Offline consumption behavior prediction method and system based on user geographic position

Abstract

The invention provides an offline consumption behavior prediction method and system based on a user geographic position, and the method comprises the steps: firstly obtaining a user geographic position data set and a user payment big data set, the former comprises a stay record sequence of a user in different offline regions, and the latter comprises a payment record sequence of the user in different offline consumption scenes; performing dynamic association processing on the two types of data to generate an association mapping relationship, constructing a user offline consumption behavior analysis model based on the association mapping relationship, mining a consumption scene adaptation degree and a payment behavior association degree, performing consumption behavior tendency deduction by combining the two, and generating a preliminary prediction result; and according to the preliminary prediction result, the associated parameters are reversely adjusted and deduced again, and a final offline consumption behavior prediction result is generated, so that the offline consumption behavior of the user can be accurately predicted, and a merchant is assisted to optimize a marketing strategy.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method and system for predicting offline consumption behavior based on user geographic location. Background Technology

[0002] In the business world, accurately predicting offline consumer behavior is crucial for businesses to develop marketing strategies, optimize store layouts, and enhance the consumer experience. However, existing methods for predicting offline consumer behavior have several limitations.

[0003] Some methods rely on a single data source, such as analyzing a user's historical payment records alone. While these methods can obtain information such as past spending amounts and product categories, they cannot understand the specific scenarios and contexts in which the user's spending behavior occurred. For example, a user may make purchases in different geographical locations due to different needs; it is difficult to determine the location and scenario of the user's next purchase based solely on payment records.

[0004] Some methods, while considering geographic location information, simply make a rough correlation between geographic location and consumption without delving into the dynamic relationship between geographic location data and payment data. For example, they merely count the number of times a user makes a purchase in a certain area without analyzing the intrinsic connection between the user's dwell time, frequency of stay, and consumption behavior in that area. This results in inaccurate and incomplete predictions, failing to meet the growing demand for precision marketing from businesses. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for predicting offline consumption behavior based on user geographic location, the method comprising: Acquire user geolocation data set and user payment big data set, wherein the user geolocation data set contains a sequence of user stay records in different offline areas, and the user payment big data set contains a sequence of user payment records in different offline consumption scenarios; The user geographic location data set and the user payment big data set are dynamically correlated to generate a correlation mapping relationship between user geographic location and payment behavior. The correlation mapping relationship is used to represent the time synchronization correspondence between the stay record sequence and the payment record sequence. Based on the aforementioned association mapping relationship, a user offline consumption behavior analysis model is constructed. This consumption behavior analysis model is used to mine the consumption scenario adaptability hidden in the dwell record sequence and the payment behavior correlation hidden in the payment record sequence. By combining the adaptability of the consumption scenario and the correlation of the payment behavior, the user's offline consumption behavior tendency is inferred, and preliminary consumption behavior prediction results are generated. Based on the preliminary consumer behavior prediction results, the correlation parameters in the correlation mapping relationship are adjusted in reverse, and the consumer behavior tendency inference is performed again based on the adjusted correlation mapping relationship to generate the final offline consumer behavior prediction results.

[0006] In another aspect, embodiments of the present invention also provide an offline consumption behavior prediction system based on user geolocation, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0007] Based on the above, this embodiment of the invention acquires user geographic location data sets and user payment big data sets, dynamically correlates these two types of data, generates a correlation mapping relationship between user geographic location and payment behavior, and can capture the time synchronization correspondence between the stay record sequence and the payment record sequence. Based on this correlation mapping relationship, a user offline consumption behavior analysis model is constructed, which can effectively mine the consumption scenario adaptability implicit in the stay record sequence and the payment behavior correlation implicit in the payment record sequence. It analyzes the user's consumption behavior characteristics from different perspectives, combines the consumption scenario adaptability and payment behavior correlation to perform user offline consumption behavior tendency inference, generates preliminary consumption behavior prediction results, adjusts the correlation parameters in the correlation mapping relationship in reverse according to the preliminary prediction results, and performs consumption behavior tendency inference again to generate the final prediction result, which greatly improves the accuracy and reliability of offline consumption behavior prediction and helps merchants better understand user needs. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the offline consumption behavior prediction method based on user geolocation provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the offline consumption behavior prediction system based on user geolocation provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for predicting offline consumption behavior based on user geographic location, as provided in one embodiment of the present invention. The following is a detailed description of this method.

[0011] Step S110: Obtain a user geolocation data set and a user payment big data set. The user geolocation data set contains a sequence of user stay records in different offline areas, and the user payment big data set contains a sequence of user payment records in different offline consumption scenarios.

[0012] In this embodiment, the first step is to acquire user geolocation data sets and user payment big data sets. The core component of the user geolocation data set is a sequence of user stay records in different offline areas, with each sequence corresponding to information related to the user's stay in a specific offline area. The core component of the user payment big data set is a sequence of user payment records in different offline consumption scenarios, with each sequence corresponding to information related to the user's payment completed within a specific offline consumption scenario. Next, to ensure the validity and usability of the acquired data, the following sub-steps are implemented: Step S111: Receive location information generated by the user in different offline areas from the geographic location acquisition terminal, process the location information into a continuous time series, and obtain a residence record sequence of the user in different offline areas. The residence record sequence includes the time information of the user entering the offline area, the time information of leaving the offline area, and the location change trajectory information during the residence period.

[0013] In this embodiment, location information generated in different offline areas is continuously received from the geographic location acquisition terminal. This location information is generated after the geographic location acquisition terminal collects the user's location at preset time intervals. After receiving the location information, it is organized into a continuous time series according to the chronological order. Location information belonging to the same offline area and being temporally consecutive is grouped together to form a user stay record sequence in that offline area. Each stay record sequence explicitly includes the time information when the user enters the offline area (i.e., the time when the user is first captured in the offline area), the time information when the user leaves the offline area (i.e., the time when the user is last captured in the offline area), and the location change trajectory information corresponding to each collection time point during the stay period. This location change trajectory information can reflect the user's movement path during the stay period.

[0014] Step S112: Extract payment information generated by the user after completing payment in different offline consumption scenarios from the payment data management system, arrange the payment information in chronological order to obtain the payment record sequence of the user in different offline consumption scenarios, the payment record sequence includes the consumption scenario type information corresponding to the payment, the product category information corresponding to the payment, and the time information of the payment operation completion.

[0015] In this embodiment, payment information generated after payments are completed in different offline consumption scenarios is extracted from the payment data management system. This payment information is automatically recorded and stored by the payment data management system after the user completes the payment operation. After extracting the payment information, it is arranged according to the chronological order of the payment operation completion time. Payment information belonging to the same offline consumption scenario is grouped together to form a payment record sequence for that offline consumption scenario. Each payment record sequence clearly includes the consumption scenario type information corresponding to the payment (used to distinguish different types of offline consumption scenarios), the product category information corresponding to the payment (used to identify the category of the product purchased in this payment), and the time information of the payment operation completion (i.e., the time when the user completed the payment confirmation).

[0016] Step S113: Perform time dimension calibration on the time information in the stay record sequence and the payment operation completion time information in the payment record sequence, extract the valid stay records in the calibrated stay record sequence whose user stay time exceeds a preset time threshold, form a valid stay record subsequence, and use the valid stay record subsequence as the core component of the user geolocation data set.

[0017] In this embodiment, since the time information in the stay record sequence and the payment operation completion time information in the payment record sequence may originate from different time acquisition devices, resulting in inconsistencies in the time base, it is necessary to perform time dimension calibration on both types of time information. Specifically, using a unified time standard (such as standard time zone time) as the base, the entry time information and exit time information in the stay record sequence, as well as the payment operation completion time information in the payment record sequence, are adjusted to ensure that all time information is under the same time base. After completing the time dimension calibration, the user stay duration corresponding to each stay record in the calibrated stay record sequence is calculated (stay duration = time information of leaving the offline area - time information of entering the offline area), and the calculated stay duration is compared with a preset time threshold. If the stay duration corresponding to a certain stay record exceeds the preset time threshold, the stay record is determined to be a valid stay record; if the stay duration does not exceed the preset time threshold, the stay record is determined to be an invalid stay record and is removed. All valid stay records are organized in chronological order to form a valid stay record subsequence, and this valid stay record subsequence is used as the core component of the user geographic location data set.

[0018] Step S114: Extract the valid payment records corresponding to offline physical consumption scenarios from the calibrated payment record sequence to form a valid payment record subsequence, and use the valid payment record subsequence as the core component of the user payment big data set.

[0019] In this embodiment, after completing the time dimension calibration, each payment record in the calibrated payment record sequence is judged for its scenario type to distinguish whether the consumption scenario corresponding to the payment record is an offline physical consumption scenario. Specifically, by identifying the consumption scenario type information in the payment record, if the scenario corresponding to the consumption scenario type information is a scenario that requires the user to actually go to the site to make the consumption (such as offline shopping malls, supermarkets, restaurants, etc.), then the payment record is determined to be a valid payment record corresponding to an offline physical consumption scenario; if the scenario corresponding to the consumption scenario type information is an online consumption scenario (such as online shopping platforms, online service platforms, etc.), then the payment record is determined to be an invalid payment record and is removed. All valid payment records are organized in chronological order to form a valid payment record subsequence, and this valid payment record subsequence is used as a core component of the user payment big data set.

[0020] Step S120: Perform dynamic correlation processing on the user geographic location data set and the user payment big data set to generate a correlation mapping relationship between user geographic location and payment behavior. The correlation mapping relationship is used to characterize the time synchronization correspondence between the stay record sequence and the payment record sequence.

[0021] In this embodiment, after acquiring the user's geographic location data set and the user's payment big data set, dynamic association processing is required for the two types of data sets to establish a correspondence between user geographic location and payment behavior. The generated association mapping relationship must be able to clearly represent the time synchronization correspondence between the residence record sequence and the payment record sequence. To achieve this dynamic association processing, the following sub-steps are detailed: Step S121: Extract the time interval information of the dwell record sequence from the user's geographic location data set. The time interval information is determined by the time information of the user entering the offline area and the time information of leaving the offline area.

[0022] In this embodiment, the time interval information corresponding to each stay record sequence is extracted from the user's geographic location data set. Since each stay record sequence contains the time information of the user entering the offline area and the time information of the user leaving the offline area, the time information of entering the offline area is taken as the starting point of the time interval, and the time information of leaving the offline area is taken as the ending point of the time interval. The time period from the starting point to the ending point is determined as the time interval information of the stay record sequence, and each stay record sequence corresponds to a unique time interval information.

[0023] Step S122: Extract payment time information from the payment record sequence in the user payment big data set, wherein the payment time information is the time information when the payment operation is completed.

[0024] In this embodiment, the payment time point information corresponding to each payment record sequence is extracted from the user payment big data set. Since each payment record sequence contains the time information of the payment operation completion, the time information of the payment operation completion is directly determined as the payment time point information of the payment record sequence, and each payment record sequence corresponds to a unique payment time point information.

[0025] Step S123: Determine whether each payment time point falls within the time interval information of any stop record sequence. If the payment time point falls within the time interval information of any stop record sequence, then mark the payment record and the stop record as a preliminary association pair.

[0026] In this embodiment, the payment time point information of each payment record sequence is compared and judged one by one with the time interval information of all dwell record sequences. The specific judgment logic is as follows: if a certain payment time point information is between the start point (inclusive) and the end point (inclusive) of the time interval information of a certain dwell record sequence, then it is determined that the payment time point information falls within the time interval information of the dwell record sequence. At this time, the payment record corresponding to the payment record sequence and the dwell record corresponding to the dwell record sequence are marked as a preliminary association pair; if a certain payment time point information does not fall within the time interval information of any dwell record sequence, then the payment record is not marked as a preliminary association pair with any dwell record for the time being.

[0027] Step S124: Perform scenario matching verification on the offline area information corresponding to the stay record and the consumption scenario type information corresponding to the payment record in the preliminary association pair. Determine whether the offline area corresponding to the stay record matches the consumption scenario type corresponding to the payment record. If they match, retain the preliminary association pair; otherwise, remove the preliminary association pair.

[0028] In this embodiment, after obtaining the initial association pairs, it is necessary to perform scene matching verification on each initial association pair to ensure the rationality of the association. This step is further refined into the following sub-steps: Step S1241: Construct a target matching rule base, which stores the mapping relationship between different offline area identifiers and corresponding consumption scenario types, with each offline area identifier corresponding to at least one consumption scenario type.

[0029] In this embodiment, a target matching rule base is first constructed. The core content of this rule base is the mapping relationship between different offline area identifiers and corresponding consumption scenario types. During the construction process, each offline area is first assigned a unique identifier to form an offline area identifier. Then, based on the actual functional positioning of each offline area (e.g., the area is mainly used for commercial retail, catering services, leisure and entertainment, etc.), the consumption scenario type corresponding to each offline area identifier is determined. Each offline area identifier corresponds to at least one consumption scenario type. If an offline area has multiple functions (e.g., a mixed-use commercial complex that includes retail, catering, entertainment, etc.), then that offline area identifier corresponds to multiple consumption scenario types. These mapping relationships are stored one by one in the target matching rule base, forming structured rule data.

[0030] Step S1242: Extract the offline area identifier of the preliminary association record and the consumption scenario type information of the payment record.

[0031] In this embodiment, for each preliminary association pair, the corresponding offline area identifier is extracted from the stay record (this identifier is a unique identifier assigned to the offline area when constructing the target matching rule base), and the corresponding consumption scenario type information is extracted from the payment record (this information is used to identify the consumption scenario category to which this payment belongs), ensuring that the corresponding offline area identifier and consumption scenario type information can be extracted for each preliminary association pair.

[0032] Step S1243: Input the offline area identifier into the target matching rule base and query all consumption scenario types corresponding to the offline area identifier.

[0033] In this embodiment, the offline area identifier extracted from the initial association pair is used as the query keyword and input into the constructed target matching rule base. Through the query function of the rule base, all corresponding consumption scenario types of the offline area identifier stored in the rule base are retrieved. The query results are returned in the form of a list, which includes all consumption scenario types corresponding to the offline area identifier.

[0034] Step S1244: Determine whether the consumption scenario type information of the payment record belongs to one of the various consumption scenario types obtained in the query. If it does, determine that the offline area corresponding to the stay record matches the consumption scenario type corresponding to the payment record, and retain the preliminary association pair.

[0035] In this embodiment, the consumption scenario type information of the payment record extracted from the preliminary association pair is compared with all consumption scenario types corresponding to the offline area identifier obtained in step S1243. If the consumption scenario type information of the payment record is completely consistent with a certain consumption scenario type in the query result, it is determined that the offline area corresponding to the stay record in the preliminary association pair matches the consumption scenario type corresponding to the payment record. At this time, the preliminary association pair is retained and proceeds to the subsequent processing flow.

[0036] Step S1245: If the consumption scenario type information of the payment record does not belong to any of the queried consumption scenario types, then further analyze the auxiliary consumption scenario types around the offline area corresponding to the stay record to determine whether the consumption scenario type information of the payment record belongs to the auxiliary consumption scenario type.

[0037] In this embodiment, if the judgment result of step S1244 is that the consumption scenario type information of the payment record does not belong to any of the queried consumption scenario types, it is necessary to further analyze the ancillary consumption scenario types around the offline area corresponding to the stay record. Specifically, by obtaining the functional information of other offline areas within a preset range around the offline area, the consumption scenario types corresponding to these surrounding areas are determined, and these consumption scenario types are defined as ancillary consumption scenario types of the offline area. Then, the consumption scenario type information of the payment record is compared with these ancillary consumption scenario types to determine whether it belongs to an ancillary consumption scenario type.

[0038] Step S1246: If it belongs to the auxiliary consumption scenario type, then supplement and update the target matching rule library, establish a mapping relationship between the auxiliary consumption scenario type and the offline area identifier, and retain the preliminary association pair; if it does not belong to the auxiliary consumption scenario type, then determine that it does not match and remove the preliminary association pair.

[0039] In this embodiment, if the judgment result of step S1245 is that the consumption scenario type information of the payment record belongs to the auxiliary consumption scenario type, then the target matching rule base is supplemented and updated, the auxiliary consumption scenario type is added to the consumption scenario type list corresponding to the offline area identifier, a new mapping relationship between the auxiliary consumption scenario type and the offline area identifier is established, and the update result of the rule base is saved, while the preliminary association pair is retained; if the judgment result of step S1245 is that the consumption scenario type information of the payment record does not belong to the auxiliary consumption scenario type, then it is determined that the offline area corresponding to the stay record in the preliminary association pair does not match the consumption scenario type corresponding to the payment record, and at this time the preliminary association pair is removed from the association list and no longer enters the subsequent processing flow.

[0040] Step S125: Arrange the retained preliminary association pairs in chronological order, construct an association mapping table containing stay record identifiers, payment record identifiers, time correspondences and scene matching results, and use the association mapping table as the association mapping relationship between user geographical location and payment behavior.

[0041] In this embodiment, after completing the scene matching verification of all preliminary association pairs and retaining valid preliminary association pairs, these preliminary association pairs are sorted according to the order of the entry time information of the stay records (or the payment time point information of the payment records) corresponding to all retained preliminary association pairs. Next, a unique identifier (i.e., stay record identifier and payment record identifier) ​​is assigned to each stay record and each payment record, and then an association mapping table is constructed. Each entry in this association mapping table includes a stay record identifier (used to associate the corresponding stay record), a payment record identifier (used to associate the corresponding payment record), a time correspondence (recording the specific correspondence between the payment time point information and the stay time interval information, such as the specific location of the payment time point within the stay time interval), and a scene matching result (recording whether the association pair was retained through direct matching or auxiliary scene matching). The constructed association mapping table serves as the association mapping relationship between user geographical location and payment behavior, used for subsequent model construction and analysis.

[0042] Step S130: Construct a user offline consumption behavior analysis model based on the aforementioned association mapping relationship, and use this consumption behavior analysis model to mine the consumption scenario adaptability implicit in the dwell record sequence and the payment behavior correlation implicit in the payment record sequence.

[0043] In this embodiment, based on the association mapping relationship generated in step S120, a user offline consumption behavior analysis model is constructed to analyze user offline consumption behavior. This model is used to mine hidden key indicators from the dwell time record sequence and payment record sequence, namely, consumption scenario suitability and payment behavior correlation. To achieve model construction and indicator mining, the following sub-steps are detailed: Step S131: Using the stay record identifier in the association mapping relationship as an index, associate the location change trajectory information and offline area information in the corresponding stay record sequence to construct a stay feature dataset.

[0044] In this embodiment, the stay record identifier in the association mapping relationship is used as an index keyword to search the stay record sequence in the user's geographic location data set, find the stay record sequence corresponding to the stay record identifier, and extract the location change trajectory information and offline area information from the stay record sequence. The location change trajectory information and offline area information corresponding to each stay record identifier are treated as a set of feature data. All sets of feature data are arranged in the order of the stay record identifiers to form a structured stay feature dataset. Each data entry in the stay feature dataset contains a stay record identifier, the corresponding location change trajectory information, and offline area information.

[0045] Step S132: Using the payment record identifier in the association mapping relationship as an index, associate the consumption scenario type information and product category information in the corresponding payment record sequence to construct a payment feature dataset.

[0046] In this embodiment, the payment record identifier in the association mapping relationship is used as an index keyword to search the payment record sequence in the user payment big data set, find the payment record sequence corresponding to the payment record identifier, and extract the consumption scenario type information and product category information from the payment record sequence. The consumption scenario type information and product category information corresponding to each payment record identifier are treated as a set of feature data. All sets of feature data are arranged in the order of payment record identifiers to form a structured payment feature dataset. Each data entry in the payment feature dataset contains a payment record identifier, the corresponding consumption scenario type information, and product category information.

[0047] Step S133: Input the dwell feature dataset and the payment feature dataset into the model building module to initialize the basic framework of the user offline consumption behavior analysis model. The basic framework includes a dwell feature processing layer, a payment feature processing layer and a feature association layer.

[0048] In this embodiment, the dwell time feature dataset constructed in step S131 and the payment feature dataset constructed in step S132 are jointly input into the model building module. Upon receiving the two types of feature datasets, the model building module first initializes the basic framework of the user offline consumption behavior analysis model. This basic framework includes three core layers: the dwell time feature processing layer, the payment feature processing layer, and the feature association layer. Specifically, the dwell time feature processing layer is dedicated to processing various types of information in the dwell time feature dataset and extracting key features related to dwell time; the payment feature processing layer is dedicated to processing various types of information in the payment feature dataset and extracting key features related to payment; and the feature association layer is used to establish the association between the output results of the dwell time feature processing layer and the payment feature processing layer.

[0049] Step S134: In the dwell feature processing layer, path analysis is performed on the location change trajectory information in the dwell feature dataset to extract the complexity features of the user's movement path during the dwell period. Combined with the scene attributes corresponding to the offline area information, after standardization processing, the user's consumption scene adaptability in the offline area is calculated. The consumption scene adaptability is used to reflect the degree of matching between the user's dwell area and potential consumption needs.

[0050] In this embodiment, within the dwell feature processing layer, path analysis is first performed on the position change trajectory information of each data entry in the dwell feature dataset. This analysis process is detailed into the following sub-steps: Step S1341: Extract trajectory points from the location change trajectory information in the dwell feature dataset to obtain the coordinates of multiple continuous trajectory points of the user during the dwell period.

[0051] In this embodiment, trajectory point extraction is performed on each location change trajectory information in the dwell feature dataset. Based on the time interval of location collection, the location coordinates corresponding to each collection time point are extracted from the location change trajectory information. These location coordinates are arranged in chronological order of collection time, forming multiple continuous trajectory point coordinates of the user during the dwell period. These continuous trajectory point coordinates can completely reflect the user's location movement in the dwell area from entry to exit. For example, during the dwell period in a certain offline area, the coordinate sequence corresponding to the entire path from near the area entrance to the front of different merchants, and then to near the area exit.

[0052] Step S1342: Calculate the distance between the coordinates of adjacent trajectory points, sum the distances of all adjacent trajectory points as the total length of the trajectory, and count the number of trajectory points as the total number of trajectory points.

[0053] In this embodiment, for the multiple continuous trajectory point coordinates obtained in step S1341, adjacent trajectory point coordinates are selected sequentially according to time sequence. A preset distance calculation method (such as a distance calculation method based on a Cartesian coordinate system) is used to calculate the distance between each pair of adjacent trajectory point coordinates. After calculating the distances of all adjacent trajectory points, these distance values ​​are summed, and the sum is the total trajectory length corresponding to the location change trajectory information. Simultaneously, the number of extracted trajectory point coordinates is counted, and the resulting number is the total number of trajectory points corresponding to the location change trajectory information. The total trajectory length and the total number of trajectory points can preliminarily reflect the user's movement range and the density of location data collection during the dwell period.

[0054] Step S1343: Calculate the average step length of the trajectory based on the total length of the trajectory and the total number of trajectory points. At the same time, analyze the degree of clustering of trajectory points in spatial distribution, calculate the dispersion coefficient of trajectory point distribution, and use the average step length and dispersion coefficient of the trajectory as the core parameters of the movement path complexity feature.

[0055] In this embodiment, the average step length of the trajectory is first calculated. The calculation logic is: Average step length of trajectory = Total length of trajectory ÷ Total number of trajectory points. The average step length of the trajectory obtained by this calculation reflects the average distance moved by the user between two adjacent collection time points during the dwell period. Next, the degree of clustering of trajectory points in spatial distribution is analyzed, specifically by calculating the coefficient of variation of the trajectory point distribution. When calculating the coefficient of variation, the center position coordinates of all trajectory point coordinates in space are first determined (e.g., by calculating the average of the x-coordinates and y-coordinates of all trajectory points to form the center position coordinates). Then, the distance between each trajectory point coordinate and the center position coordinates is calculated, resulting in multiple distance values. Then, the standard deviation of these distance values ​​is calculated, and the average of these distance values ​​is also calculated. Finally, the coefficient of variation of the trajectory point distribution is obtained through the calculation logic of coefficient of variation = standard deviation ÷ average. The larger the coefficient of variation, the more dispersed the trajectory points are in space, and the lower the degree of clustering. The smaller the coefficient of variation, the more concentrated the trajectory points are in space, and the higher the degree of clustering. The calculated average step length of the trajectory and the coefficient of variation are used together as the core parameters of the movement path complexity feature. These two parameters reflect the complexity of the user's movement path from different dimensions.

[0056] Step S1344: Extract the offline area information corresponding to the stay records in the stay feature dataset, and determine the scene attributes of the offline area. The scene attributes include the area function type, the density of merchants in the area, and the main service target group of the area.

[0057] In this embodiment, content related to scene attributes is extracted from the offline area information of each data entry in the dwell feature dataset to determine the scene attributes of the offline area. Specifically, the area function type is determined based on the planned use and actual operation of the offline area, such as the area being specifically for providing catering services, specifically for selling clothing, or combining shopping and leisure functions; the merchant density within the area is calculated by statistically analyzing the number of actually operating merchants in the offline area and combining it with the physical area of ​​the offline area (e.g., number of merchants ÷ area), reflecting the density of merchants within the area; and the main target customer group of the area is determined based on factors such as the target customer positioning of the merchants in the offline area and the surrounding population structure, such as primarily serving young people, families, or business people. The area function type, merchant density within the area, and the main target customer group of the area are collectively considered as the scene attributes of the offline area.

[0058] Step S1345: Standardize the core parameters of the mobile path complexity feature and the scene attributes of the offline area to obtain standardized parameters.

[0059] In this embodiment, since the core parameters of the movement path complexity characteristics (average trajectory step length, coefficient of variation) and the scene attributes of the offline area (merchant density within the area, etc.) may have different dimensions and numerical ranges, direct use in calculations would lead to result deviations. Therefore, these parameters need to be standardized. The standardization process adopts a preset linear standardization method. The specific logic is as follows: for each parameter to be standardized, first determine the maximum and minimum values ​​of the parameter in the dataset corresponding to all stay records; then, through the calculation logic of standardized parameter = (original parameter value - minimum value) ÷ (maximum value - minimum value), transform each original parameter value so that the standardized parameter values ​​are all within the range of 0 to 1. For example, for the average trajectory step length, first find the maximum and minimum values ​​of the average trajectory step length in all stay records, then substitute the original value of the average trajectory step length of a certain stay record into the above formula to obtain the standardized parameter of the average trajectory step length; for the merchant density within the area, the same standardization process is performed to ensure that all parameters involved in subsequent calculations are within a uniform dimension and numerical range.

[0060] Step S1346: Construct a consumption scenario adaptation calculation model, using the standardized parameters as input parameters, and assign corresponding calculation weights to each input parameter.

[0061] In this embodiment, a computational model specifically designed for calculating the adaptability of consumption scenarios is constructed. The input parameters of this model are all the standardized parameters obtained in step S1345, including the standardized average trajectory step size, the standardized coefficient of variation, and the standardized merchant density within the region. After determining the input parameters, a corresponding computational weight is assigned to each parameter. The weight allocation is determined based on the degree of influence of each parameter on the adaptability of the consumption scenario. For example, if the region's functional type has the highest correlation with the user's potential consumption needs, then the standardized parameters related to the region's functional type (such as standardized parameters converted through encoding) are assigned higher weights; if the average trajectory step size has a relatively small impact on the adaptability of the consumption scenario, then lower weights are assigned. After the weight allocation is completed, the weight values ​​of each input parameter are stored in the parameter configuration module of the consumption scenario adaptability computational model for subsequent calculations.

[0062] Step S1347: The input parameters are weighted and calculated using the consumption scenario adaptation calculation model to generate the consumption scenario adaptation degree of the user in the offline area. The numerical range of the consumption scenario adaptation degree is positively correlated with the complexity of the movement path and the degree of matching of scenario attributes.

[0063] In this embodiment, the standardized input parameters obtained in step S1345 and the corresponding calculation weights assigned in step S1346 are input into the consumption scenario adaptation calculation model. The model, following a preset weighted operation logic, first multiplies each input parameter by its corresponding calculation weight to obtain a weighted value for each parameter; then, it sums the weighted values ​​of all parameters, and the sum is the user's consumption scenario adaptation in that offline area. The numerical range of the consumption scenario adaptation is positively correlated with the complexity of the movement path and the degree of matching of scenario attributes. That is, the higher the complexity of the movement path (indicating a greater likelihood that the user will actively explore merchants within the area) and the higher the degree of matching between the scenario attributes and the user's potential consumption needs, the larger the value of the consumption scenario adaptation; conversely, the smaller the value.

[0064] Step S135: In the payment feature processing layer, the consumption scenario type information and product category information in the payment feature dataset are analyzed for correlation. The user's product selection preference features in different consumption scenarios are extracted, and the correlation degree between the user's payment behavior in the current payment scenario and the historical payment scenario is calculated. The correlation degree of payment behavior is used to reflect the similarity between the user's current payment behavior and the historical payment behavior.

[0065] In this embodiment, within the payment feature processing layer, the consumption scenario type information and product category information of each data entry in the payment feature dataset are first analyzed for correlation. Specifically, all payment records of the same user are categorized according to consumption scenario type information, and the frequency of different product categories purchased by the user under each consumption scenario type is counted. For example, under the catering consumption scenario type, the number of times the user purchases staple food, beverages, snacks, and other product categories is counted. Based on the frequency distribution obtained from the statistics, the user's product selection preference features under different consumption scenarios are extracted. For example, if a user purchases beverages most frequently under the catering consumption scenario, then the user's product selection preference feature under the catering scenario shows a tendency to purchase beverages. Next, the correlation of payment behavior is calculated. The calculation process is as follows: First, the consumption scenario type and product category information corresponding to the current payment scenario are determined. Then, historical payment scenarios with the same or similar consumption scenario type information as the current payment scenario are selected from historical payment records. Next, the product category information of the current payment scenario is compared with the product category information of the selected historical payment scenarios, and the proportion of the number of identical or similar product categories in the two is calculated to the total number of product categories in the current payment scenario. At the same time, the correlation factors between the current payment scenario and the historical payment scenario are analyzed in terms of payment time interval (such as the difference between the current payment time and the historical payment time) and payment amount range (such as the degree of overlap between the current payment amount range and the historical payment amount range), and a corresponding weight is assigned to each correlation factor. Finally, the product category similarity ratio and the weighted value of each correlation factor are integrated by weighted summation to obtain the correlation of the user's payment behavior in the current payment scenario and the historical payment scenario. The larger the value of the correlation of payment behavior, the higher the similarity between the user's current payment behavior and the historical payment behavior; the smaller the value, the lower the similarity.

[0066] Step S136: In the feature association layer, establish association calculation rules for the degree of adaptation of consumption scenarios and the degree of correlation of payment behavior.

[0067] In this embodiment, within the feature association layer, historical data samples of the consumption scenario fit calculated in step S134 and the payment behavior correlation calculated in step S135 are first collected. Each sample contains a set of consumption scenario fit values, a corresponding payment behavior correlation value, and a result identifier indicating whether the actual consumption behavior corresponding to that set of values ​​has occurred (marked as valid if it has occurred, and marked as invalid if it has not occurred). Based on these historical data samples, statistical analysis methods (such as correlation analysis) are used to determine the inherent correlation between consumption scenario fit and payment behavior correlation. For example, analysis reveals that when both consumption scenario fit and payment behavior correlation exceed a certain threshold, the probability of actual consumption behavior occurring significantly increases. Based on the determined correlation rules, association calculation rules for consumption scenario fit and payment behavior correlation are established. For example, the rule content includes "when the product of the consumption scenario fit value and the payment behavior correlation value is greater than a preset product threshold, the association is determined to be valid and can be used for subsequent consumption behavior tendency inference; otherwise, the association is determined to be invalid, and the feature extraction process needs to be re-examined." The established association calculation rules are stored in the rule base of the feature association layer.

[0068] Step S140: Combine the consumption scenario adaptability and the payment behavior correlation to perform user offline consumption behavior tendency inference and generate preliminary consumption behavior prediction results.

[0069] In this embodiment, after obtaining the consumption scenario suitability and payment behavior correlation, and establishing the relationship between the two through a feature correlation layer, the user's offline consumption behavior tendency is inferred by combining these two indicators to generate preliminary consumption behavior prediction results. This step is further broken down into the following sub-steps: Step S141: Input the consumption scenario adaptability into the scenario weight allocation module of the user offline consumption behavior analysis model. The scenario weight allocation module assigns scenario influence weights to the consumption scenarios corresponding to different offline regions based on the value of the consumption scenario adaptability.

[0070] In this embodiment, the consumption scenario suitability calculated in step S134 is input into the scenario weight allocation module of the user offline consumption behavior analysis model. The scenario weight allocation module has a pre-set weight allocation algorithm that determines the corresponding scenario influence weight based on the value of the consumption scenario suitability. Specifically, the value range of the consumption scenario suitability is first divided into multiple intervals, each interval corresponding to a pre-set scenario influence weight value. For example, when the consumption scenario suitability value is in the 0.8-1.0 interval, the corresponding scenario influence weight is 0.9; when it is in the 0.6-0.8 interval, the corresponding scenario influence weight is 0.7; when it is in the 0.4-0.6 interval, the corresponding scenario influence weight is 0.5; and when it is in the 0-0.4 interval, the corresponding scenario influence weight is 0.3. After receiving the consumption scenario suitability value, the scenario weight allocation module first determines the interval to which the value belongs, and then, based on the correspondence between the interval and the weight, assigns a scenario influence weight to the consumption scenario of the offline area corresponding to the consumption scenario suitability. The higher the value of the scene influence weight, the greater the influence of the offline consumption scene on users' consumption behavior tendencies.

[0071] Step S142: Input the payment behavior correlation degree into the payment weight allocation module of the user offline consumption behavior analysis model. The payment weight allocation module assigns payment influence weights to the consumption behaviors corresponding to different payment records according to the magnitude of the payment behavior correlation degree.

[0072] In this embodiment, the payment behavior correlation calculated in step S135 is input into the payment weight allocation module of the user's offline consumption behavior analysis model. The payment weight allocation module uses a similar weight allocation logic to the scenario weight allocation module. It first divides the numerical range of the payment behavior correlation into multiple intervals, each interval corresponding to a preset payment influence weight value. For example, when the payment behavior correlation value is in the 0.8-1.0 interval, the corresponding payment influence weight is 0.85; when it is in the 0.6-0.8 interval, the corresponding payment influence weight is 0.65; when it is in the 0.4-0.6 interval, the corresponding payment influence weight is 0.45; and when it is in the 0-0.4 interval, the corresponding payment influence weight is 0.25. After receiving the payment behavior correlation value, the payment weight allocation module determines the interval to which the value belongs, and then, based on the correspondence between the interval and the weight, assigns a payment influence weight to the consumption behavior of the payment record corresponding to that payment behavior correlation. The larger the payment influence weight value, the higher the reference value of the consumption behavior for the user's current consumption behavior tendency.

[0073] Step S143: Call the behavior tendency inference module of the user's offline consumption behavior analysis model. This behavior tendency inference module takes the scene influence weight and payment influence weight as inputs, and combines the time synchronization correspondence in the association mapping relationship to construct the consumption behavior tendency evaluation function.

[0074] In this embodiment, the behavioral tendency inference module of the user's offline consumption behavior analysis model is invoked. This step is further refined into the following sub-steps: Step S1431: In the behavior tendency inference module, create an evaluation parameter input interface, receive the scenario impact weight and payment impact weight through the evaluation parameter input interface, and read the time synchronization correspondence in the association mapping relationship.

[0075] In this embodiment, an evaluation parameter input interface is created within the behavior tendency inference module. This parameter input interface supports receiving two types of parameters simultaneously: scenario influence weight and payment influence weight. Furthermore, it can classify and store the input parameters according to their identifiers (such as "scenario weight" and "payment weight"). While receiving parameters, the behavior tendency inference module reads the time synchronization correspondence in the association mapping relationship generated in step S120 through a data interface. This time synchronization correspondence contains details of the correspondence between payment time point information and dwell time interval information in each association pair, such as whether the payment time point is in the first half, second half, or middle of the dwell time interval.

[0076] Step S1432: Based on the time synchronization correspondence, determine the correlation strength between the time interval corresponding to the influence weight of each scenario and the time point corresponding to the influence weight of each payment, and generate the time correlation coefficient.

[0077] In this embodiment, based on the read time synchronization correspondence, the correlation strength between the dwell time interval corresponding to each scenario influence weight and the payment time point corresponding to each payment influence weight is analyzed. Specifically, the judgment logic is as follows: if the payment time point is in the middle segment of the dwell time interval (e.g., the dwell time interval is T1-T2, and the middle segment is defined as the time period of (T1+T2) / 2±(T2-T1) / 4), it indicates that the user completed the payment during the core period of the dwell time, resulting in the highest correlation strength, and the corresponding time correlation coefficient is assigned the maximum value; if the payment time point is in the first half (not the middle segment) or the second half (not the middle segment) of the dwell time interval, the correlation strength is second, and the time correlation coefficient is assigned the middle value; if the payment time point is near the start point (e.g., the time difference from T1 is less than a preset short-time threshold) or near the end point (e.g., the time difference from T2 is less than a preset short-time threshold), it indicates that the user may have completed the payment when just entering or about to leave the area, resulting in a lower correlation strength, and the time correlation coefficient is assigned the minimum value. Through the above method, a corresponding time correlation coefficient is generated for each combination of scenario influence weight and payment influence weight.

[0078] Step S1433: Multiply the scene impact weight with the corresponding time correlation coefficient to obtain the time-calibrated scene weight; multiply the payment impact weight with the corresponding time correlation coefficient to obtain the time-calibrated payment weight.

[0079] In this embodiment, for each scenario impact weight, it is multiplied by the corresponding time correlation coefficient. The result is the time-calibrated scenario weight. This calibration process can correct the deviation caused by the different correlation strengths between the payment time point and the dwell time interval. For example, if the original scenario impact weight is high, but the corresponding payment time point is near the beginning of the dwell time interval, the time correlation coefficient is low. The calibrated scenario weight will be lower accordingly, better reflecting the actual correlation. Similarly, for each payment impact weight, it is multiplied by the corresponding time correlation coefficient to obtain the time-calibrated payment weight, thus achieving time dimension calibration of the payment impact weight.

[0080] Step S1434: Analyze users' historical offline consumption behavior data, extract users' actual consumption behavior records under different combinations of scenario weights and payment weights, and establish a sample library of correspondence between consumption behavior and weight combinations.

[0081] In this embodiment, historical offline consumption behavior data of users over a period of time is acquired. This data includes records of users' stays in different offline areas, corresponding payment records, and the actual types of consumption behavior (such as purchasing a certain type of goods or using a certain service). Scene weights and payment weights in the historical data are back-calculated to obtain combinations of scene weights and payment weights at different historical moments. Then, each combination of scene weights and payment weights is associated with the corresponding actual consumption behavior record. For example, a weight combination (scene weight 0.7, payment weight 0.6) corresponds to the actual consumption behavior record "purchasing beverages in a catering setting". All associated records are organized and archived to establish a sample library of correspondences between consumption behavior and weight combinations. Each sample in the sample library contains a weight combination (scene weight, payment weight) and the corresponding actual consumption behavior type.

[0082] Step S1435: Based on the corresponding sample library, a consumer behavior tendency assessment function is constructed using statistical analysis methods. This consumer behavior tendency assessment function uses the time-calibrated scenario weight and the time-calibrated payment weight as independent variables, and the probability assessment score of potential consumer behavior as the dependent variable.

[0083] In this embodiment, a consumer behavior propensity assessment function is constructed based on sample data from a correspondence database using statistical analysis methods (such as multiple linear regression analysis). First, the time-calibrated scenario weights and time-calibrated payment weights from the correspondence database are used as independent variables, and the probability of actual consumption behavior occurring in the samples (obtained by statistically analyzing the proportion of actual consumption behavior occurrences under this weight combination to the total number of records) is used as the dependent variable, establishing a regression relationship between the variables. During the construction process, the coefficient parameters in the function are determined through iterative calculation to minimize the deviation between the probability assessment score output by the function and the probability of actual consumption behavior occurring in the samples. For example, if coefficient parameters a and b, and a constant term c are obtained through regression analysis, the consumer behavior propensity assessment function can be expressed as: Probability Assessment Score = a × Time-calibrated Scenario Weight + b × Time-calibrated Payment Weight + c. This consumer behavior propensity assessment function uses the time-calibrated scenario weights and time-calibrated payment weights as independent variables; inputting different weight values ​​will output the corresponding probability assessment score for potential consumption behavior, achieving a quantitative assessment of the probability of consumption behavior occurring.

[0084] Step S1436: Verify the consumer behavior propensity assessment function by inputting the known scenario weights and payment weights, comparing the assessment scores output by the consumer behavior propensity assessment function with the actual occurrence of consumer behavior, and adjusting the coefficient parameters in the consumer behavior propensity assessment function.

[0085] In this embodiment, some sample data that did not participate in the function construction are extracted from the corresponding relationship sample library as verification samples. These verification samples contain known scenario weights, payment weights, and corresponding actual consumption behavior occurrences. The scenario weights and payment weights in the verification samples are input into the constructed consumption behavior tendency evaluation function to obtain the probability evaluation score output by the function. Then, the evaluation score is compared with the actual consumption behavior occurrences in the verification samples. For example, if the evaluation score is higher than a preset verification threshold, the proportion of actual consumption behavior should reach a preset proportion standard; if the evaluation score is lower than the preset verification threshold, the proportion of actual consumption behavior not occurring should reach a preset proportion standard. If the comparison result shows that the deviation between the evaluation score and the actual situation exceeds a preset deviation threshold, the coefficient parameters in the consumption behavior tendency evaluation function are adjusted (such as adjusting the values ​​of a, b, and c in step S1435), the evaluation score is recalculated, and the comparison is repeated until the deviation is less than the preset deviation threshold, ensuring that the consumption behavior tendency evaluation function has high accuracy and reliability.

[0086] Step S144: The potential consumption behavior of users in different offline areas is assessed using the consumption behavior tendency assessment function, and an assessment score is generated for each potential consumption behavior.

[0087] In this embodiment, the time-calibrated scene weights corresponding to the user's current stay records in different offline areas, and the time-calibrated payment weights corresponding to the current payment records, are collected. These weight values ​​are then input into a validated consumer behavior tendency assessment function. For each offline area, corresponding to a potential consumer behavior type (such as the types of behavior that may occur in that area, such as purchasing goods or using services), the function calculates and outputs a corresponding probability assessment score based on the input weight values. For example, in a certain offline catering area, potential consumer behavior types include "purchasing staple food," "purchasing drinks," and "purchasing snacks." After inputting the time-calibrated scene weights of that area and the time-calibrated payment weights of the corresponding payment records into the function, probability assessment scores corresponding to "purchasing staple food," "purchasing drinks," and "purchasing snacks" can be generated respectively. Each potential consumer behavior type corresponds to a unique assessment score, which directly reflects the probability of the potential consumer behavior occurring.

[0088] Step S145: Sort the evaluation scores in descending order, extract the top K potential consumer behaviors and their corresponding evaluation scores, and form a preliminary consumer behavior prediction result that includes the type of consumer behavior, the corresponding offline area, and the evaluation score.

[0089] In this embodiment, after obtaining the evaluation scores for all potential consumption behaviors, the behaviors are sorted in descending order of score. After sorting, the top K potential consumption behaviors are extracted based on a preset extraction quantity K (K being the pre-defined number of high-probability consumption behaviors to be retained). Simultaneously, the offline region information (i.e., the offline region where the potential consumption behavior might occur) and the corresponding evaluation score are linked together for each extracted potential consumption behavior. This information is integrated into a structured data format, where each data entry contains three core fields: "Consumption Behavior Type - Corresponding Offline Region - Evaluation Score". All integrated data is then processed to form a preliminary consumption behavior prediction result. This preliminary prediction result clearly presents the K most likely consumption behaviors that users will engage in in different offline regions and their quantitative indicators of probability.

[0090] Step S150: Based on the preliminary consumption behavior prediction results, the correlation parameters in the correlation mapping relationship are adjusted in reverse, and the consumption behavior tendency inference is performed again based on the adjusted correlation mapping relationship to generate the final offline consumption behavior prediction results.

[0091] In this embodiment, to further improve the accuracy of consumer behavior prediction, it is necessary to reverse-adjust the correlation parameters in the correlation mapping relationship based on the preliminary consumer behavior prediction results, and then re-derive the final prediction result based on the adjusted correlation mapping relationship. This step is further detailed into the following sub-steps: Step S151: Analyze the evaluation scores in the preliminary consumer behavior prediction results and extract low-confidence consumer behavior prediction items whose evaluation scores are lower than the preset score threshold.

[0092] In this embodiment, a preset score threshold is first set. This threshold is determined based on the output range of the consumer behavior tendency assessment function and actual application requirements, and is used to distinguish between high-confidence and low-confidence consumer behavior prediction items. The assessment score corresponding to each prediction item in the preliminary consumer behavior prediction results is analyzed and compared with the preset score threshold. If the assessment score of a prediction item is lower than the preset score threshold, it is determined to be a low-confidence consumer behavior prediction item, extracted from the preliminary prediction results, and stored separately. If the assessment score is higher than or equal to the preset score threshold, it is determined to be a high-confidence consumer behavior prediction item, not processed temporarily, and retained in the preliminary prediction results. This step can filter out prediction items with lower accuracy in the preliminary prediction results.

[0093] Step S152: Analyze the associated mapping relationship entries corresponding to the low-reliability consumption behavior prediction items, and determine the association parameters of the stay records and payment records corresponding to the associated mapping relationship entries. The association parameters include time matching accuracy parameters and scene matching degree parameters.

[0094] In this embodiment, for each low-reliability consumer behavior prediction item, the associated mapping relationship entries it relied on during the initial prediction process are traced (i.e., based on which set of stay records and payment records the prediction item was calculated). After finding the corresponding associated mapping relationship entries, association parameters are extracted from the entry information. These association parameters include time matching accuracy parameters and scenario matching degree parameters. The time matching accuracy parameter characterizes the accuracy of the match between the time interval information of the stay record and the payment time point information of the payment record. For example, a higher parameter value indicates that the payment time point is closer to the core time period within the stay time interval, and the time matching is more accurate. The scenario matching degree parameter characterizes the degree of match between the offline area corresponding to the stay record and the consumption scenario type corresponding to the payment record. A higher scenario matching degree parameter indicates a more reasonable scenario match. By analyzing the associated parameters corresponding to low-reliability prediction items, parameter links in the associated mapping relationship that may have deviations can be located.

[0095] Step S153: Based on the number of low-confidence consumer behavior prediction items and the corresponding deviation of the associated parameters, formulate an association parameter adjustment strategy. The association parameter adjustment strategy includes the correction range of the time matching accuracy parameter and the correction range of the scene matching degree parameter.

[0096] In this embodiment, the total number of low-reliability consumer behavior prediction items is first counted, along with the number of items with low time matching accuracy parameters and low scene matching accuracy parameters. The distribution of these two types of parameter deviations in the low-reliability prediction items is then analyzed. Next, for each low-reliability prediction item, the deviation degree of its associated parameter from the preset standard parameter value is calculated (deviation degree = |actual parameter value - preset standard parameter value| ÷ preset standard parameter value). The larger the deviation degree, the greater the gap between the parameter and the ideal state. Taking into account both the distribution of the number of low-reliability prediction items and the deviation degree of the associated parameters, an adjustment strategy for the associated parameters is formulated. For example, if more than 70% of the low-reliability prediction items are due to low time matching accuracy parameters, and the average deviation degree of these parameters reaches 30%, then the adjustment strategy sets the correction magnitude for the time matching accuracy parameter to be increased by 20%; if the proportion of prediction items with low scene matching accuracy parameters is 20%, and the average deviation degree is 15%, then the correction magnitude for the scene matching accuracy parameter is set to be increased by 10%. The adjustment strategy clearly specifies the correction magnitudes for the time matching accuracy parameter and the scene matching accuracy parameter.

[0097] Step S154: According to the aforementioned association parameter adjustment strategy, the time matching precision parameter and scene matching degree parameter of all associated entries in the association mapping relationship are uniformly adjusted to generate the adjusted association mapping relationship.

[0098] In this embodiment, based on the correlation parameter adjustment strategy established in step S153, the time matching precision parameter and scene matching degree parameter of all correlation entries in the correlation mapping relationship are uniformly adjusted. Specifically, for the time matching precision parameter, according to the correction range specified in the strategy, a corresponding correction amount is added to the original time matching precision parameter value of each correlation entry (corrected time matching precision parameter = original parameter value + original parameter value × correction range); for the scene matching degree parameter, similarly, according to the corresponding correction range, a corresponding correction amount is added to the original scene matching degree parameter value (corrected scene matching degree parameter = original parameter value + original parameter value × correction range). If the adjusted parameter value exceeds the preset maximum parameter value, it is limited to the maximum parameter value; if the adjusted parameter value is lower than the preset minimum parameter value, it is limited to the minimum parameter value, ensuring that the parameter values ​​are within a reasonable and effective range. After completing the parameter adjustment of all correlation entries, the adjusted correlation entries are reorganized and archived to generate an adjusted correlation mapping relationship. The correlation parameters in this adjusted correlation mapping relationship better conform to the correlation patterns of actual consumer behavior.

[0099] Step S155: Input the adjusted association mapping relationship into the user's offline consumption behavior analysis model, and re-mine the consumption scenario adaptability implied in the dwell record sequence and the payment behavior correlation implied in the payment record sequence.

[0100] In this embodiment, the adjusted association mapping relationship generated in step S154 is input into the user's offline consumption behavior analysis model. The model will repeat the core process in step S130, re-extracting features and mining indicators. Specifically, using the stay record identifier in the adjusted association mapping relationship as an index, the location change trajectory information and offline area information in the stay record sequence are re-associated to construct a new stay feature dataset; using the payment record identifier in the adjusted association mapping relationship as an index, the consumption scenario type information and product category information in the payment record sequence are re-associated to construct a new payment feature dataset. Then, in the stay feature processing layer, the movement path complexity feature is recalculated based on the new stay feature dataset, and after combining it with the offline area scenario attributes and standardization processing, the consumption scenario fit is regenerated; in the payment feature processing layer, the product selection preference feature is re-analyzed based on the new payment feature dataset, and the payment behavior correlation is recalculated based on the historical payment scenarios. Since the association parameters have been adjusted, the re-mined consumption scenario fit and payment behavior correlation can more accurately reflect the relationship between user stay and payment.

[0101] Step S156: Combining the re-mined consumption scenario adaptability and payment behavior correlation, perform the user's offline consumption behavior tendency inference again to generate the final offline consumption behavior prediction result, which includes consumption behavior type, corresponding offline region, evaluation score and credibility label.

[0102] In this embodiment, the newly mined consumption scenario suitability and payment behavior correlation are input into the user's offline consumption behavior analysis model, and the behavior tendency inference process in step S140 is repeated. First, the scenario weight allocation module assigns new scenario influence weights based on the newly obtained consumption scenario suitability, and the payment weight allocation module assigns new payment influence weights based on the newly obtained payment behavior correlation. Next, the behavior tendency inference module calls the time synchronization correspondence in the adjusted correlation mapping relationship, recalculates the time correlation coefficient, and performs time calibration on the scenario influence weights and payment influence weights. Then, the calibrated weights are input into the consumption behavior tendency evaluation function to regenerate the evaluation score corresponding to each potential consumption behavior. Finally, the evaluation scores are sorted and the top K potential consumption behaviors are extracted. Based on this, a credibility label is added to each extracted potential consumer behavior. The credibility label is determined according to the evaluation score and a preset credibility level threshold. For example, when the evaluation score is in the range of 0.8-1.0, the credibility label is "high credibility"; when it is in the range of 0.6-0.8, the credibility label is "medium credibility"; and when it is in the range of 0.4-0.6, the credibility label is "low credibility". The consumer behavior type, corresponding offline region, evaluation score, and credibility label are integrated to form the final offline consumer behavior prediction result. This final offline consumer behavior prediction result not only includes the predicted content and probability quantification indicators of the consumer behavior, but also clarifies the reliability of the prediction result through the credibility label.

[0103] Figure 2 The illustration shows exemplary hardware and software components of a user geolocation-based offline consumer behavior prediction system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the user geolocation-based offline consumer behavior prediction system 100 and to perform the functions in this application.

[0104] The offline consumption behavior prediction system 100 based on user geolocation can be a general-purpose server or a special-purpose server; both can be used to implement the offline consumption behavior prediction method based on user geolocation of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0105] For example, the user location-based offline consumer behavior prediction system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the user location-based offline consumer behavior prediction system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The user location-based offline consumer behavior prediction system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0106] For ease of explanation, only one processor is described in the user location-based offline consumption behavior prediction system 100. However, it should be noted that the user location-based offline consumption behavior prediction system 100 of this application may also include multiple processors, and therefore the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the user location-based offline consumption behavior prediction system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0107] Furthermore, embodiments of the present invention also provide a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned method for predicting offline consumption behavior based on user geographical location is implemented.

[0108] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1.A method for predicting offline consumer behavior based on user geographic location, characterized in that, The method comprises: obtaining a user geographic location data set and a user payment big data set, the user geographic location data set containing a user stay record sequence in different offline areas, and the user payment big data set containing a user payment record sequence in different offline consumption scenarios; performing dynamic association processing on the user geographic location data set and the user payment big data set to generate an association mapping relationship between user geographic location and payment behavior, the association mapping relationship being used to represent a time synchronization correspondence relationship between the stay record sequence and the payment record sequence; constructing a user offline consumption behavior analysis model based on the association mapping relationship, and mining a consumption scenario adaptation degree implied in the stay record sequence and a payment behavior association degree implied in the payment record sequence through the consumption behavior analysis model; combining the consumption scenario adaptation degree and the payment behavior association degree to perform user offline consumption behavior tendency deduction to generate a preliminary consumption behavior prediction result; adjusting association parameters in the association mapping relationship according to the preliminary consumption behavior prediction result, and performing consumption behavior tendency deduction again based on the adjusted association mapping relationship to generate a final offline consumption behavior prediction result. 2.The user-geolocation-based offline consumer behavior prediction method of claim 1, wherein, The method comprises: receiving position information generated by a user in different offline areas from a geographic location collection terminal, arranging the position information in a continuous time sequence to obtain a stay record sequence of the user in different offline areas, the stay record sequence containing time information of the user entering an offline area, time information of the user leaving the offline area, and position change trajectory information during the stay; extracting payment information generated after the user completes payment in different offline consumption scenarios from a payment data management system, arranging the payment information in a time sequence to obtain a payment record sequence of the user in different offline consumption scenarios, the payment record sequence containing consumption scenario type information corresponding to payment, product category information corresponding to payment, and time information of payment operation completion; performing time dimension calibration on the time information in the stay record sequence and the time information of payment operation completion in the payment record sequence, extracting valid stay records in the calibrated stay record sequence whose stay duration exceeds a preset time threshold to form a valid stay record subsequence, and taking the valid stay record subsequence as a core component of the user geographic location data set; extracting valid payment records corresponding to offline entity consumption scenarios in the calibrated payment record sequence to form a valid payment record subsequence, and taking the valid payment record subsequence as a core component of the user payment big data set. 3.The user-geolocation-based offline consumer behavior prediction method of claim 1, wherein, The method comprises: extracting time interval information of the stay record sequence in the user geographic location data set, the time interval information being determined by the time information of the user entering the offline area and the time information of the user leaving the offline area; extracting payment time point information of the payment record sequence in the user payment big data set, the payment time point information being time information of completion of a payment operation; determining whether each payment time point information falls within time interval information of any one of the stay record sequences, and if the payment time point information falls within the time interval information of any one of the stay record sequences, marking the payment record and the stay record as a preliminary association pair; performing scene matching verification on offline area information corresponding to the stay record in the preliminary association pair and consumption scene type information corresponding to the payment record, determining whether the offline area corresponding to the stay record matches the consumption scene type corresponding to the payment record, and if so, retaining the preliminary association pair, and if not, eliminating the preliminary association pair; arranging the retained preliminary association pairs in chronological order, constructing an association mapping table containing stay record identifiers, payment record identifiers, time corresponding relationships and scene matching results, and taking the association mapping table as an association mapping relationship between user geographic positions and payment behaviors. 4.The method of claim 3, wherein, The scene matching verification on offline area information corresponding to the stay record in the preliminary association pair and consumption scene type information corresponding to the payment record, determining whether the offline area corresponding to the stay record matches the consumption scene type corresponding to the payment record, and if so, retaining the preliminary association pair, and if not, eliminating the preliminary association pair, includes: constructing a target matching rule library, the target matching rule library storing mapping relationships between different offline area identifiers and corresponding consumption scene types, each offline area identifier corresponding to at least one consumption scene type; extracting offline area identifiers of the stay record in the preliminary association pair and consumption scene type information of the payment record; inputting the offline area identifier into the target matching rule library to query all consumption scene types corresponding to the offline area identifier; determining whether the consumption scene type information of the payment record belongs to one of all the consumption scene types obtained by the query, and if so, determining that the offline area corresponding to the stay record matches the consumption scene type corresponding to the payment record, and retaining the preliminary association pair; if the consumption scene type information of the payment record does not belong to one of all the consumption scene types obtained by the query, further analyzing the affiliated consumption scene types around the offline area corresponding to the stay record, and determining whether the consumption scene type information of the payment record belongs to the affiliated consumption scene types; if the consumption scene type information belongs to the affiliated consumption scene types, supplementing and updating the target matching rule library to establish a mapping relationship between the offline area identifier and the affiliated consumption scene types, and retaining the preliminary association pair; if the consumption scene type information does not belong to the affiliated consumption scene types, determining that the offline area does not match the consumption scene type, and eliminating the preliminary association pair. 5.The user-geolocation-based offline consumer behavior prediction method of claim 1, wherein, The user offline consumption behavior analysis model is constructed based on the association mapping relationship, and the consumption behavior analysis model is used to mine the consumption scene adaptation degree implied in the stay record sequence and the payment behavior association degree implied in the payment record sequence, including: taking the stay record identifier in the association mapping relationship as an index, associating position change trajectory information and offline area information in the corresponding stay record sequence, and constructing a stay feature data set; Taking the payment record identifier in the association mapping relationship as an index, the consumption scene type information and the commodity category information in the corresponding payment record sequence are associated to construct a payment feature dataset; The stay feature dataset and the payment feature dataset are input into a model construction module to initialize a basic framework of a user offline consumption behavior analysis model, and the basic framework includes a stay feature processing layer, a payment feature processing layer and a feature association layer; In the stay feature processing layer, path analysis is performed on the location change trajectory information in the stay feature dataset, the moving path complexity feature of the user during the stay is extracted, the scene attribute corresponding to the offline area information is combined, and the consumption scene adaptation degree of the user in the offline area is calculated after standardization processing, and the consumption scene adaptation degree is used to reflect the matching degree of the user stay area and the potential consumption demand; In the payment feature processing layer, the consumption scene type information and the commodity category information in the payment feature dataset are associated and analyzed, the commodity selection preference feature of the user in different consumption scenes is extracted, and the payment behavior association degree of the user in the current payment scene and the historical payment scene is calculated, and the payment behavior association degree is used to reflect the similarity between the current payment behavior and the historical payment behavior of the user; In the feature association layer, an association calculation rule of the consumption scene adaptation degree and the payment behavior association degree is established. 6.The method of claim 5, wherein, In the stay feature processing layer, path analysis is performed on the location change trajectory information in the stay feature dataset, the moving path complexity feature of the user during the stay is extracted, the scene attribute corresponding to the offline area information is combined, and the consumption scene adaptation degree of the user in the offline area is calculated after standardization processing, and the consumption scene adaptation degree is used to reflect the matching degree of the user stay area and the potential consumption demand; Trajectory point extraction is performed on the location change trajectory information in the stay feature dataset to obtain a plurality of continuous trajectory point coordinates of the user during the stay; The distance between adjacent trajectory point coordinates is calculated, the sum of the distances of all adjacent trajectory points is taken as the total length of the trajectory, and the total number of trajectory points is counted; The average step length of the trajectory is calculated according to the total length of the trajectory and the total number of trajectory points, the aggregation degree of the trajectory points in the spatial distribution is analyzed, the dispersion coefficient of the trajectory point distribution is calculated, and the average step length and the dispersion coefficient are taken as the core parameters of the moving path complexity feature; The offline area information corresponding to the stay record in the stay feature dataset is extracted, and the scene attribute of the offline area is determined, and the scene attribute includes the area function type, the density of the merchants in the area and the main service object group of the area; The core parameters of the moving path complexity feature and the scene attribute of the offline area are standardized to obtain the standardized parameters; A consumption scene adaptation degree calculation model is constructed, the standardized parameters are taken as input parameters, and each input parameter is assigned a corresponding calculation weight; The input parameters are weighted and operated by the consumption scene adaptation degree calculation model to generate the consumption scene adaptation degree of the user in the offline area, and the numerical range of the consumption scene adaptation degree is positively correlated with the moving path complexity and the matching degree of the scene attribute. 7.The user-geolocation-based offline consumer behavior prediction method of claim 1, wherein, The user offline consumption behavior tendency deduction is performed by combining the consumption scene adaptation degree and the payment behavior correlation degree, and a preliminary consumption behavior prediction result is generated, including: The consumption scene adaptation degree is input into a scene weight distribution module of a user offline consumption behavior analysis model. The scene weight distribution module distributes scene influence weights for different offline regions corresponding to the consumption scene according to the numerical size of the consumption scene adaptation degree; The payment behavior correlation degree is input into a payment weight distribution module of the user offline consumption behavior analysis model. The payment weight distribution module distributes payment influence weights for different payment records corresponding to the consumption behavior according to the numerical size of the payment behavior correlation degree; A behavior tendency deduction module of the user offline consumption behavior analysis model is called. The behavior tendency deduction module takes the scene influence weights and the payment influence weights as inputs, combines the time synchronization corresponding relationship in the association mapping relationship, and constructs a consumption behavior tendency evaluation function; The potential consumption behaviors of the user in different offline regions are evaluated for possibility by the consumption behavior tendency evaluation function, and an evaluation score corresponding to each potential consumption behavior is generated; The evaluation scores are sorted in descending order, and the K potential consumption behaviors with the highest evaluation scores and the corresponding evaluation scores are extracted to form a preliminary consumption behavior prediction result including the consumption behavior type, the corresponding offline region, and the evaluation score. 8.The method of claim 7, wherein, The behavior tendency deduction module of the user offline consumption behavior analysis model is called. The behavior tendency deduction module takes the scene influence weights and the payment influence weights as inputs, combines the time synchronization corresponding relationship in the association mapping relationship, and constructs a consumption behavior tendency evaluation function, including: In the behavior tendency deduction module, an evaluation parameter input interface is created. The scene influence weights and the payment influence weights are received through the evaluation parameter input interface, and the time synchronization corresponding relationship in the association mapping relationship is read at the same time; Based on the time synchronization corresponding relationship, the association strength between the time interval corresponding to each scene influence weight and the time point corresponding to each payment influence weight is determined, and a time association coefficient is generated; The scene influence weights and the corresponding time association coefficients are multiplied to obtain time-calibrated scene weights; the payment influence weights and the corresponding time association coefficients are multiplied to obtain time-calibrated payment weights; The user historical offline consumption behavior data is analyzed, the actual consumption behavior records of the user under different combinations of scene weights and payment weights are extracted, and a corresponding relationship sample library of consumption behavior and weight combination is established; Based on the corresponding relationship sample library, a statistical analysis method is used to construct a consumption behavior tendency evaluation function. The consumption behavior tendency evaluation function takes the time-calibrated scene weights and the time-calibrated payment weights as independent variables, and takes the possibility evaluation score of the potential consumption behavior as the dependent variable; The consumption behavior tendency evaluation function is verified. The known scene weights and payment weights are input, and the evaluation score output by the consumption behavior tendency evaluation function is compared with the actual consumption behavior occurrence. The coefficient parameters in the consumption behavior tendency evaluation function are adjusted. 9.The user-geolocation-based offline consumer behavior prediction method of claim 1, wherein, The association parameter in the association mapping relationship is adjusted according to the preliminary consumption behavior prediction result, the consumption behavior tendency deduction is performed again based on the adjusted association mapping relationship, and a final offline consumption behavior prediction result is generated, including: Analyzing the association mapping relationship item corresponding to the low-credibility consumption behavior prediction item, determining the association parameter of the stay record and the payment record corresponding to the association mapping relationship item, and the association parameter includes a time matching accuracy parameter and a scene matching degree parameter; According to the number of low-credibility consumption behavior prediction items and the deviation degree of the corresponding association parameters, an association parameter adjustment strategy is formulated, the association parameter adjustment strategy includes a correction amplitude of the time matching accuracy parameter and a correction amplitude of the scene matching degree parameter; According to the association parameter adjustment strategy, the time matching accuracy parameter and the scene matching degree parameter of all association items in the association mapping relationship are uniformly adjusted, and an adjusted association mapping relationship is generated; The adjusted association mapping relationship is input into the user offline consumption behavior analysis model, and the consumption scene adaptation degree hidden in the stay record sequence and the payment behavior association degree hidden in the payment record sequence are re-mined; Combined with the consumption scene adaptation degree and the payment behavior association degree obtained by re-mining, the user offline consumption behavior tendency deduction is performed again to generate a final offline consumption behavior prediction result including a consumption behavior type, a corresponding offline area, an evaluation score and a credibility identifier. A processor and a memory are included, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to realize the offline consumption behavior prediction method based on user geographic position in any one of claims 1-9. 10.A system for predicting offline consumer behavior based on user geographic location, the system comprising: ​

Images