are insufficient, the three attribution weights use the same initial value to ensure that semantic consistency, object consistency, and temporal consistency have equal initial contributions. Core Needs Attribution Degree This is used to reduce the contribution of user behavior that has already been explained by core requirements during the process of building potential requirements, and to prevent the same user behavior from being used repeatedly to characterize core requirements and potential requirements.

[0055] S23: Read user behavior set Behavioral credibility Degree of core demand attribution Based on the scenario data, the scenario correlation degree, spatial proximity, temporal overlap and substitution degree are calculated and normalized to between zero and one according to the spatial proximity between the service location of the auxiliary service corresponding to the user behavior and the current travel node, the temporal overlap between the available time of the auxiliary service and the expected travel time, and the degree of substitution of the auxiliary service for the real-time facility status. Then, the scenario correlation degree is obtained by weighted summation according to the preset non-negative weights.

[0056] Scene relevance Used to determine the first Whether the auxiliary service indicated by the user's behavior is related to the current recommendation task. Future breakpoints are only used to determine the degree to which auxiliary services substitute for real-time facility status; they do not directly generate potential demands, nor do they replace personalized recommendation algorithms. Subsequently, the credibility of the behavior will be... As a control metric for the reliability of user behavior, the maximum value of the attribution degree of core needs is taken as the degree to which user behavior has been explained by core needs, and the degree of scenario relevance is taken as the degree of relevance between user behavior and the current recommendation task. The three together form the potential demand evidence weight.

[0057] No. Weight of evidence of potential needs for user behavior Calculated using the following formula:

[0058]

[0059] in, Indicates user At any moment The Weight of evidence of potential demand for a user behavior item Indicates user At any moment The The credibility of user behavior. Indicates user At any moment The The degree of contextual relevance of user behaviors. Indicates user The core demand quantity Indicates user At any moment The User behavior can be detected by the first The degree of attribution of core requirements explained in the core requirements section Indicates the first The maximum value is taken from all the core needs attribution levels corresponding to each user behavior.

[0060] The above approach assigns lower potential demand evidence weights to user behaviors with low credibility, those already fully explainable by core needs, or those with low relevance to the current recommendation task. Only reliable user behaviors, those not explainable by core needs, and relevant to the current recommendation task are retained. The potential demand evidence weights are then used to aggregate the behavior representations to obtain the potential demand representation:

[0061]

[0062] in, Indicates user At any moment Potential demand indicates, Indicated by user The The behavior representation is obtained by encoding the behavior type, corresponding auxiliary service function, and behavior result of a user behavior. This represents the set of user behaviors output by the data preprocessing submodule. The summation symbol indicates that all user behaviors in the set are summed. This indicates a preset positive number to prevent the denominator from being zero. Indicates user At any moment The The latent demand evidence weighting of user behaviors allows different user behaviors to collectively form a latent demand representation based on their reliability, degree of unexplainedness, and relevance to the context. The latent demand representation retains the common direction of multiple user behaviors and does not directly use a single user behavior as a latent demand.

[0063] To reduce the impact of a single incidental action on potential demand, a confidence coefficient for potential demand is calculated based on the total amount of evidence for potential demand and the degree of consistency between the directions of the evidence for potential demand, as detailed below:

[0064]

[0065] in, Indicates user At any moment Confidence coefficient of potential demand This represents the minimum total amount of evidence determined based on the average total amount of evidence from historical valid potential demand samples. Indicates user At any moment The Weight of evidence of potential demand for a user behavior item Indicated by user The The behavior representation is obtained by encoding the behavior type, corresponding auxiliary service function, and behavior result of a user behavior. This represents the set of user behaviors output by the data preprocessing submodule. This indicates a preset positive number to prevent the denominator from being zero. Represents the L2 norm, The summation symbol indicates that the smaller of two values ​​is taken, and the summation symbol indicates that all user behaviors in the set of user behaviors are summed.

[0066] S3: The personalized recommendation module performs behavioral demand matching, core demand protection scoring, and credible exposure range ranking, specifically including the following steps:

[0067] S31: Encode the service objectives, service limitations, and effective time periods in the normalized core requirement data according to the field order to obtain the core requirement representation. The core requirement representation remains unchanged unless the user actively modifies the core requirements.

[0068] The auxiliary service representation is obtained by retrieving the service function, applicable conditions, service location, available time period, preparation time, service price, historical fulfillment status, and text description based on the auxiliary service identifier. Category fields are categorized, time, price, and fulfillment rate are normalized, and text descriptions are semantically encoded. The encoded results are then concatenated in a fixed field order and mapped to the same vector dimension as the core and potential requirement representations to obtain the auxiliary service representation. When the real-time availability or historical fulfillment status of an auxiliary service changes, only the corresponding auxiliary service representation is updated, without altering other auxiliary service representations.

[0069] The degree of matching between auxiliary services and core requirements, and the degree of matching between auxiliary services and potential requirements are calculated separately, as shown below:

[0070]

[0071]

[0072] in, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment The core requirement is stated as follows: Indicates user At any moment Potential demand indicates, This represents the core requirement matching matrix. Represents the potential demand matching matrix. Indicates at time The auxiliary services indicate that This represents a logical function that converts the calculation result to a value between zero and one. This represents the transpose of a vector. The core demand matching matrix and the potential demand matching matrix are the training parameters of the recommendation model. The initial values ​​are randomly initialized and updated during the process of reducing the interval ranking loss in the credible exposure interval ranking submodule.

[0073] S32: Read the core requirement matching degree, potential requirement matching degree, potential requirement confidence coefficient, auxiliary service data, and scenario data. Compare the applicable conditions of auxiliary services with the service restrictions in the core requirements item by item, and compare the available time of auxiliary services with the expected service time of the current recommendation task. When all service restrictions are met and the auxiliary service is available within the expected service time, set the real-time recommendable flag to one; if any service restriction is not met or the auxiliary service is unavailable, set the real-time recommendable flag to zero.

[0074] Real-time recommendable tags are used to exclude auxiliary services that cannot be executed. Core requirement matching degree is used to indicate the degree to which auxiliary services meet the user's explicit needs. Potential requirement matching degree is used to indicate the degree to which auxiliary services meet potential needs. Potential requirement confidence coefficient is used to control the strength of the influence of potential needs on personalized matching values.

[0075] Personalized matching values ​​are calculated using the following formula:

[0076]

[0077] in, Indicates user For ancillary services At any moment Personalized matching value, Indicates user For ancillary services At any moment Recommended tags, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment Confidence coefficient of potential demand This represents the potential demand over-weighting suppression coefficient, and its value is determined to be greater than or equal to zero based on historical validation samples. This represents the larger of the values ​​obtained by subtracting the core demand matching degree from the potential demand matching degree when taking zero.

[0078] S33: Read personalized matching values, user behavior sets, potential demand evidence weights, auxiliary service status sequences, and historical recommendation records. First, determine whether each user behavior is related to the auxiliary service to be ranked based on the auxiliary service identifier in the user behavior. Then, calculate the potential demand evidence weights related to the auxiliary service to be ranked. At the same time, calculate the degree of fluctuation in auxiliary service status based on the frequency of change of auxiliary services between available and unavailable states within a preset historical period.

[0079] The more relevant evidence of potential demand, the more fully the model understands the matching relationship between the user and the auxiliary service; the higher the volatility of the auxiliary service's state, the less stable the service's feasibility at the time of recommendation. The matching uncertainty is calculated based on these two data points, as shown below:

[0080]

[0081] in, Indicates user For ancillary services At any moment The degree of uncertainty in matching, Indicates the first User behavior at any time Is it related to auxiliary services? Relevant values ​​are taken as 1 if relevant, and 0 if irrelevant. Indicates auxiliary services At any moment The degree of fluctuation in the state, Indicates user At any moment The Weight of evidence of potential demand for a user behavior item This represents the set of user behaviors output by the data preprocessing submodule. The summation symbol indicates that all user behaviors in the set are summed.

[0082] Subsequently, valid auxiliary service comparison samples are extracted from historical recommendation records. The availability of an auxiliary service is determined based on the recommendation result display records, its status at the time of display, and its selection by the user. Auxiliary services that have been displayed, were available at the time of display, and were selected by the user are considered positive services; those that have been displayed, were available at the time of display, but were not selected by the user are considered negative services. Auxiliary services that are not displayed, unavailable, or have an unknown status are not considered negative services.

[0083] The weighting for trusted exposure between positive and negative services is calculated using the following formula:

[0084]

[0085] in, Indicates user Correct service and negative services The resulting comparison samples at time The weighting of credible exposure; and These represent the current service. and negative services Have users been notified? If something is already displayed, return 1; otherwise, return 0. and Each represents the current service. and negative services To users Whether it is available during display; if available, return 1; otherwise, return 0. Indicates user Do you want to select the correct service? Select 1 if you choose, otherwise select 0. Indicates user Choose negative service Select 1 if you choose, otherwise select 0. and Representing users respectively With positive service and negative services The overall credibility of the relevant information.

[0086] The above calculation process generates non-zero comparison weights only when both positive and negative services have been displayed and are available at the time of display. The contribution of the comparison samples is adjusted based on the comprehensive behavioral credibility of the related user behaviors of the two services, thus excluding erroneous negative samples caused by unexposed and unavailable auxiliary services from the data source.

[0087] The conservative lower bound and optimistic upper bound of the auxiliary service matching interval are determined based on the personalized matching value and the degree of matching uncertainty, respectively, as shown in the following calculation:

[0088]

[0089]

[0090] in, Indicates user For ancillary services At any moment The conservative lower bound, user For ancillary services At any moment The optimistic upper limit, Indicates user For ancillary services At any moment Personalized matching value, Indicates user For ancillary services At any moment The degree of uncertainty in matching, This represents the matching uncertainty adjustment coefficient, and its value is greater than or equal to zero.

[0091] The matching uncertainty adjustment coefficient is determined based on the interval coverage of the recommended results in the historical validation samples. Increasing this coefficient expands the matching interval and improves the conservatism of the ranking, while decreasing it narrows the matching interval. The smallest candidate coefficient that makes the proportion of actual selection results falling into the corresponding matching interval reach the preset coverage ratio is selected as the matching uncertainty adjustment coefficient.

[0092] The interval ranking model is trained using conservative lower bound, optimistic upper bound, and reliable exposure comparison weights, as shown below:

[0093]

[0094] in, This represents the ranking loss for reliable exposure intervals. Represents a set of users. Indicates user A comparative set of effective auxiliary services, Indicates user Correct service and negative services The resulting comparison samples at time The weighting of credible exposure comparison Indicates user Correct service At any moment The conservative lower bound, user negative service At any moment The optimistic upper limit, Represents the natural logarithm. This represents the natural exponential function.

[0095] The effective auxiliary service comparison set consists of positive and negative services with non-zero trusted exposure comparison weights obtained from the above calculation. By reducing the trusted exposure interval ranking loss, the core demand matching matrix, potential demand matching matrix, and recommendation model parameters are updated to ensure that the conservative lower bound of positive services is higher than the optimistic upper bound of negative services. After training, the final recommendation value of the auxiliary services is calculated as follows:

[0096]

[0097] in, Indicates user For ancillary services At any moment The final recommended value, Indicates user For ancillary services At any moment A conservative lower bound is used as the final recommendation value, prioritizing auxiliary services with higher personalized matching values ​​and lower matching uncertainty. Auxiliary services marked as "one" that can be recommended in real-time are ranked from highest to lowest according to their final recommendation values, generating personalized recommendation results.

[0098] S4: The feedback update module executes the recommended output, user confirmation, and model feedback updates, specifically including the following steps:

[0099] S41: Read the final recommendation value and real-time recommended tagging According to the final recommended value The system displays auxiliary services from largest to smallest, simultaneously showing the core and potential demand matching criteria for each auxiliary service. After a user selects an auxiliary service, a selection record is generated, containing the user identifier, task identifier, auxiliary service identifier, selection time, and selection result. Once the auxiliary service is completed, the actual usage and service fulfillment results are retrieved from the service order and service fulfillment records.

[0100] The displayed records, selected records, actual usage results, and service performance results are linked according to user identifiers, task identifiers, and auxiliary service identifiers to form recommendation feedback data.

[0101] S42: Update the corresponding fields based on the recommendation feedback data. For auxiliary services that have been selected and actually used, use the selection record and actual usage result as subsequent confirmation records to improve the confirmation of subsequent behaviors of users with the same auxiliary service. For auxiliary services that have been displayed and are available but not selected, combine the corresponding display record, auxiliary service status and selection result into a negative service record, and include it in the effective auxiliary service comparison set during the next model training.

[0102] For auxiliary services that are not displayed, unavailable, or have an unknown status, only the auxiliary service status is updated; negative service records are not generated based on user-unselected results. For core requirements that users actively modify and reconfirm, the modification results are written into the core requirement data, and the core requirement representation is regenerated. Potential requirement representations are only recalculated based on the updated set of user behaviors and are not directly written into the core requirement data.

[0103] After the feedback update is completed, the updated user behavior data, core demand data, auxiliary service status, and recommendation feedback data are re-inputted. Data preprocessing reconstructs the user behavior set according to the original three association keys, so that new feedback enters the next recommendation calculation along the original processing chain of behavior credibility, core demand attribution, potential demand construction, and personalized recommendation.

[0104] The preferred embodiments of this application have been described in detail above with reference to the accompanying drawings. Typical known structures and common knowledge techniques in the preferred embodiments have not been described in detail here. Those skilled in the art can improve and implement the technical solutions of this invention based on the guidance provided in these embodiments and their own capabilities. Some typical known structures, known methods or common knowledge techniques should not be obstacles for those skilled in the art to implement this application.

[0105] The scope of protection claimed in this application shall be determined by the contents of its claims, and the contents described in the invention description, specific embodiments and drawings shall be used to interpret the claims.

[0106] Within the scope of the technical concept of this application, several modifications can be made to the specific implementation of this application, and these modified implementations should also be considered within the protection scope of this application.

Claims

1. A collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking, characterized in that: Includes the following steps: S1: The data construction module performs multi-source data acquisition and preprocessing, specifically including: S11: Obtain user behavior, core needs, auxiliary services, scenarios, and recommendation feedback data; S12: Form a recommendation sample with uniform behavior type, auxiliary service status, time benchmark and numerical range; S2: The requirement representation learning module performs behavior credibility calculation, core requirement attribution, and potential requirement construction; S3: The personalized recommendation module performs behavioral demand matching, core demand protection scoring, and credible exposure range sorting. S4: The feedback update module executes the recommended output, user confirmation, and model feedback updates; Specifically, S3 includes: S31: Encode the service objectives, service limitations, and effective time periods in the normalized core requirement data according to the field order to obtain the core requirement representation; read the service function, applicable conditions, service location, available time period, preparation time, service price, historical performance status, and text description according to the auxiliary service identifier, perform category coding on the category field, normalize the time, price, and performance rate, and perform semantic coding on the text description. Then, concatenate the coding results according to the fixed field order and map them to the same vector dimension as the core requirement representation and potential requirement representation to obtain the auxiliary service representation. The degree of matching between auxiliary services and core requirements, and the degree of matching between auxiliary services and potential requirements are calculated separately, as shown below: ， ， in, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment The core requirement is stated as follows: Indicates user At any moment Potential demand indicates, This represents the core requirement matching matrix. Represents the potential demand matching matrix. Indicates at time The auxiliary services indicate that This represents a logical function that converts the calculation result to a value between zero and one. Indicates vector transpose; S32: Read the core requirement matching degree, potential requirement matching degree, potential requirement confidence coefficient, auxiliary service data, and scenario data. Compare the applicable conditions of auxiliary services with the service restrictions in the core requirements item by item, and compare the available time of auxiliary services with the expected service time of the current recommendation task. When all service restrictions are met and the auxiliary service is available within the expected service time, set the real-time recommendable flag to 1; when any service restriction is not met or the auxiliary service is unavailable, set the real-time recommendable flag to 0. Personalized matching values ​​are calculated using the following formula: ， in, Indicates user For ancillary services At any moment Personalized matching value, Indicates user For ancillary services At any moment Recommended tags, Indicates user Core needs and auxiliary services At any moment The degree of matching of core needs Indicates user Potential needs and support services At any moment The degree of matching of potential needs Indicates user At any moment Confidence coefficient of potential demand This represents the potential demand over-weighting inhibition coefficient; S33: Read personalized matching values, user behavior sets, potential demand evidence weights, auxiliary service status sequences, and historical recommendation records. First, determine whether each user behavior is related to the auxiliary service to be ranked based on the auxiliary service identifier in the user behavior. Then, calculate the potential demand evidence weights related to the auxiliary service to be ranked. At the same time, calculate the degree of fluctuation of the auxiliary service status based on the frequency of change of the auxiliary service between the available and unavailable states within a preset historical period.

2. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 1, characterized in that, S33 further includes: The degree of uncertainty in the matching is calculated as follows: in, Indicates user For ancillary services At any moment The degree of uncertainty in matching, Indicates the first User behavior at any time Is it related to auxiliary services? Relevant values ​​are taken as 1 if relevant and 0 if irrelevant. Indicates auxiliary services At any moment The degree of fluctuation in the state, Indicates user At any moment The Weight of evidence of potential demand for a user behavior item This represents the set of user behaviors output by the data preprocessing submodule.

3. The core and potential demand collaborative recommendation method based on behavioral attribution residuals and credible exposure interval ranking according to claim 2, characterized in that, S33 further includes: Extract valid auxiliary service comparison samples from historical recommendation records, determine whether an auxiliary service has been displayed based on the recommendation result display record, determine whether an auxiliary service is available based on the auxiliary service status associated with the display time, determine whether an auxiliary service has been selected based on the user selection record, and designate auxiliary services that have been displayed, are available at the time of display, and have been selected by the user as positive services, and designate auxiliary services that have been displayed, are available at the time of display, but have not been selected by the user as negative services. The weighting for trusted exposure between positive and negative services is calculated using the following formula: ， in, Indicates user Correct service and negative services The resulting comparison samples at time The weighting of credible exposure comparison; and These represent the current service. and negative services Have users been notified? If something has already been displayed, return 1; otherwise, return 0. and Each represents the current service. and negative services To users Whether it is available during display; if available, return 1; otherwise, return 0. Indicates user Do you want to select the correct service? Select 1 if you choose, otherwise select 0. Indicates user Do you want to choose negative service? Select 1 if you choose, otherwise select 0. and Representing users respectively With positive service and negative services The overall credibility of the relevant information.

4. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 3, characterized in that, S33 further includes: The interval ranking model is trained using conservative lower bound, optimistic upper bound, and reliable exposure comparison weights, as shown in the following formula: ， in, This represents the ranking loss for reliable exposure intervals. Represents a set of users. Indicates user A comparative set of effective auxiliary services, Indicates user Correct service and negative services The resulting comparison samples at time The weighting of credible exposure comparison Indicates user Correct service At any moment The conservative lower bound, user negative service At any moment The optimistic upper limit, Represents the natural logarithm. Represents the natural exponential function; The effective auxiliary service comparison set consists of positive and negative services with non-zero credible exposure comparison weights obtained from the above calculation. By reducing the credible exposure interval ranking loss, the core demand matching matrix, potential demand matching matrix, and recommendation model parameters are updated to ensure that the conservative lower bound of positive services is higher than the optimistic upper bound of negative services. After training, the final recommendation value of the auxiliary services is calculated as follows: ， in, Indicates user For ancillary services At any moment The final recommended value, Indicates user For ancillary services At any moment The conservative lower bound.

5. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 1, characterized in that, S2 specifically includes the following steps: S21: Read user behavior set Regarding the first For each user behavior, the basic credibility of the behavior type is obtained from the behavior type field, the degree of behavior completion is obtained from the behavior duration and operation completion status, the degree of confirmation of subsequent behavior is obtained from subsequent user behaviors with the same auxiliary service identifier, and the degree of consistency of auxiliary service status is obtained from the auxiliary service status at the time of the behavior and the behavior result. S22: Read user behavior set Behavioral credibility The core requirements indicate that for each user behavior, the semantic consistency, object consistency, and temporal consistency between the user behavior and each core requirement are calculated based on the auxiliary service data associated with the user behavior and the time when the behavior occurs. S23: Read user behavior set Behavioral credibility Degree of core demand attribution Based on the scenario data, firstly, the scenario correlation degree, spatial proximity, temporal overlap, and substitution degree are calculated by normalizing all factors to between zero and one, according to the spatial proximity between the service location of the corresponding auxiliary service and the current travel node, the temporal overlap between the available time of the auxiliary service and the expected travel time, and the degree to which the auxiliary service substitutes for the real-time facility status. Then, a weighted sum is performed according to preset non-negative weights to obtain the scenario correlation degree. The behavior credibility is then... As a control metric for the reliability of user behavior, the maximum value of the attribution degree of core needs is taken as the degree to which user behavior has been explained by core needs, and the degree of scenario relevance is taken as the degree of relevance between user behavior and the current recommendation task. The three together form the potential demand evidence weight.

6. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 5, characterized in that... S21 further includes: No. The credibility of a user's behavior is calculated using the following formula: in, Indicates user At any moment The The credibility of user behavior. Indicates the first The basic credibility of the behavior type to which the user behavior belongs. Indicates the first User behavior types Indicates user At any moment The The degree of completion of user behavior. Indicates user At any moment The The degree of confirmation of subsequent user behavior. Indicates user At any moment The The degree of consistency between the status of auxiliary services for user behavior. , and These represent the non-negative weights for the degree of completion of the action, the degree of confirmation of subsequent actions, and the degree of consistency of the status of auxiliary services, respectively.

7. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 6, characterized in that, S22 further includes: No. User behavior can be detected by the first The degree of core requirement attribution in the explanation of core requirements is calculated using the following formula: ， in, Indicates user At any moment The User behavior can be detected by the first The degree of attribution of core requirements explained in the core requirements section Indicates user At any moment The User behavior can be detected by the first The degree of semantic consistency in the explanation of the core requirements of the item. Indicates user At any moment The User behavior can be detected by the first The degree of consistency between the objects explained in the core requirements of the item Indicates user At any moment The User behavior can be detected by the first The degree of temporal consistency in the explanation of the core requirements of the project. , and These represent the non-negative attribution weights for semantic consistency, object consistency, and temporal consistency, respectively.

8. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 7, characterized in that, S23 further includes: No. Weight of evidence of potential needs for user behavior Calculated using the following formula: ， in, Indicates user At any moment The Weight of evidence of potential demand for a user behavior item Indicates user At any moment The The credibility of user behavior. Indicates user At any moment The The degree of contextual relevance of user behaviors. Indicates user The core demand quantity Indicates user At any moment The User behavior can be detected by the first The degree of attribution of core requirements explained in the core requirements section Indicates the first Take the maximum value among all the core needs attribution levels corresponding to each user behavior; The behavioral representations are aggregated using latent demand evidence weights to obtain the latent demand representations: ， in, Indicates user At any moment Potential demand indicates, Indicated by user The The behavior representation is obtained by encoding the behavior type, corresponding auxiliary service function, and behavior result of a user behavior. This represents the set of user behaviors output by the data preprocessing submodule. The summation symbol indicates that all user behaviors in the set are summed. This indicates a preset positive number to prevent the denominator from being zero. Indicates user At any moment The Weight of evidence of potential demand for a user behavior item.

9. The collaborative recommendation method for core and potential needs based on behavioral attribution residuals and credible exposure interval ranking according to claim 8, characterized in that, S23 further includes: The confidence coefficient for potential demand is calculated based on the total amount of evidence of potential demand and the degree of consistency between the direction of the evidence of potential demand, as shown in the following formula: ， in, Indicates user At any moment Confidence coefficient of potential demand This represents the minimum total amount of evidence determined based on the average total amount of evidence from historical valid potential demand samples. Indicates user At any moment The Weight of evidence of potential demand for a user behavior item Indicated by user The The behavior representation is obtained by encoding the behavior type, corresponding auxiliary service function, and behavior result of a user behavior. This represents the set of user behaviors output by the data preprocessing submodule. This indicates a preset positive number to prevent the denominator from being zero. Represents the L2 norm, This means taking the smaller of the two values.

10. The core and potential demand collaborative recommendation method based on behavioral attribution residuals and credible exposure interval ranking according to claim 4, characterized in that, S4 specifically includes the following steps: S41: Based on the final recommended value The auxiliary services are displayed from largest to smallest, and the core and potential demand matching criteria for each auxiliary service are displayed simultaneously. After a user selects an auxiliary service, a selection record is generated and written with the user identifier, task identifier, auxiliary service identifier, selection time, and selection result. After the auxiliary service is completed, the actual usage result and service performance result are obtained from the service order record and service performance record. S42: Update the corresponding fields based on the recommendation feedback data. For auxiliary services that have been selected and actually used, use the selection record and actual usage result as subsequent confirmation records. For auxiliary services that have been displayed and are available but not selected, use the corresponding display record, auxiliary service status and selection result to form a negative service record, and enter it into the effective auxiliary service comparison set during the next model training.

Images