Delivery person identification method and device, electronic equipment and storage medium
By using a deliveryman identification model based on signaling and internet data, combined with a shrinking autoencoder and location information, the problem of low accuracy in deliveryman identification in existing technologies has been solved, achieving more efficient and accurate deliveryman identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA MOBILE GRP BEIJING
- Filing Date
- 2023-06-21
- Publication Date
- 2026-06-23
AI Technical Summary
Existing delivery person identification methods rely on manual offline sample collection, which leads to inaccurate and untimely information, poor model training performance, and the fact that conventional machine learning algorithms such as decision trees are prone to overfitting and ignoring data correlations, resulting in low identification accuracy.
A deliveryman identification model based on user signaling data and internet access data is adopted. It is trained using a shrinking autoencoder deep neural network, combined with location information and base station data. Noise is suppressed by L1 and L2 regularization of the autoencoder, and L1 regularization is added for feature selection. The DBSCAN algorithm is used for data segmentation and clustering to optimize model parameters.
It improves the accuracy and efficiency of delivery personnel identification. By verifying location information, it can screen out more accurate delivery personnel, reduce the risk of model overfitting, and improve recognition accuracy.
Smart Images
Figure CN116910641B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of communication technology, specifically to a delivery person identification method, device, electronic device, and storage medium. Background Technology
[0002] Currently, delivery driver identification relies on manually collecting delivery driver samples offline and then using a model to learn the characteristics of these samples for prediction. However, this manual collection method carries risks of inaccurate and untimely information, and is also difficult, time-consuming, and labor-intensive. This results in poor model training performance, insufficient delivery driver sample data, and the model learning incorrect information, leading to inaccurate predictions.
[0003] The delivery driver identification model is based on user call data, such as call frequency, call duration, and call time distribution, and uses conventional machine learning decision trees for identification. However, conventional machine learning decision trees are prone to overfitting when making predictions, leading to inaccurate data output in real-world scenarios. Furthermore, existing machine learning decision trees tend to overlook the interrelationships between attributes in the dataset, such as the relationship between call frequency and call duration. In cases of imbalanced positive and negative samples, different decision rules can lead to different attribute selection biases when the decision tree performs attribute segmentation, resulting in low delivery driver identification accuracy. Summary of the Invention
[0004] This application provides a delivery person identification method, device, electronic device, and storage medium to solve the technical problem of low delivery person identification accuracy.
[0005] In a first aspect, embodiments of this application provide a delivery person identification method, including:
[0006] Based on the user's initial signaling data and internet access data, the first suspected delivery person was identified;
[0007] The identification information of the first suspected delivery person is input into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification model; the delivery person identification model is obtained by training a shrinking autoencoder deep neural network using suspected delivery person samples;
[0008] Based on the second signaling data of the first suspected delivery person and the location information of each delivery locker, a third suspected delivery person was identified;
[0009] Based on the second suspected delivery person and the third suspected delivery person, the delivery person is determined.
[0010] In one embodiment, determining the third suspected delivery person based on the second signaling data of the first suspected delivery person and the location information of each delivery locker includes:
[0011] The location information of the first suspected delivery person and the base station data of the base station that interacted with the first suspected delivery person are obtained based on the second signaling data.
[0012] Based on the location information of the first suspected delivery person, the base station data, and the location information of each delivery locker, determine the number of times the suspected delivery person stays at the base stations to which each delivery locker belongs within a set time period and the duration of their stay.
[0013] Determine the number of base stations whose dwell time falls within a first preset range;
[0014] If the number of base stations is within a second set range, then the first suspected delivery person is determined to be the third suspected delivery person.
[0015] In one embodiment, determining the delivery person based on the second suspected delivery person and the third suspected delivery person includes:
[0016] Determine the intersection of the second suspected delivery person and the third suspected delivery person;
[0017] The suspected delivery personnel within the intersection are identified as the delivery personnel.
[0018] In one embodiment, the deliveryman identification model is trained based on the following steps:
[0019] Identify suspected delivery personnel samples;
[0020] A shrinking autoencoder is used to map the suspected delivery person samples into low-dimensional data, and the data features of the low-dimensional data are extracted.
[0021] The deliveryman identification model is obtained by training a neural network based on the extracted data features.
[0022] Specifically, the loss function of the shrinking autoencoder includes a regularization penalty term to suppress data noise; L1 regularization and L2 regularization are added to the shrinking autoencoder to select data features.
[0023] In one embodiment, determining the first suspected delivery person based on the user's first signaling data and internet access data includes:
[0024] The first signaling data and the internet access data are segmented to obtain segmented data;
[0025] The segmented data is subjected to variable derivation to obtain derived data;
[0026] The derived data is clustered, and users whose similarity to the delivery person reaches a set value based on the clustering results are identified as the first suspected delivery person.
[0027] In one embodiment, before obtaining the number of times and duration of time the suspected delivery person stayed at the base stations belonging to each delivery locker within a set time period based on the location information of the first suspected delivery person and the location information of each delivery locker, the method further includes:
[0028] Determine the coverage area of each base station;
[0029] Based on the location information of each delivery locker and the coverage area of each base station, the delivery lockers covered by each base station are determined.
[0030] In one embodiment, after training the neural network based on the extracted data features to obtain the deliveryman identification model, the process includes:
[0031] Obtain the recognition result of the deliveryman recognition model;
[0032] The deliveryman identification model is optimized based on the identification results.
[0033] Secondly, embodiments of this application provide a delivery person identification device, including:
[0034] The first suspected delivery person identification module is used to identify the first suspected delivery person based on the user's first signaling data and internet access data;
[0035] The second suspected deliveryman identification module is used to input the identification information of the first suspected deliveryman into the deliveryman identification model to obtain the second suspected deliveryman output by the deliveryman identification model; the deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples;
[0036] The third suspected delivery person identification module is used to identify a third suspected delivery person based on the second signaling data of the first suspected delivery person and the location information of each delivery locker.
[0037] The delivery person identification module is used to identify a delivery person based on the second suspected delivery person and the third suspected delivery person.
[0038] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the delivery person identification method described in the first aspect.
[0039] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the deliveryman identification method of the first aspect.
[0040] The deliveryman identification method, device, electronic device, and storage medium provided in this application embodiment determine a first suspected deliveryman based on the user's first signaling data and internet access data; input the identification information of the first suspected deliveryman into a deliveryman identification model to obtain a second suspected deliveryman output by the deliveryman identification model; the deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples; a third suspected deliveryman is determined based on the second signaling data of the first suspected deliveryman and the location information of each delivery locker; and the deliveryman is determined based on the second and third suspected deliverymen. This application embodiment improves the accuracy of deliveryman identification by using a shrinking autoencoder deep neural network model to identify the second suspected deliveryman, obtaining the deliveryman's location information through the second signaling data, identifying the third suspected deliveryman based on the location information, and determining the deliveryman based on the second and third suspected deliverymen. Attached Figure Description
[0041] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 This is one of the flowcharts illustrating the delivery person identification method provided in the embodiments of this application;
[0043] Figure 2 This is a schematic diagram of the structure of the shrinking autoencoder provided in the embodiments of this application;
[0044] Figure 3 This is a schematic diagram of the structure of the CAE-DNN model provided in the embodiments of this application;
[0045] Figure 4 This is a second schematic flowchart of the delivery person identification method provided in the embodiments of this application;
[0046] Figure 5 This is a schematic diagram showing the location relationship between the base station and the delivery locker provided in this embodiment;
[0047] Figure 6 This is a schematic diagram of the delivery person identification device provided in the embodiments of this application;
[0048] Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0050] Figure 1 This is one of the flowcharts illustrating the delivery person identification method provided in this application. (Refer to...) Figure 1 This application provides a delivery person identification method, which may include:
[0051] S100 identifies the first suspected delivery person based on the user's initial signaling data and internet access data.
[0052] It should be noted that delivery personnel can be couriers, food delivery workers, transportation personnel, etc. This application embodiment uses couriers as an example to analyze and explain the delivery personnel identification method.
[0053] When mobile communication users perform actions such as powering on / off, making calls, sending text messages, updating their location, and switching base stations, they generate "signaling data." Therefore, the first signaling data includes the identifier of the base station interacting with the terminal, the interaction time, call data, text message data, the delivery person's location information, base station data, and the content of the information interaction. Among these, call data includes call frequency, call duration, and call time distribution.
[0054] Internet access data refers to data generated by users browsing the internet, such as usage data of user-end apps and web browsing data on user-end devices.
[0055] The system extracts the user's initial signaling data and basic information from the mobile database. It then searches online for business registration data to identify courier companies and uses their group numbers to find the courier numbers within the network; alternatively, it collects courier numbers from courier companies through web scraping. Deep Packet Inspection (DPI) is used to analyze the user's mobile application (APP) to obtain usage time and data usage data for courier-related apps. By matching the usage time and data usage data of courier-related apps between the user and the delivery person, potential delivery persons are identified. Finally, the system matches the potential delivery person's number with the user's basic information and initial signaling data. Data segmentation, variable derivation, and clustering are then performed on the potential delivery person's initial signaling data and internet access data to obtain the first potential delivery person.
[0056] S200, input the identification information of the first suspected delivery person into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification model; the delivery person identification model is obtained by training a shrinking autoencoder deep neural network using suspected delivery person samples.
[0057] It should be noted that the identification information includes phone numbers, ID card numbers, and other information. After inputting the identification information of the first suspected delivery person into the delivery person identification model, the model obtains the characteristic information of the first suspected delivery person based on this identification information, then analyzes the characteristic information, and based on the analysis results, filters out the second suspected delivery person from the first suspected delivery person and outputs the second suspected delivery person. The characteristic information consists of data representing the delivery person's call behavior and internet behavior, such as the number of users on the other end of the call, call duration, call frequency, and APP usage data.
[0058] For example, by analyzing call data, the delivery driver identification model can identify characteristics such as a higher number of users on the other end of the call compared to normal users, shorter average call duration, higher call frequency, and more concentrated call behavior. Analyzing internet data reveals that delivery drivers spend a significant amount of time using delivery-related apps and consume large amounts of data. Based on the analysis results of these characteristics, a second suspected delivery driver is identified.
[0059] The delivery driver identification model is obtained by training a shrinking autoencoder deep neural network using samples of suspected delivery drivers. A shrinking autoencoder is a type of unsupervised learning-based neural network that can map high-dimensional data to low-dimensional data. After training the shrinking autoencoder, its weights and biases are initialized, and the parameters of the last layer are randomly initialized. Features extracted by the feature extraction layer are then used to classify delivery driver users, thus obtaining the delivery driver identification model.
[0060] S300, based on the second signaling data of the first suspected delivery person and the location information of each delivery locker, determines the third suspected delivery person.
[0061] It's important to note that some delivery drivers don't deliver packages directly to recipients; instead, they place them in parcel lockers and then notify the recipient via a delivery app. Therefore, the call data for these delivery drivers is not significantly different from that of regular users.
[0062] To further improve the accuracy of delivery personnel identification, this application embodiment identifies the delivery personnel who notify the recipient to pick up the package via a courier APP by using second signaling data and the location information of the delivery locker.
[0063] The second signaling data can be extracted from the mobile database. The second signaling data includes the location information of the first suspected delivery person and the base station data of the base station that interacted with the first suspected delivery person; the location information of the express locker can be obtained from the Internet.
[0064] By analyzing the location information of the first suspected deliveryman, base station data, and the location information of each delivery locker, deliverymen who stay at a certain number of base stations belonging to delivery lockers within a set time period can be identified, and these deliverymen can be identified as the third suspected deliverymen.
[0065] S400, based on the second and third suspected delivery personnel, identifies the delivery person.
[0066] The delivery person is determined by taking the intersection of the second and third suspected delivery persons or by assigning weights to the second and third suspected delivery persons.
[0067] The deliveryman identification method provided in this application embodiment determines a first suspected deliveryman based on the user's first signaling data and internet access data; inputs the identification information of the first suspected deliveryman into a deliveryman identification model to obtain a second suspected deliveryman output by the deliveryman identification model; the deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples; a third suspected deliveryman is determined based on the second signaling data of the first suspected deliveryman and the location information of each delivery locker; and the deliveryman is determined based on the second and third suspected deliverymen. This application embodiment improves the accuracy of deliveryman identification by using a shrinking autoencoder deep neural network model to identify the second suspected deliveryman, obtaining the deliveryman's location information through the second signaling data, identifying the third suspected deliveryman based on the location information, and determining the deliveryman based on the second and third suspected deliverymen.
[0068] Based on the above embodiments, determining a third suspected delivery person based on the second signaling data of the first suspected delivery person and the location information of each delivery locker includes:
[0069] S310, Based on the second signaling data, obtain the location information of the first suspected delivery person and the base station data of the base station interacted with by the first suspected delivery person;
[0070] S320, based on the location information of the first suspected delivery person, base station data, and the location information of each delivery locker, determine the number of times the first suspected delivery person stays at the base station to which each delivery locker belongs and the duration of stay within a set time period;
[0071] S330, determine the number of base stations whose dwell time is within the first set range;
[0072] S340, if the number of base stations is within the second set range, then the first suspected delivery person is determined to be the third suspected delivery person.
[0073] The second signaling data includes the identifier of the base station interacting with the terminal, the interaction time, call data, SMS data, the location information of the first suspected delivery person, base station data, and the content of the information exchange. The location information of the first suspected delivery person includes spatial location and movement trajectory.
[0074] Optionally, the location information of the first suspected delivery person can be obtained through GPS positioning.
[0075] The base station data includes the base station number, the base station's longitude, the base station's latitude, the time when the first suspected delivery person entered the base station and the time when they left the base station, as well as the base station's name and type.
[0076] Based on the location information of the first suspected delivery person, the nearest delivery locker is determined. Then, based on the location information of the delivery locker, the corresponding base station covering the location of the delivery locker is obtained. Base station data is acquired through second signaling data.
[0077] Set a time period, for example, the working hours (08:00~20:00). By analyzing base station data and the location information and interaction time of the first suspected delivery person, the system can determine the number of times the first suspected delivery person stayed at each base station of the delivery locker and the duration of their stay within the set time period.
[0078] Determine the number of base stations whose dwell time falls within a first predetermined range. Set a first predetermined range for dwell time by counting the duration the delivery person spends at the base station associated with the delivery locker while placing the package inside. For example, if the first predetermined range is M1-M2 minutes, remove base stations with dwell times shorter than M1 minutes and longer than M2 minutes to obtain the number of base stations with dwell times within the M1-M2 minute range. A dwell time shorter than M1 minutes indicates that the suspected delivery person only briefly passed by the delivery locker and did not have time to place the package inside. A dwell time longer than M2 minutes indicates that the suspected delivery person lives in the vicinity long-term. Identify the base stations with dwell times within the M1-M2 minute range and count their number.
[0079] Identify the first suspected delivery person corresponding to the number of base stations within a second defined range. Set the second defined range based on the number of base stations the delivery person is stationed at while placing the package into the delivery locker. For example, if the second defined range is K1-K2, find the number of base stations within the K1-K2 range; the first suspected delivery person corresponding to this number of base stations is the third suspected delivery person.
[0080] This application embodiment identifies the delivery person by using location information and base station data to deliver the package directly to the delivery locker, and then notifies the recipient of the delivery person via a courier app, thereby improving the accuracy of the delivery person identification.
[0081] Based on the above embodiments, determining the delivery person based on the second and third suspected delivery persons includes:
[0082] S410, determine the intersection of the second suspected delivery person and the third suspected delivery person;
[0083] S420 identifies suspected delivery personnel within the intersection as actual delivery personnel.
[0084] It should be noted that the second suspected delivery person identified by the delivery person identification model includes both suspected delivery persons who deliver packages directly to the recipient and those who deliver packages directly to delivery lockers and then notify the recipient to pick up the package through a courier app. The third suspected delivery person mainly includes those who deliver packages directly to delivery lockers and then notify the recipient to pick up the package through a courier app. Users who belong to both the second and third suspected delivery persons have a relatively high probability of being a delivery person. The suspected delivery persons are output by taking the intersection of the second and third suspected delivery persons and sorting them.
[0085] By taking the intersection of the second and third suspected deliverymen, the process of initially screening deliverymen through the deliveryman identification model and then further verifying the deliverymen through location information was achieved.
[0086] The embodiments of this application improve the accuracy of delivery personnel identification by using a delivery personnel identification model for screening and location information verification.
[0087] Based on the above embodiments, the deliveryman identification model is trained using the following steps:
[0088] S210, identifying suspected delivery worker samples;
[0089] S220 uses a shrinking autoencoder to map suspected delivery personnel samples into low-dimensional data and extracts the data features of the low-dimensional data;
[0090] S230, The neural network is trained based on the extracted data features to obtain the deliveryman identification model;
[0091] Among them, the loss function of the shrinking autoencoder adds a regularization penalty term to suppress data noise; L1 regularization and L2 regularization are added to the shrinking autoencoder to select data features.
[0092] It should be noted that the suspected delivery person sample includes the suspected delivery person's sample identification information and sample characteristic data.
[0093] Extract first signaling data and user information of suspected delivery personnel from the mobile database. Search online for business registration data to match courier companies, and then find the courier numbers within the network using the courier company's group number; alternatively, use web scraping to collect courier numbers from courier companies. Analyze user terminal apps using DPI to obtain usage time and data usage data for user courier-related apps. Compile data on delivery personnel's usage time and data usage for courier-related apps. By matching the usage time and data usage data of user and delivery personnel's courier-related apps, identify suspected delivery personnel.
[0094] Suspected delivery personnel are assigned numbers to obtain their sample identification information. This sample identification information is then input into a shrinking autoencoder, which extracts sample feature information based on the identification information and maps it to low-dimensional data.
[0095] like Figure 2 As shown, the shrinking autoencoder mainly consists of an encoding network and a decoding network. Its hierarchical structure includes an input layer, hidden layers, and an output layer. The output of the hidden layer is the compressed feature. The encoding network encodes the input user feature number X to obtain the encoded feature h, and the decoding network decodes h into X'. Wherein,
[0096] X'=g(h=g(f(X))≈X;
[0097] Here, f() and g() are activation functions.
[0098] The contraction autoencoder (CAE) adds a regularization penalty term to the loss function, thereby reducing the sensitivity of the terminal to noisy data, suppressing data noise, and making the model more robust. The regularization penalty term includes the F-norm of the Jacobian matrix. The loss function is:
[0099]
[0100] Among them, J CAE (X) is the loss function, X is the input, and D is the loss function. n For the training set, Let λ be the mean square error function, and λ be a hyperparameter controlling the penalty intensity.
[0101] The regular expression penalty term is:
[0102]
[0103] in, This is a regular expression penalty term.
[0104] The discrete data in the sample feature information is dummy-coded into a data type. Then, the basic information, monthly call information, SMS information, data usage information, tariff information, and APP information parsed by DPI in the sample feature information are put into a shrinking autoencoder for processing to eliminate noise and outliers in the data itself before training, thereby reducing the impact of data defects on the model.
[0105] Adding L2 regularization to the shrinking autoencoder can reduce the model's complexity and prevent overfitting, which can lead to inaccurate predictions. In constructing the delivery driver identification model, the weights should be kept as small as possible. Constructing a model with relatively small parameters allows for good generalization ability.
[0106] like Figure 4 As shown, the deliveryman identification model in this application embodiment is a model combining a shrinking autoencoder and a fully connected neural network (Dynamic Neural Network, DNN) (CAE-DNN model).
[0107] Training a CAE is a continuous process of learning data features and extracting features, while also providing reasonable initial parameters for the DNN, such as... Figure 3 As shown. After the CAE completes training, the weights and biases of the CAE are initialized, and then the parameters of the last layer of the network are randomly initialized. The features extracted by the feature extraction layer are used to classify the deliveryman users. During the user classification process, the backpropagation algorithm can be used to fine-tune the network parameters to improve the generalization ability of the deliveryman recognition model. During the model training phase, the parameters of the entire network are optimized to speed up the network training. Due to the complexity of the DNN model itself, batch normalization layers (BN), dropout, and L2 regularization can be added during training to reduce the difficulty of the model and avoid overfitting.
[0108] L1 regularization selects model features. It produces a sparse weight matrix, i.e., a sparse model, which can be used for feature selection. In a sparse model, only a few features contribute to the model.
[0109] By introducing L1 and L2 regularization into the autoencoder, the autoencoder becomes a shrinking encoder capable of feature selection, solving the problems of overfitting and difficulty in feature selection in pre-trained models.
[0110] This application embodiment improves the accuracy of the courier identification model by adding a regularization penalty term to the model to suppress data noise; and by adding L1 regularization and L2 regularization to solve the problems of model overfitting and difficulty in feature selection.
[0111] Based on the above embodiments, the first suspected delivery person is identified based on the user's first signaling data and internet access data, including:
[0112] S110, performs data segmentation on the first signaling data and internet access data to obtain segmented data;
[0113] S120, perform variable derivation on the segmented data to obtain derived data;
[0114] S130, cluster the derived data, and based on the clustering results, identify users whose similarity to the delivery person reaches a set value as the first suspected delivery person.
[0115] An improved density-based clustering method for applications with noise (DBSCN) is used for data segmentation. The improved DBSCN algorithm incorporates the basic idea of density detection, which addresses the problem of the DBSCAN algorithm generating a large number of outliers when processing high-dimensional, non-uniformly dense data. The data segmentation process includes: dividing the first signaling data and internet access data into n nodes; querying the precision (Epsilon, Eps) neighborhood of the n nodes to obtain the coefficient of variation (cv) of each node; using the cv value to partition the data into segmented data; then, performing variable derivation on data such as the average monthly number of calls, average call duration, and average monthly number of base stations in the segmented data to obtain derived data; finally, using the improved DBSCAN algorithm to cluster the derived data to obtain the first suspected delivery person.
[0116] This application embodiment expands the scope of the first suspected delivery person and improves the efficiency of identifying delivery persons by segmenting, deriving variables, and clustering the user's first signaling data and Internet access data.
[0117] Based on the above embodiments, before obtaining the number of times a suspected delivery person stayed at each delivery locker's base station and the duration of their stay within a set time period, based on the location information of the first suspected delivery person and the location information of each delivery locker, the method further includes:
[0118] S321, determine the coverage area of each base station;
[0119] S322, based on the location information of each delivery locker and the coverage area of each base station, determines the delivery lockers covered by each base station.
[0120] like Figure 4 As shown, the coverage area of a base station is obtained through base station data. Optionally, the coverage area of a base station can be obtained through the network.
[0121] When the location of the delivery locker is within the coverage area of the base station, the delivery locker is a delivery locker covered by the base station.
[0122] like Figure 4 As shown, delivery lockers 5 and 6 are covered by base station 3, delivery locker 1 is covered by base station 1, and delivery locker 2 is covered by base station 4. Delivery locker 3 is not covered by any base station and is therefore not a delivery locker covered by any base station.
[0123] This application embodiment determines the delivery locker covered by the base station by using the coverage area of the base station and the location information of the delivery locker, thereby improving the accuracy of identifying a third suspected delivery person.
[0124] Based on the above embodiments, after training the neural network based on the extracted data features to obtain the deliveryman identification model, the process includes:
[0125] S231, Obtain the recognition results of the deliveryman recognition model;
[0126] S232, optimize the deliveryman identification model based on the identification results.
[0127] CAE-DNN model training is a continuous process of training, learning, and optimization. By obtaining the recognition results of the delivery person identification model, the relevant parameters of the delivery person identification model are optimized based on the recognition results. When the accuracy of the recognition results is lower than the preset value, the model parameters are adjusted and training continues until the accuracy of the recognition results reaches the preset value, thereby achieving the effect of optimizing the delivery person identification model.
[0128] The CAE-DNN model can learn from all features during model training and prediction. It learns data features through forward propagation and automatically adjusts model hyperparameters through backpropagation, eliminating the need for manual parameter tuning and further improving model accuracy.
[0129] The embodiments in this application optimize the deliveryman identification model based on the identification results, thereby improving the accuracy of deliveryman identification.
[0130] To further explain the deliveryman identification method provided in the embodiments of this application, the following embodiments are used as examples. Figure 6 Explanation:
[0131] (1) Data collection and preparation: Collect the user's first signaling data and Internet access data.
[0132] (2) Data segmentation, variable derivation, and clustering: The improved DBSCAN algorithm is used to segment, derive, and cluster the user's first signaling data and Internet access data to obtain the first suspected delivery person.
[0133] (3) Neural Network Recognition: The feature data of the first suspected delivery person is fed into a shrinking autoencoder for processing to eliminate noise and outliers in the feature data. The shrinking autoencoder maps high-dimensional data to low-dimensional data for user classification. The second suspected delivery person is then identified and output through a neural network.
[0134] (4) Based on location identification: the location information of the first suspected deliveryman, the base station data of the base station interacted by the first suspected deliveryman, and the location information of the delivery locker, obtain the base station corresponding to the location of the delivery locker, calculate the number of times each first suspected deliveryman passes through the base station to which the delivery locker belongs during the working period, remove base stations with a stay time of less than M1 minutes or greater than M2 minutes, obtain the number of base stations of the first suspected deliveryman staying at the delivery locker, and select users whose number of base stations staying at the delivery locker is in the range of K1-K2 as the third suspected deliveryman.
[0135] (5) Compile the list of delivery personnel: take the intersection of the second and third suspected delivery personnel.
[0136] (6) Result verification: Data verification is carried out through business means and manual verification: The summary list of delivery personnel obtained in step (5) is verified to confirm whether they are real delivery personnel. For example, the list of real delivery personnel is obtained through business means, and the delivery personnel in the summary list are manually verified to see if they are real delivery personnel.
[0137] This application embodiment identifies a second suspected delivery person using a delivery person identification model. By optimizing the autoencoder and adding regularization terms, noise and outliers in the original data are reduced, improving data quality and thus the model's accuracy. Simultaneously, this application embodiment identifies a third suspected delivery person using location data. Taking the intersection of the second and third suspected delivery persons further refines the delivery person identification, improving the accuracy of delivery person recognition.
[0138] The delivery person identification device provided in the embodiments of this application is described below. The delivery person identification device described below can be referred to in correspondence with the delivery person identification method described above. Reference Figure 6 , Figure 6 This is a schematic diagram of the structure of the delivery person identification device provided in an embodiment of this application. A delivery person identification device includes:
[0139] The first suspected delivery person identification module 601 is used to identify the first suspected delivery person based on the user's first signaling data and Internet access data;
[0140] The second suspected deliveryman identification module 602 is used to input the identification information of the first suspected deliveryman into the deliveryman identification model to obtain the second suspected deliveryman output by the deliveryman identification model; the deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples;
[0141] The third suspected delivery person identification module 603 is used to identify the third suspected delivery person based on the second signaling data of the first suspected delivery person and the location information of each delivery locker;
[0142] Deliveryman identification module 604 is used to identify a deliveryman based on a second suspected deliveryman and a third suspected deliveryman.
[0143] The deliveryman identification device provided in this application embodiment identifies a first suspected deliveryman based on the user's first signaling data and internet access data; inputs the identification information of the first suspected deliveryman into a deliveryman identification model to obtain a second suspected deliveryman output by the deliveryman identification model; the deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples; a third suspected deliveryman is identified based on the second signaling data of the first suspected deliveryman and the location information of each delivery locker; and the deliveryman is determined based on the second and third suspected deliverymen. This application embodiment improves the accuracy of deliveryman identification by identifying the second suspected deliveryman through a shrinking autoencoder deep neural network model, obtaining the deliveryman's location information through the second signaling data, identifying the third suspected deliveryman based on the location information, and determining the deliveryman based on the second and third suspected deliverymen.
[0144] In one embodiment, the third suspected delivery person determination module 603 is used to: obtain the location information of the first suspected delivery person and the base station data of the base station that interacts with the first suspected delivery person based on the second signaling data; determine the number of base stations to which a suspected delivery person stays and the duration of stay within a set time period based on the location information of the first suspected delivery person, the base station data, and the location information of each delivery locker; determine the number of base stations to which the suspected delivery person stays within a first set range; and if the number of base stations is within a second set range, determine the first suspected delivery person as the third suspected delivery person.
[0145] In one embodiment, the delivery person determination module 604 is used to: determine the intersection of a second suspected delivery person and a third suspected delivery person; and determine the suspected delivery person within the intersection as a delivery person.
[0146] In one embodiment, the deliveryman identification model is trained based on the following steps: identifying suspected deliveryman samples; using a shrinking autoencoder to map the suspected deliveryman samples into low-dimensional data and extracting data features from the low-dimensional data; training the neural network based on the extracted data features to obtain the deliveryman identification model; wherein, the loss function of the shrinking autoencoder adds a regularization penalty term to suppress data noise; and L1 regularization and L2 regularization are added to the shrinking autoencoder to select data features.
[0147] In one embodiment, the first suspected delivery person determination module 601 is used to: segment the first signaling data and internet access data to obtain segmented data; perform variable derivation on the segmented data to obtain derived data; cluster the derived data, and determine users whose similarity to the delivery person reaches a set value based on the clustering results as the first suspected delivery person.
[0148] In one embodiment, the third suspected deliveryman identification module 603 is further configured to: determine the coverage area of each base station; and determine the delivery lockers covered by each base station based on the location information of each delivery locker and the coverage area of each base station.
[0149] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 can call a computer program in the memory 730 to execute the steps of the delivery person identification method, such as including:
[0150] Based on the user's first signaling data and internet access data, a first suspected delivery person is identified; the identification information of the first suspected delivery person is input into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification model; the delivery person identification model is obtained by training a shrinking autoencoder deep neural network using suspected delivery person samples; based on the second signaling data of the first suspected delivery person and the location information of each delivery locker, a third suspected delivery person is identified; based on the second and third suspected delivery persons, the delivery person is identified.
[0151] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0152] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the delivery person identification method provided in the above embodiments, such as including:
[0153] Based on the user's first signaling data and internet access data, a first suspected delivery person is identified; the identification information of the first suspected delivery person is input into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification model; the delivery person identification model is obtained by training a shrinking autoencoder deep neural network using suspected delivery person samples; based on the second signaling data of the first suspected delivery person and the location information of each delivery locker, a third suspected delivery person is identified; based on the second and third suspected delivery persons, the delivery person is identified.
[0154] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to execute the steps of the delivery person identification method provided in the above embodiments, such as including:
[0155] Based on the user's first signaling data and internet access data, a first suspected delivery person is identified; the identification information of the first suspected delivery person is input into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification model; the delivery person identification model is obtained by training a shrinking autoencoder deep neural network using suspected delivery person samples; based on the second signaling data of the first suspected delivery person and the location information of each delivery locker, a third suspected delivery person is identified; based on the second and third suspected delivery persons, the delivery person is identified.
[0156] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).
[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0158] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A deliveryman identification method, characterized in that, include: Based on the user's initial signaling data and internet access data, the first suspected delivery person was identified; Input the identification information of the first suspected delivery person into the delivery person identification model to obtain the second suspected delivery person output by the delivery person identification; The deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples; Based on the second signaling data of the first suspected delivery person, the location information of the first suspected delivery person and the base station data of the base station that interacted with the first suspected delivery person are obtained; Based on the location information of the first suspected delivery person, the base station data, and the location information of each delivery locker, the number of times the suspected delivery person stayed at the base stations to which each delivery locker belongs and the duration of their stay were determined within a set time period. Determine the number of base stations whose dwell time falls within a first preset range; If the number of base stations is within a second preset range, then the first suspected delivery person is determined to be the third suspected delivery person; Based on the second suspected delivery person and the third suspected delivery person, the delivery person is determined.
2. The deliveryman identification method according to claim 1, characterized in that, The process of determining the delivery person based on the second suspected delivery person and the third suspected delivery person includes: Determine the intersection of the second suspected delivery person and the third suspected delivery person; The suspected delivery personnel within the intersection are identified as the delivery personnel.
3. The deliveryman identification method according to claim 1, characterized in that, The deliveryman identification model is trained based on the following steps: Identify suspected delivery personnel samples; A shrinking autoencoder is used to map the suspected delivery person samples into low-dimensional data, and the data features of the low-dimensional data are extracted. The deliveryman identification model is obtained by training a neural network based on the extracted data features. Specifically, the loss function of the shrinking autoencoder includes a regularization penalty term to suppress data noise; L1 regularization and L2 regularization are added to the shrinking autoencoder to select data features.
4. The deliveryman identification method according to claim 1, characterized in that, The process of identifying the first suspected delivery person based on the user's first signaling data and internet access data includes: The first signaling data and the internet access data are segmented to obtain segmented data; The segmented data is subjected to variable derivation to obtain derived data; The derived data is clustered, and users whose similarity to the delivery person reaches a set value based on the clustering results are identified as the first suspected delivery person.
5. The deliveryman identification method according to claim 1, characterized in that, Before obtaining the number of times and duration of time a suspected delivery person stayed at the base stations belonging to each delivery locker within a set time period, based on the location information of the first suspected delivery person and the location information of each delivery locker, the process further includes: Determine the coverage area of each base station; Based on the location information of each delivery locker and the coverage area of each base station, the delivery lockers covered by each base station are determined.
6. The deliveryman identification method according to claim 3, characterized in that, After training the neural network based on the extracted data features to obtain the deliveryman identification model, the process includes: Obtain the recognition result of the deliveryman recognition model; The deliveryman identification model is optimized based on the identification results.
7. A deliveryman identification device, characterized in that, include: The first suspected delivery person identification module is used to identify the first suspected delivery person based on the user's first signaling data and internet access data; The second suspected deliveryman identification module is used to input the identification information of the first suspected deliveryman into the deliveryman identification model and obtain the second suspected deliveryman output by the deliveryman identification. The deliveryman identification model is obtained by training a shrinking autoencoder deep neural network using suspected deliveryman samples; The third suspected deliveryman identification module is used to obtain the location information of the first suspected deliveryman and the base station data of the base station that interacts with the first suspected deliveryman based on the second signaling data of the first suspected deliveryman; and to determine the number of times the suspected deliveryman stays at the base station to which each delivery locker belongs and the duration of stay within a set time period based on the location information of the first suspected deliveryman, the base station data, and the location information of each delivery locker. The number of base stations whose dwell time is within a first set range is determined; if the number of base stations is within a second set range, then the first suspected delivery person is determined to be the third suspected delivery person. The delivery person identification module is used to identify a delivery person based on the second suspected delivery person and the third suspected delivery person.
8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the delivery person identification method according to any one of claims 1 to 6.
9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the delivery person identification method according to any one of claims 1 to 6.