User travel behavior label identification method and device

Abstract

The present application relates to the field of data processing, and particularly relates to a user travel behavior label identification method and device, the method comprising: obtaining target user information and converting into a user feature vector; determining a travel behavior label corresponding to the user feature vector based on a plurality of preset travel behavior labels respectively corresponding to a plurality of determination conditions and a plurality of travel behavior label prediction models; if the user feature vector and any determination condition have a preset association relationship, then the travel behavior label corresponding to the determination condition is taken as the result; otherwise, the user feature vector is input into the plurality of travel behavior label prediction models, and the result is determined according to the model output. The present application sets different key feature subsets as determination conditions for different travel behaviors, and combines the two-stage architecture of rule priority and model bottoming, thereby ensuring the identification accuracy, reducing the calling frequency of complex models, and improving the overall identification efficiency.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method and apparatus for identifying user travel behavior tags. Background Technology

[0002] In scenarios such as intelligent transportation, large-scale event management, and urban emergency dispatch, it is necessary to identify user travel behaviors at specific times to specific areas (such as event venues, transportation hubs, etc.) in order to predict traffic pressure in advance, formulate diversion plans, or optimize resource allocation. Travel behavior tags can include: using private vehicles, using public transportation, using floating transportation (such as ride-hailing services, taxis), walking, etc.

[0003] Existing methods for identifying user travel behavior mainly fall into two categories: The first category is pure rule-based matching methods. These methods make hard judgments based on preset thresholds or conditions (such as the distance between a user's residence and destination, whether they own a private car, or the installation status of specific applications). Their advantages are fast computation speed and strong interpretability. However, due to the complexity and diversity of user behavior in reality, rules often cannot cover all situations, resulting in low recognition coverage and the inability to output valid labels for users who do not conform to the rules.

[0004] The second category is pure machine learning model methods. This method inputs user feature vectors into a pre-trained classification model (such as a binary or multi-classifier), and the model outputs travel behavior labels. Its advantages include strong generalization ability and the capacity to handle complex boundary conditions. However, its disadvantage is that it requires performing model inference for each user, resulting in high computational overhead, especially when the user base reaches hundreds of thousands or even millions. This can create significant computational pressure in scenarios with high real-time requirements (such as completing label identification for all users within hours before an event).

[0005] Therefore, how to reduce the number of model calls and improve the overall recognition efficiency while ensuring recognition accuracy is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned technical problems, the technical solution adopted by this invention is as follows: According to a first aspect of the present invention, a method for identifying user travel behavior tags is provided, the method comprising the following steps: Obtain target user information, which includes device information and attribute information. The target user is the user associated with the target grid location at a specified time node.

[0007] The target user information is converted into a user feature vector.

[0008] Based on the judgment conditions corresponding to multiple preset travel behavior tags and multiple travel behavior tag prediction models, the travel behavior tag corresponding to the user feature vector is determined: if the user feature vector has a preset correlation with any judgment condition, then the travel behavior tag corresponding to the judgment condition is taken as the result; otherwise, the user feature vector is input into the multiple travel behavior tag prediction models, and the result is determined according to the output of the travel behavior tag prediction models.

[0009] The determination criteria are set based on a subset of key features in the user feature vector, and the subset of key features on which the determination criteria for different travel behavior tags are based are different.

[0010] According to a second aspect of the present invention, a user travel behavior tag identification device is provided, the device comprising: The user information acquisition module is used to acquire target user information, which includes device information and attribute information. The target user is a user associated with a target grid location at a specified time node.

[0011] The feature vector generation module is used to convert the target user information into user feature vectors.

[0012] The determination module is used to determine the travel behavior label corresponding to the user feature vector based on the determination conditions corresponding to multiple preset travel behavior labels and multiple travel behavior label prediction models: if the user feature vector has a preset correlation with any determination condition, then the travel behavior label corresponding to the determination condition is taken as the result; otherwise, the prediction module is triggered; wherein, the determination conditions are set based on the key feature subset in the user feature vector, and the key feature subsets on which the determination conditions corresponding to different travel behavior labels are based are different.

[0013] The prediction module is used to input the user feature vector into multiple travel behavior label prediction models when triggered, and determine the travel behavior label corresponding to the user feature vector based on the output of the travel behavior label prediction models.

[0014] The present invention has at least the following beneficial effects: This invention sets different subsets of key features for different travel behavior tags, allowing each judgment condition to focus only on features related to that travel behavior (e.g., walking behavior focuses on activity levels in motion monitoring apps, driving behavior focuses on distance and vehicle ownership status). Compared to using the same full set of features for matching, this reduces the introduction of features irrelevant to specific tags, lowering computational redundancy in the rule matching stage. Based on this, a two-stage architecture of rule-first and model-backup is adopted: for users with a pre-defined association with any judgment condition, the result is directly output without calling a prediction model; for users not covered by the rules, multiple binary classification models are then launched for prediction. Through this differentiated feature selection and rule-first mechanism, the number of users requiring prediction model calls can be reduced, thereby lowering the overall computational cost under the same hardware conditions while maintaining relatively high recognition coverage and accuracy.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a user travel behavior tag recognition method provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] It should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the steps as sequential processes, many of these steps can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the steps can be rearranged. A process can be terminated when its operation is complete, but it may also have additional steps not included in the figures. A process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0021] This embodiment provides a method for identifying user travel behavior tags, which can be applied to servers or cloud platforms. For example... Figure 1 As shown, the method includes the following steps: S100, Obtain target user information.

[0022] Obtain information on N target users (N≥1). The target user information includes the target user's device information and attribute information. The target user is the user associated with the target grid location at a specified time node.

[0023] 1. Definitions of relevant concepts Specify a time point: This can be a specific time on a particular date, such as "year x month x day 19:00".

[0024] Target grid location: This can be a geographical area or space that provides a specific service, such as a grid area for a large event or a specific building. Grid locations can be uniquely identified by their boundary latitude and longitude coordinates.

[0025] Target users can be identified through internet interactions (such as online ticket purchases and event registrations) or through spatiotemporal targeting. Specifically, user information can be retrieved from a pre-defined user information table based on the target grid location ID and a specified time point. Each record in the pre-defined user information table can include: grid location ID, time point, and user identifier.

[0026] 2. Attribute Information The attribute information includes: gender, age, demand level, first stop grid location, second stop grid location, and private vehicle ownership status.

[0027] Demand Level: This characterizes a user's level of demand for living resources (such as consumption level). A higher demand level indicates stronger purchasing power or a greater demand for high-end resources; a lower demand level indicates a greater preference for cost-effective or discounted resources. Demand levels can be obtained through existing prediction models. Input features include: mobile device model, the property / commercial value corresponding to the grid location where the mobile terminal spends the most time, the frequency of use of high-end consumer applications (such as luxury shopping and high-end hotel booking), and the frequency of use of affordable / discount applications (such as group buying, cashback, and coupons). For example, demand levels can be divided into three levels: Level 1 (high demand), Level 2 (medium demand), and Level 3 (low demand). First Dwell Grid Location: Within a preset historical time period (e.g., within one year), the grid location where the mobile terminal spends the most time outside of a specified time period (e.g., 8 AM to 6 PM). This typically corresponds to a home address.

[0028] Second dwell grid location: Within the same preset historical time period, the grid location where the mobile terminal spends the longest time during a specified time period (e.g., from 8:00 AM to 6:00 PM). This usually corresponds to an office address.

[0029] Private vehicle ownership status: This indicates whether or not one owns a private vehicle (e.g., a car), including first ownership status (ownership) and second ownership status (non-ownership).

[0030] The location information of a mobile terminal can be obtained through location data reported by the user after authorization.

[0031] 3. Equipment Information The equipment information includes first equipment information and second equipment information.

[0032] (1) First equipment information: Statistically analyze the following data within the first preset time period (e.g., the last 3 months): Installation volume and activity level of public transportation apps (bus, subway); Installation volume and activity levels of floating transportation apps (ride-hailing, taxi); Activity levels of lightweight apps related to public transportation; Activity level of lightweight applications in the category of floating vehicles; The number of installations of transportation service applications (such as car maintenance applications); The installation volume and activity level of motion monitoring apps (such as fitness apps); The installation volume and activity of applications in the first, second, and third level venue categories.

[0033] Among them, the number of installations refers to the total number of newly installed applications of this type within the first preset time period; the activity level refers to the ratio of the number of active days of the application to the total number of days within the first preset time period.

[0034] Venue rating: Level 1 > Level 2 > Level 3, determined by the level of service provided by the venue (the higher the service level, the higher the rating).

[0035] In this application, an application refers to a software program that needs to be installed independently on an operating system; a lightweight program refers to a lightweight program that does not require independent installation and runs on a host platform (such as WeChat).

[0036] (2) Second equipment information: Statistically analyze the following data within the second preset time period (e.g., the past month): Set the visit density for activity venues: Activity venues refer to high-end locations such as golf courses, resorts, and five-star hotels or above. The visit density equals the total number of visits to this type of activity venue within the second preset time period divided by the total number of days included in the second preset time period.

[0037] Visitor density at transportation service locations: such as gas stations and 4S stores. The calculation formula is the same as above.

[0038] Travel range: Within the second preset time period, the furthest distance (e.g., 5 km, 10 km) from the first stationary grid location among the grid locations traversed by the mobile terminal. The distance between grid locations is calculated using latitude and longitude coordinates.

[0039] Travel time: refers to the average travel time required to travel from the first stop grid position to the second stop grid position, which can be calculated using historical location information reported by the mobile terminal.

[0040] S200, the target user information is converted into a corresponding user feature vector.

[0041] Each target user's information is encoded to obtain a corresponding user feature vector. Encoding methods include: one-hot or sequential encoding for categorical features, and binning encoding for continuous features (e.g., dividing intervals based on quantiles). The user feature vector contains the following three types of encoded values: User attribute feature encoding values: including at least gender encoding values, age encoding values, demand level encoding values, and private vehicle ownership status encoding values.

[0042] Device usage characteristic coding values: including at least the installation volume coding value and activity coding value of travel-related applications within a first preset time period.

[0043] Location movement feature encoding values: including at least distance encoding values, visit density encoding values, travel range encoding values ​​and travel time encoding values, wherein the distance encoding values ​​are used to characterize the positional relationship between the first dwell grid position and the target grid position.

[0044] The following provides a detailed explanation of each encoded value.

[0045] (1) User attribute feature encoding value Gender coding values: include primary gender coding value (male) and secondary gender coding value (female).

[0046] Age coding values: The age is divided into multiple intervals, and each interval corresponds to a coding value. The division method can be determined based on the quantiles of the sample distribution.

[0047] Demand level coding values: including first demand level coding values, second demand level coding values, and third demand level coding values, which correspond to high demand, medium demand, and low demand, respectively, with the demand level decreasing in that order.

[0048] Private vehicle ownership status code values: including first ownership status code value (owning a private vehicle, such as a car) and second ownership status code value (not owning).

[0049] (2) Equipment usage feature code value The device usage feature encoding values ​​are derived from various application data in the first device information and the second device information. Specifically, these include: Installation volume and activity levels of public transportation apps; Install count and activity level codes for floating vehicle applications; Activity coding values ​​for lightweight public transportation apps; The activity code value for lightweight applications of the floating vehicle type; The installation volume encoding value for transportation service applications; Installation volume and activity level codes for motion monitoring applications; Installation volume and activity levels of applications in Level 1, Level 2, and Level 3 venues.

[0050] For the aforementioned application types, both the installation volume and activity level codes employ a segmented coding method: the installation volume or activity level within a first preset time period is divided into multiple numerical ranges, each corresponding to a code value. The division boundaries can be determined based on the quantiles of the sample (e.g., 25%, 50%, 75%), and the specific number of code values ​​can be set according to actual needs. Taking floating transportation applications as an example, their installation volume code values ​​may include multiple floating installation volume code values, each corresponding to an installation volume range; the activity level code values ​​are similarly structured.

[0051] (3) Location movement feature encoding value The location movement feature encoding values ​​include: Distance encoding value: Used to characterize the positional relationship between the first dwelling grid position and the target grid position. Specifically includes: First distance code value: The first dwelling grid position and the target grid position do not belong to the same geographical area; Second distance code value: They belong to the same geographical area, but the distance between them is greater than the set distance (e.g., 2 kilometers). The third distance code value: belongs to the same geographical area, and the distance between the two is less than or equal to the set distance.

[0052] The distance encoding value can be calculated based on the latitude and longitude coordinates of the first dwell grid position and the target grid position.

[0053] The visit density coding value includes the visit density coding values ​​for designated activity venues and transportation service venues within the second preset time period, as well as the travel range coding value and travel time coding value. The visit density is equal to the total number of visits to this type of venue within the time period divided by the total number of days in the time period. The calculated density value is divided into multiple intervals, with each interval corresponding to a coding value.

[0054] Travel range encoding value: The travel range refers to the farthest distance from the first stationary grid position among the grid positions traversed by the mobile terminal within the second preset time period. This distance value is divided into multiple intervals, each interval corresponding to an encoding value.

[0055] Travel time encoding value: Travel time refers to the average travel time required to travel from the first stop grid position to the second stop grid position. This time value is divided into multiple intervals, and each interval corresponds to a encoding value.

[0056] (4) Determination of encoding rules The coding rules (i.e., the division of interval boundaries) for the aforementioned continuous features such as age, installation volume, activity level, visit density, travel range, and travel time can be obtained based on sample data analysis. Specifically, multiple sample information from historical data on travel to the target grid location at specified time points are collected. Each sample includes user information and its corresponding travel behavior tag. Then, a custom interval binning method or quantile method is used to encode each feature, resulting in the coding rule for each feature. Different coded values ​​correspond to different value ranges.

[0057] S300, based on the judgment conditions corresponding to multiple preset travel behavior tags and multiple travel behavior tag prediction models, determine the travel behavior tag corresponding to the user feature vector.

[0058] 1. Identification of key features Key features are those strongly correlated with travel behavior, selected from user feature vectors. In this embodiment, key features include: demand level encoding value, distance encoding value, private vehicle ownership status encoding value, activity encoding value of public transportation applications, activity encoding value of floating transportation applications, activity encoding value of lightweight public transportation applications, installation volume encoding value and activity encoding value of motion monitoring applications.

[0059] The key feature subset corresponding to each travel behavior label can be determined in the following ways: sort the feature weights in the trained travel behavior label prediction model from high to low, and select the top K features, or the top certain percentage, or features with weights greater than a set threshold (such as 0.6) as key features; or obtain them based on feature analysis of sample information.

[0060] 2. Definition of Decision Criteria Each preset travel behavior tag corresponds to a judgment condition. The judgment condition is set based on a subset of key features in the user's feature vector, and the subset of key features on which the judgment condition is based is different for different travel behavior tags.

[0061] By setting different subsets of key features for different travel behavior labels, this invention establishes a differentiated rule matching strategy. The discriminative features relied upon by different travel behaviors are fundamentally different: for example, walking behavior is strongly correlated with the intensity of a user's use of fitness and health apps, while driving behavior is strongly correlated with the distance to the user's home address, private car ownership status, and usage of vehicle service apps. If the same full set of features is used for matching all labels, on the one hand, a large number of noisy features unrelated to specific labels will be introduced, reducing matching accuracy; on the other hand, feature redundancy will increase unnecessary computational overhead. This application, through targeted screening, focuses each judgment condition on only a small number of key features most relevant to the travel behavior, thereby achieving higher hit accuracy and lower computational complexity in the rule matching stage, effectively reducing the number of users requiring complex prediction models, and significantly improving overall recognition efficiency.

[0062] In this embodiment, four travel behavior labels are preset: using floating transportation (such as ride-hailing), using public transportation (bus / subway), walking, and using private transportation (private car). Corresponding judgment conditions (i.e., combinations of values ​​for key feature subsets) are set for each, as shown in the following example: (1) Criteria for determining floating vehicles Its key feature subset includes: a first distance coding value (the home address and the event venue are not in the same area), a coding value representing ownership of private vehicles, a coding value representing the activity level of floating vehicle applications above a first threshold, a coding value representing the activity level of public transportation applications below a second threshold, and a coding value representing the activity level of lightweight public transportation applications below a third threshold.

[0063] In a specific instance, the criteria can be expressed as follows: the home address is not at the location of the event; the applicant owns a car; the activity level of ride-hailing apps is higher than 0.75 in the past 3 months; and the activity level of public transportation apps and lightweight apps is lower than 0.1 in the past 3 months.

[0064] (2) Criteria for determining public transportation vehicles Its key feature subset includes: a first distance coding value, a coding value representing not owning private vehicles, a coding value representing the activity level of public transportation applications above the fourth threshold, a coding value representing the activity level of lightweight public transportation applications above the fifth threshold, a coding value representing the activity level of floating transportation applications below the sixth threshold, and a coding value representing the demand level below a preset level (such as the second or third demand level).

[0065] In a specific instance, the criteria can be expressed as follows: the home address is not at the location of the event; the applicant does not own a car; the activity level of public transportation apps and lightweight apps is higher than 0.75 in the past 3 months; the activity level of ride-hailing apps is lower than 0.1 in the past 3 months; and the demand level is the second or third demand level.

[0066] (3) Walking determination criteria Its key feature subset includes: the third distance encoding value (the home address and the event venue are in the same area and the distance is ≤ a set distance, such as 2 kilometers), and the encoding value that represents the installation volume of motion status monitoring applications that is higher than the seventh threshold and the activity level that is higher than the eighth threshold.

[0067] In a specific instance, the criteria could be expressed as: the home address is within 2 kilometers of the event location; and the installation volume and activity level of motion monitoring applications are both higher than 0.25 within the past 3 months.

[0068] (4) Criteria for determining private vehicles Its key feature subset includes: second distance coding value (home address and event location are in the same area but distance > set distance), coding value representing ownership of private vehicles, coding value representing demand level higher than the ninth threshold (such as first or second demand level), coding value representing activity level of floating vehicle application below the tenth threshold, coding value representing activity level of public transportation application below the eleventh threshold, and coding value representing activity level of lightweight public transportation application below the twelfth threshold.

[0069] In a specific instance, the criteria can be expressed as follows: the home address is more than 2 kilometers away from the event location; the applicant owns a car; the demand level is either the first or second demand level; the activity level of ride-hailing apps is less than 0.1 in the past 3 months; and the activity level of public transportation apps and lightweight apps is less than 0.1 in the past 3 months.

[0070] The thresholds (from the first threshold to the twelfth threshold) involved in the above-mentioned judgment conditions are all configurable parameters, and their specific values ​​can be preset according to the actual application scenario and the distribution of sample data. For example, the threshold value can be determined based on the quantiles of the corresponding features within the first preset time period (such as 25%, 50%, and 75%), or the feature distribution of positive and negative samples in the sample data can be used as a reference. Those skilled in the art can reasonably set the above thresholds according to specific business needs and data characteristics, and the specific values ​​will not be listed here.

[0071] 3. Identification based on decision conditions Determine whether a pre-defined correlation exists between the user feature vector and each decision condition. The pre-defined correlation must include at least one or more of the following: The user feature vector contains all the key feature subsets defined by the judgment criteria; The similarity (e.g., cosine similarity) between the user feature vector and the key feature subset defined by the judgment criteria must be greater than or equal to a similarity threshold. This similarity threshold can be preset according to the actual application scenario, for example, set to 0.8. When the similarity reaches this threshold, an association is considered to exist. In the user feature vector, all feature values ​​related to the judgment conditions fall within a preset numerical range. This numerical range can be predetermined based on the distribution of the sample data (e.g., quantiles), and each key feature corresponding to a judgment condition can have its own reasonable range set.

[0072] Judgment Process: The judgment conditions corresponding to each travel behavior tag are sequentially traversed. For each judgment condition, it is checked whether the user feature vector has a preset correlation with that condition (e.g., whether it contains all the key feature subsets defined by the condition, whether the similarity reaches a threshold, or whether the feature value falls within a preset range). Once a preset correlation is found, the traversal is immediately stopped, and the travel behavior tag corresponding to that condition is taken as the recognition result; if no preset correlation is found after traversing all judgment conditions, it is determined that the model prediction stage needs to be entered.

[0073] 4. Identification based on travel behavior label prediction models If the user feature vector has no preset correlation with any of the decision conditions, then proceed to the model prediction stage.

[0074] In this embodiment, four travel behavior label prediction models are set up, all of which are binary classification models (such as logistic regression, XGBoost, support vector machines, etc.), corresponding to floating vehicles, public transportation, walking, and private transportation, respectively. Each model is trained using multiple sample information and is used to predict whether a user has the corresponding travel behavior label. During training, historical user data is used as samples, and the corresponding travel behavior labels are used as supervision signals. Conventional machine learning training methods (such as gradient descent and cross-validation) are used to obtain model parameters. The specific structure of the above models can adopt the conventional binary classification model architecture in this field, which will not be elaborated here.

[0075] Specifically, the user's feature vector is input into four prediction models, resulting in four predicted scores (i.e., the probability that the user belongs to a corresponding label). The label corresponding to the highest predicted score is taken as the user's final travel behavior label. For example, if the user's feature vector input into the floating model yields 0.85, into the public transportation model 0.10, into the walking model 0.03, and into the private transportation model 0.02, then the user is ultimately determined to have used floating transportation.

[0076] 5. Optional reference library matching stage (three-level recognition architecture) As a further optimization, a reference library matching stage can be added after the judgment conditions are not met and before the travel behavior label prediction model is called.

[0077] Specifically: If the user feature vector has no preset correlation with any of the judgment conditions, then based on the preset user travel behavior tag reference library, it is determined whether there is a travel behavior tag corresponding to the user feature vector. If it exists, the corresponding travel behavior tag is used as the recognition result; if it does not exist, the model prediction step continues.

[0078] The user travel behavior tag reference library stores multiple reference feature vectors and their corresponding travel behavior tags (e.g., high-confidence samples selected from historical data). The determination method is as follows: calculate the similarity (e.g., cosine similarity) between the current user feature vector and each reference vector. If there is a reference vector with a similarity greater than a preset threshold, then the travel behavior tag corresponding to that reference vector is taken as the matching result.

[0079] In one specific embodiment, the user travel behavior tag reference library can be organized according to travel behavior tag categories, with each category of tags corresponding to a set of reference feature vectors. For the user feature vector to be identified, its similarity with each category of reference vectors is calculated, the maximum similarity in each category is taken, and then the maximum value among these maximum similarities is taken. The tag corresponding to the maximum value is the matching result.

[0080] This three-level recognition architecture (judgment conditions → reference library → prediction model) can further reduce the number of calls to complex models and improve overall recognition efficiency.

[0081] Example 2

[0082] Based on the same inventive concept, embodiments of the present invention provide a user travel behavior tag identification device. This device includes the following modules: User information acquisition module: used to acquire target user information. The target user information includes device information and attribute information, and the target user is the user associated with the target grid location at a specified time node.

[0083] Feature vector generation module: used to convert the target user information into user feature vectors.

[0084] The determination module is used to determine the travel behavior label corresponding to the user feature vector based on determination conditions corresponding to multiple preset travel behavior labels and multiple travel behavior label prediction models. Specifically, if the user feature vector has a preset correlation with any determination condition, the travel behavior label corresponding to that determination condition is taken as the result; otherwise, the prediction module is triggered. The determination conditions are set based on a subset of key features in the user feature vector, and the subset of key features used for the determination conditions corresponding to different travel behavior labels is different.

[0085] Prediction module: When triggered, it inputs the user feature vector into multiple travel behavior label prediction models and determines the travel behavior label corresponding to the user feature vector based on the output of the travel behavior label prediction models.

[0086] Optionally, the device further includes a reference library matching module: used to determine whether there is a travel behavior tag corresponding to the user feature vector based on a preset user travel behavior tag reference library when the determination module determines that there is no preset correlation between the user feature vector and all determination conditions. If it exists, the corresponding travel behavior tag is used as the travel behavior tag corresponding to the user feature vector; if it does not exist, the prediction module is triggered. The preset user travel behavior tag reference library includes multiple reference feature vectors and their corresponding travel behavior tags.

[0087] This invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in this invention.

[0088] This invention also provides a non-transitory computer-readable storage medium storing computer-executable instructions for performing the methods described in this invention.

[0089] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0090] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for identifying user travel behavior tags, characterized in that, The method includes the following steps: Obtain target user information, which includes device information and attribute information, and the target user is the user associated with the target grid location at a specified time node; Convert the target user information into a user feature vector; Based on the judgment conditions corresponding to multiple preset travel behavior tags and multiple travel behavior tag prediction models, the travel behavior tag corresponding to the user feature vector is determined: if the user feature vector has a preset correlation with any judgment condition, then the travel behavior tag corresponding to the judgment condition is taken as the result; otherwise, the user feature vector is input into the multiple travel behavior tag prediction models, and the result is determined according to the output of the travel behavior tag prediction models. The determination criteria are set based on a subset of key features in the user feature vector, and the subset of key features on which the determination criteria for different travel behavior tags are based are different. The device information includes first device information and second device information; The first device information includes: the number of installations and activity levels of travel-related applications within a first preset time period, as well as the activity level of lightweight applications; The second device information includes: the visit density of designated activity locations, the visit density of transportation service locations, the travel range, and the travel time within a second preset time period. The travel time refers to the average travel time required to travel from the first dwelling grid location to the second dwelling grid location. The first dwelling grid location is the grid location where the mobile terminal has stayed the longest during the time outside the specified time period within the preset historical time period. The second dwelling grid location is the grid location where the mobile terminal has stayed the longest during the specified time period within the preset historical time period.

2. The method according to claim 1, characterized in that, The preset association relationship includes at least one or more of the following: The user feature vector contains all the key feature subsets defined by the judgment criteria; The similarity between the user feature vector and the key feature subset defined by the judgment condition is greater than or equal to the similarity threshold; The feature values ​​in the user feature vector that are related to the determination condition all fall within a preset numerical range.

3. The method according to claim 1, characterized in that, Determining the travel behavior label corresponding to the user feature vector based on the output of the travel behavior label prediction model includes: The user feature vector is input into multiple travel behavior label prediction models to obtain multiple corresponding prediction scores, wherein each travel behavior label prediction model corresponds to a travel behavior label; The travel behavior label corresponding to the largest predicted score among the multiple predicted scores is determined as the travel behavior label corresponding to the user feature vector.

4. The method according to claim 1, characterized in that, The method further includes: If there is no preset correlation between the user feature vector and all the judgment conditions, then based on the preset user travel behavior tag reference library, it is determined whether there is a travel behavior tag corresponding to the user feature vector; if there is, then the corresponding travel behavior tag is used as the travel behavior tag corresponding to the user feature vector; if there is, then the step of inputting the user feature vector into multiple travel behavior tag prediction models is executed. The preset user travel behavior tag reference library includes multiple reference feature vectors and their corresponding travel behavior tags.

5. The method according to claim 1, characterized in that, The attribute information includes: gender, age, demand level, first stop grid location, second stop grid location, and private vehicle ownership status; The demand level is used to characterize the degree of a user's demand for living resources.

6. The method according to claim 1, characterized in that, The user feature vector includes: User attribute feature codes include at least gender codes, age codes, demand level codes, and private vehicle ownership status codes; The device uses feature encoding values, including at least the installation volume encoding values ​​and activity encoding values ​​of travel-related applications within a first preset time period, as well as the activity encoding values ​​of lightweight applications; The location movement feature encoding value includes at least a distance encoding value, a visit density encoding value, a travel range encoding value, and a travel time encoding value, wherein the distance encoding value is used to characterize the positional relationship between the first dwell grid position and the target grid position.

7. The method according to claim 1, characterized in that, The determination conditions corresponding to the multiple preset travel behavior tags include at least one of the following: determination conditions corresponding to travel on floating vehicles, determination conditions corresponding to travel on public transportation, determination conditions corresponding to travel on foot, and determination conditions corresponding to travel on private vehicles.

8. A user travel behavior tag recognition device, characterized in that, include: The user information acquisition module is used to acquire target user information, which includes device information and attribute information. The target user is a user associated with a target grid location at a specified time node. The device information includes first device information and second device information; The first device information includes: the installation volume and activity level of travel-related applications within a first preset time period, as well as the activity level of lightweight applications; the second device information includes: the visit density of designated activity locations, the visit density of transportation service locations, the travel range, and the travel time within a second preset time period, wherein the travel time refers to the average travel time required to travel from the first stop grid location to the second stop grid location; the first stop grid location is the grid location where the mobile terminal has stayed the longest during a time period other than the specified time period within a preset historical time period, and the second stop grid location is the grid location where the mobile terminal has stayed the longest during the specified time period within a preset historical time period; The feature vector generation module is used to convert the target user information into a user feature vector; The determination module is used to determine the travel behavior label corresponding to the user feature vector based on the determination conditions corresponding to multiple preset travel behavior labels and multiple travel behavior label prediction models: if the user feature vector has a preset correlation with any determination condition, then the travel behavior label corresponding to the determination condition is taken as the result; otherwise, the prediction module is triggered; wherein, the determination condition is set based on the key feature subset in the user feature vector, and the key feature subsets on which the determination conditions corresponding to different travel behavior labels are based are different; The prediction module is used to input the user feature vector into multiple travel behavior label prediction models when triggered, and determine the travel behavior label corresponding to the user feature vector based on the output of the travel behavior label prediction models.

9. The apparatus according to claim 8, characterized in that, It also includes a reference library matching module, which is used to: if the determination module determines that there is no preset correlation between the user feature vector and all the determination conditions, then determine whether there is a travel behavior tag corresponding to the user feature vector based on a preset user travel behavior tag reference library; If it exists, the corresponding travel behavior tag is used as the travel behavior tag corresponding to the user feature vector; if it does not exist, the prediction module is triggered; wherein, the preset user travel behavior tag reference library includes multiple reference feature vectors and their corresponding travel behavior tags.

Images