Driver order receiving willingness training method and order sending matching strategy prediction method and device

Abstract

The invention relates to the technical field of driver order dispatching, and discloses an order receiving willingness training method of a driver and an order dispatching matching strategy prediction method and device.According to the order receiving willingness training method of the driver, and the order dispatching matching strategy prediction method and device of the driver, multiple historical influence factor data influencing order receiving of the driver are applied, so that multi-dimensional data are formed and input into an order receiving willingness prediction model for training; therefore, relevance among multiple different dimension data is established, orders are accurately distributed to drivers, and the travel experience of passengers is improved. Therefore, even if the drivers rank at the back, the drivers have enough high order receiving willingness and have opportunities to normally participate in order receiving, so that the phenomenon that no order can be received due to the fact that the drivers rank at the back is avoided, and the problems that the drivers rank at the front do not have the phenomenon of frequent order receiving and safe driving is influenced due to fatigue driving are solved; the order receiving willingness prediction result is generated through the order receiving willingness prediction model, so that the order receiving efficiency of the driver is improved, and the accuracy of order allocation for the driver is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of driver order dispatching, and in particular, to a method for training a driver's willingness to accept orders, a method for predicting an order dispatching matching strategy, and a device therefor. Background Art

[0002] In recent years, when passengers travel daily, taking a taxi has become increasingly frequent. Therefore, intelligent order dispatching for drivers has received more and more attention, and different Internet intelligent order dispatching methods have emerged continuously to meet the diversified travel needs of passengers. When a passenger hails a taxi, it is not difficult to find such an application scenario, that is, when a certain passenger places an order for a taxi through an Internet platform, the system will dispatch the order to different drivers through a certain intelligent order dispatching algorithm to achieve the matching of passengers and drivers.

[0003] In the related art, most common intelligent order dispatching algorithms simply perform weighted calculations based on data in a single dimension such as the number of passengers of a driver, the favorable comment rate, the reply quality, or the reply efficiency to obtain the ranking order of the drivers, and then allocate orders to the drivers. For example, when dispatching orders according to the intelligent order dispatching algorithm, if orders are often allocated to the drivers ranked at the top, the drivers ranked at the bottom often face no orders to grab. However, the willingness of the drivers ranked at the bottom to accept orders is very high, but they cannot receive orders normally due to the ranking order, resulting in a significant reduction in the enthusiasm of the drivers to accept orders. And the drivers ranked at the top often receive orders, which will also affect safe driving due to fatigue driving, and further affect the passenger satisfaction. Therefore, since the intelligent order dispatching method cannot examine the correlation between data in multiple different dimensions, it is impossible to accurately allocate orders to the drivers, thereby affecting the travel experience of passengers. Summary of the Invention

[0004] In view of this, the present invention provides a method for training a driver's willingness to accept orders, a method for predicting an order dispatching matching strategy, and a device therefor, so as to solve the problem that since the intelligent order dispatching method cannot examine the correlation between data in multiple different dimensions, it is impossible to accurately allocate orders to the drivers, thereby affecting the travel experience of passengers.

[0005] In a first aspect, the present invention provides a method for training a driver's willingness to accept orders, the method comprising:

[0006] Obtaining historical influence factor data affecting a driver's order acceptance, the historical influence factor data including various influence factor data;

[0007] Extract historical impact factor features from historical impact factor data, and input the historical impact factor features into the order acceptance willingness prediction model for training. After calculating the weight parameter values of the historical impact factor features through multiple layers of networks in the order acceptance willingness prediction model, generate the order acceptance willingness prediction results corresponding to the historical impact factor data, as well as the loss value between the order acceptance willingness prediction results and the true labels of the historical impact factor data. The loss value is used to update the weight parameter values.

[0008] The embodiments of the present disclosure apply various historical impact factor data affecting drivers' order acceptance, thereby forming multi-dimensional data. Input the multi-dimensional data into the order acceptance willingness prediction model for continuous training, and then establish the correlation between multiple different-dimensional data, so as to accurately allocate orders for drivers and improve the travel experience of passengers. Therefore, even if the driver's ranking is relatively low, but the driver's order acceptance willingness is high enough, there is a chance to normally participate in order acceptance, and thus there will be no phenomenon that the driver has no orders to accept due to a low ranking. Moreover, drivers with a high ranking will not have the phenomenon of frequently accepting orders, and the problem of affecting safe driving due to fatigue driving. By generating the order acceptance willingness prediction results of drivers through the order acceptance willingness prediction model, it is not only beneficial to improve the order acceptance efficiency of drivers, but also beneficial to improve the accuracy of order allocation for drivers.

[0009] In some alternative embodiments, the various impact factor data include: passenger praise rate, today's order acceptance volume, cumulative order acceptance volume, cumulative visit volume, cumulative income situation, on-duty duration on the current day, cumulative working duration, cumulative refusal order times, and cumulative itinerary on the current day.

[0010] In some alternative embodiments, input the historical impact factor features into the order acceptance willingness prediction model for training. After calculating the weight parameter values of the historical impact factor features through multiple layers of networks in the order acceptance willingness prediction model, generate the order acceptance willingness prediction results of the driver, including:

[0011] Encode the historical impact factor features using the feature encoding network in the order acceptance willingness prediction model, and the feature encoding network is an XLNet encoding network;

[0012] Decode the historical impact factor features using the feature decoding network in the order acceptance willingness prediction model to generate the order acceptance willingness prediction results of the driver, and the feature encoding network is a conditional random field network.

[0013] In some alternative embodiments, obtaining the historical impact factor data affecting drivers' order acceptance includes:

[0014] Collect the original sample data of the historical impact factor data affecting drivers' order acceptance, and the original sample data includes various original impact factor data;

[0015] Filter and clean the original sample data;

[0016] Store the original sample data after filtering and cleaning;

[0017] Select a part of the data from the original sample data after filtering and cleaning as the training sample set of the order acceptance willingness prediction model, and the training sample set is the historical influencing factor data affecting the driver's order acceptance.

[0018] In some alternative embodiments, select another part of the data from the original sample data after filtering and cleaning as the test sample set of the order acceptance willingness prediction model, and the test sample set is used to test the accuracy of the order acceptance willingness prediction model.

[0019] In some alternative embodiments, the multi-layer network of the order acceptance willingness prediction model is respectively: an input layer, a hidden layer, an activation layer, and an output layer. Among them, the input layer generates a feature matrix according to the historical influencing factor features, the hidden layer generates a first weight matrix according to the feature matrix, the activation layer is used to generate an activation function according to the output parameters and input parameters of the order acceptance willingness prediction model, and the output layer generates a second weight matrix according to the activation function and the first weight matrix.

[0020] According to a second aspect, an embodiment of the present disclosure provides a method for predicting a dispatching matching strategy for a driver, the method including:

[0021] Obtain the target influencing factor data affecting the driver's order acceptance, and the target influencing factor data includes various influencing factor data;

[0022] Obtain the current satisfaction level of the target passenger with the driver, and the current satisfaction level of the target passenger with the driver is determined according to the positive rate of the target passenger with respect to the driver;

[0023] Extract the target influencing factor features from the target influencing factor data, and input the target influencing factor features into the above-mentioned order acceptance willingness prediction model for prediction. After the target influencing factor features pass through the multi-layer network to calculate the weight parameter values of the target influencing factor features in the order acceptance willingness prediction model, obtain the order acceptance willingness prediction result corresponding to the target influencing factor data;

[0024] Generate a dispatching matching strategy between the driver and the target passenger according to the order acceptance willingness prediction result and the current satisfaction level of the target passenger with the driver.

[0025] In an embodiment of the present disclosure, historical influencing factor data affecting a driver's order acceptance is input into a trained order acceptance willingness prediction model to obtain an order acceptance willingness prediction result corresponding to the target influencing factor data. According to the order acceptance willingness prediction result and the current satisfaction level of the target passenger with respect to the driver, a suitable order acceptance willingness is matched for the target passengers in different satisfaction groups, and further a dispatching matching strategy between the driver and the target passenger is determined, so as to achieve a win-win situation for both the passenger side and the driver side, and retain as many active users as possible for the dispatching platform.

[0026] In a third aspect, the present invention provides a device for training a driver's order acceptance willingness, the device comprising:

[0027] A data acquisition module, configured to acquire historical influencing factor data affecting a driver's order acceptance, where the historical influencing factor data includes various influencing factor data;

[0028] A feature extraction module, configured to extract historical influencing factor features from the historical influencing factor data, and input the historical influencing factor features into the training of the order acceptance willingness prediction model. After the historical influencing factor features pass through multiple layers of networks in the order acceptance willingness prediction model to calculate the weight parameter values of the historical influencing factor features, an order acceptance willingness prediction result corresponding to the historical influencing factor data is generated, as well as a loss value between the order acceptance willingness prediction result and the true label of the historical influencing factor data, and the loss value is used to update the weight parameter values.

[0029] In a fourth aspect, the present invention provides a computer device, comprising:

[0030] A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the method for training a driver's order acceptance willingness in the first aspect or any embodiment of the first aspect, or the method for predicting a driver's dispatching matching strategy in the second aspect or any embodiment of the second aspect.

[0031] In a fifth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the method for training a driver's order acceptance willingness in the first aspect or any embodiment of the first aspect, or the method for predicting a driver's dispatching matching strategy in the second aspect or any embodiment of the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0033] Figure 1 It is a schematic flowchart of a method for training a driver's willingness to accept orders according to an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of a neural network principle according to an embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of a step function according to an embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of a Sigmoid function according to an embodiment of the present invention;

[0037] Figure 5 It is a schematic diagram of a ReLU function according to an embodiment of the present invention;

[0038] Figure 6 It is a flowchart of a method for predicting a driver assignment matching strategy according to an embodiment of the present invention;

[0039] Figure 7 It is a schematic diagram of matching a driver's willingness to accept orders according to an embodiment of the present invention;

[0040] Figure 8 It is a simple schematic diagram of generating a driver assignment matching strategy according to an embodiment of the present invention;

[0041] Figure 9 It is a structural block diagram of a device for training a driver's willingness to accept orders according to an embodiment of the present invention;

[0042] Figure 10 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Specific Embodiments

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0044] In this embodiment, a method for training a driver's willingness to accept orders is provided, which can be used in the above-mentioned computer devices, such as mobile phones, tablet computers, desktop computers, portable notebooks, servers, etc. Figure 1 It is a flowchart of the method for training a driver's willingness to accept orders according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps:

[0045] According to an embodiment of the present invention, an embodiment of a method for training a driver's willingness to accept orders is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0046] In this embodiment, a method for training a driver's willingness to accept orders is provided, which can be used in the method for training a driver's willingness to accept orders provided in this embodiment, and can be used in computer devices, such as mobile phones, tablet computers, desktop computers, portable notebooks, servers, mobile terminals, etc. Figure 1 It is a flowchart of the method for training a driver's willingness to accept orders according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps:

[0047] Step S101, obtain historical impact factor data that affects a driver's willingness to accept orders. The historical impact factor data includes various impact factor data.

[0048] Specifically, a driver refers to one or more drivers who express their willingness to accept orders. The historical impact factor data represents some variable data that affects the order acceptance frequency of this driver. The historical impact factor data that affects a driver's willingness to accept orders can be a batch of sample data. For example, if the historical impact factor data has a high impact on a driver's willingness to accept orders, the driver's order acceptance frequency is low; if the driver impact factor data has a low impact on a driver's willingness to accept orders, the driver's order acceptance frequency is high.

[0049] In the traditional method of intelligent order dispatching for passengers, the intelligent order dispatching algorithm is used to perform simple weighted calculations based on data in a single dimension such as the driver's passenger volume, positive review rate, reply quality, or reply efficiency, to obtain the ranking order of the drivers, and then to assign orders to the drivers. Since the data in a single dimension affects the accuracy of the driver ranking and cannot examine the correlation between multiple different dimensions of data, it leads to the inability to accurately assign orders to the drivers, thereby affecting the travel experience of passengers.

[0050] In view of this, the historical impact factor data obtained in the embodiments of the present disclosure includes various impact factor data, and further combines these various impact factor data to accurately analyze the driver's willingness to accept orders.

[0051] In a specific example, the multiple impact factor data includes: passenger positive review rate, number of orders received today, cumulative number of orders received, cumulative number of visits, cumulative income situation, on-duty duration today, cumulative working hours, cumulative number of order rejections, and cumulative itinerary today.

[0052] Specifically, as shown in Table 1 below, it is a table of historical impact factor data affecting drivers' order receiving.

[0053] Table 1

[0054] Variable Name Description favorableRate Passenger Positive Comment Rate todayReceiveOrdersNum Number of Orders Received Today receiveOrdersNum Accumulative Number of Orders Received visitorNum Accumulative Number of Visits accumulatedIncome Accumulative Income Situation todayWorkHourNum Length of Time on Duty Today workTotalHourNum Accumulative Working Hours refuseOrdersNum Accumulative Number of Refused Orders totalDistanceNum Accumulative Trip Today

[0055] In some alternative embodiments, the above step S102 of obtaining historical impact factor data affecting drivers' order receiving includes:

[0056] Step a1: Collect the original sample data of historical impact factor data affecting drivers' order receiving. The original sample data includes multiple types of original impact factor data.

[0057] Specifically, for example, collect the original sample data of historical impact factor data of at least one or more drivers who participated in order receiving during a historical period. Specifically, collect the original sample data of multiple variable data in Table 1 above. In this step a1, it is equivalent to collecting user data on the driver side. For example, the driver side collects the original sample data of historical impact factor data related to the driver through the order receiving platform.

[0058] Step a2: Perform filtering and cleaning processing on the original sample data.

[0059] Specifically, for example, further process the original sample data in the above example. For example, remove illegal data through filtering and cleaning, including but not limited to removing expired data, test data, incomplete data, duplicate data, etc.

[0060] Step a3: Select a part of the data from the original sample data after filtering and cleaning processing as the training sample set for the order receiving willingness prediction model. The training sample set is the historical impact factor data affecting drivers' order receiving.

[0061] Step a4: Store the original sample data after filtering and cleaning processing.

[0062] Specifically, store the original sample data after filtering and cleaning processing in the local memory to prevent data loss.

[0063] Step a5: Select a part of the data from the original sample data after filtering and cleaning processing as the training sample set for the order receiving willingness prediction model. The training sample set is the historical impact factor data affecting drivers' order receiving.

[0064] Specifically, for example, a part of the data (80% of the original sample data) is selected from the original sample data after the filtering and cleaning process as the training sample set of the order acceptance willingness prediction model. For each training data in the training sample set, the true order acceptance willingness of the corresponding driver is manually marked.

[0065] In the embodiments of the present disclosure, the original sample data of the historical influencing factor data that affects the driver's order acceptance is filtered and cleaned, thereby ensuring that the finally obtained historical influencing factor data that affects the driver's order acceptance is more accurate.

[0066] Step S102: Extract historical influencing factor features from the historical influencing factor data, and input the historical influencing factor features into the training of the order acceptance willingness prediction model. After the historical influencing factor features pass through multiple layers of networks in the order acceptance willingness prediction model to calculate the weight parameter values of the historical influencing factor features, an order acceptance willingness prediction result corresponding to the historical influencing factor data is generated, as well as a loss value between the order acceptance willingness prediction result and the true label of the historical influencing factor data. The loss value is used to update the weight parameter values.

[0067] Specifically, the order acceptance willingness prediction model can be a multi-layer feedforward network trained by error backpropagation, also known as a BPNN network model. Its basic idea is to use the gradient descent algorithm on the basis of an artificial neural network (ANN) to minimize the mean square error between the actual output value and the expected output value of the network. As Figure 2 shown, it is a schematic diagram of a neural network.

[0068] In a specific example, the multi-layer network of the order acceptance willingness prediction model is respectively: an input layer, a hidden layer, an activation layer, and an output layer. Among them, the input layer generates a feature matrix according to the historical influencing factor features, the hidden layer generates a first weight matrix according to the feature matrix, the activation layer is used to generate an activation function according to the output parameters and input parameters of the order acceptance willingness prediction model, and the output layer generates a second weight matrix according to the activation function and the first weight matrix.

[0069] In Figure 2 , the element dimension of the input layer is related to the feature information of the input quantity (extracting historical influencing factor features from historical influencing factor data). If the number of selected feature variables is 10, the input layer is a 1*10 feature matrix. The connection relationships between the input layer and the hidden layer are W1 and b1 respectively, and the dimension of the hidden layer can be freely set.

[0070] In a specific example, the first weight matrix is calculated by the following formula:

[0071] H = X * W1 + b1

[0072] Among them, H is the first weight matrix, X is the feature matrix generated according to the historical impact factor features, W1 is the weight matrix between the input layer and the hidden layer, and b1 is the bias matrix between the input layer and the hidden layer.

[0073] Activation layer

[0074] Since the final prediction result y of the driver's willingness to accept orders and the selected variable X (the feature matrix generated according to the historical impact factor features) are not linearly related, an activation layer needs to be introduced. The commonly used activation functions are the step function, the Sigmoid function, and the ReLU function.

[0075] As Figure 3 shown, it is a schematic diagram of the step function. For the step function, when the input is less than or equal to 0, the output is 0; when the input is greater than 0, the output is 1.

[0076] As Figure 4 shown, it is a schematic diagram of the Sigmoid function. For the Sigmoid function, when the input approaches positive infinity / negative infinity, the output approaches 1 / 0 infinitely.

[0077] As Figure 5 shown, it is a schematic diagram of the ReLU function. For the ReLU function, when the input is less than 0, the output is 0; when the input is greater than 0, the output is equal to the input. Among them, the output value of the step function is discontinuous and has only two states, 0 and 1; when the absolute value of X in the Sigmoid function is relatively large, the slope of the curve changes very little, that is, the gradient disappearance phenomenon will occur.

[0078] Therefore, after passing through the activation function, the first weight matrix becomes the second weight matrix H’, which can also be called the activation matrix. The output layer generates the second weight matrix according to the activation function and the first weight matrix, which is expressed by the following formula:

[0079] Y′ = H’ * W2 + b2

[0080] Among them, H’ is the second weight matrix, W2 is the weight matrix between the output layer and the hidden layer, b2 is the bias matrix between the output layer and the hidden layer, and Y′ is the output parameter output by the output layer.

[0081] In a specific example, the historical impact factor features are input into the willingness to accept orders prediction model for training. After the historical impact factor features pass through multiple network layers to calculate the weight parameter values of the historical impact factor features in the willingness to accept orders prediction model, the prediction result of the driver's willingness to accept orders is generated, including:

[0082] Step b1, using the feature encoding network in the willingness to accept orders prediction model to encode the historical impact factor features, and the feature encoding network is the XLNet encoding network.

[0083] Furthermore, by calculating the loss value between Y' and the true label Y of the historical influence factor data and backpropagating it, the weight matrix W and the bias vector b are optimized, and continuous iteration is performed to make the residual smaller and smaller. Therefore, the specific steps of the above-mentioned order acceptance willingness prediction model are as follows:

[0084] (1) Determine the number of layers of the backpropagation neural network structure, including the input layer, the hidden layer, and the output layer.

[0085] (2) The process from the input layer to the hidden layer is the encoding process, and the specific form is shown in the following formula:

[0086]

[0087] Among them, is the historical influence factor feature, H is the hidden layer encoding of, σ is the hidden layer activation function, and W and b are the connection weights and bias vectors from the input layer to the hidden layer respectively. For the historical influence factor features of each driver Table 1 is used as the feature variable for evaluating the driver's order acceptance willingness, that is, is a 1*10 feature matrix, and the activation function σ selects the ReLU activation function.

[0088] Step b2, use the feature decoding network in the order acceptance willingness prediction model to decode the historical influence factor features to generate the order acceptance willingness prediction result of the driver, and the feature encoding network is a conditional random field network.

[0089] The process from the hidden layer to the output layer is the decoding process, and the specific form is:

[0090] R = W'H + b'

[0091] Among them, R is the actual output vector generated by the output layer after inputting , W' is the weight matrix between the output layer and the hidden layer, and b' is the bias vector between the output layer and the hidden layer.

[0092] Furthermore, randomly initialize the network parameters θ = {W, W', b, b'}, and use the gradient descent algorithm to iteratively train the neural network model, and update the parameter θ each time; the loss value calculation formula in this embodiment of the present disclosure is expressed by the following formula:

[0093]

[0094] Among them, Loss is the loss function, n is the number of training samples, y is the order acceptance willingness of the label data, and f(x i ) is the order acceptance willingness prediction result predicted by the order acceptance willingness prediction model.

[0095] Iteratively update the loss value further until the loss value meets the accuracy requirement, at which point it is considered that the prediction model training for the order acceptance willingness is completed.

[0096] In a specific example, another part of the data is selected from the original sample data after filtering and cleaning as the test sample set for the order acceptance willingness prediction model, and the test sample set is used to test the accuracy of the order acceptance willingness prediction model.

[0097] For example, a part of the data (20% of the other part of the data) is selected from the original sample data after filtering and cleaning as the test sample set for the order acceptance willingness prediction model to test the accuracy of the order acceptance willingness prediction model. The sample test data in the test sample set is input into the trained order acceptance willingness prediction model for training to determine the accuracy of the order acceptance willingness prediction model. If the accuracy meets the accuracy requirement, the training of the order acceptance willingness prediction model is correct. If the accuracy does not meet the accuracy requirement, it is necessary to repeatedly adjust different hidden layer structures, such as modifying the hidden layer dimension, etc., and retrain the order acceptance willingness prediction model.

[0098] In summary, the embodiments of the present disclosure apply historical influence factor data of various factors affecting drivers' order acceptance, thereby forming multi-dimensional data. The multi-dimensional data is input into the order acceptance willingness prediction model for continuous training, and then the correlation between multiple different-dimensional data is established to accurately allocate orders for drivers and improve the travel experience of passengers. Therefore, even if a driver's ranking is relatively low, as long as the driver's order acceptance willingness is high enough, they have the opportunity to normally participate in order acceptance, and thus there will be no phenomenon that drivers cannot receive orders due to their low ranking. Moreover, drivers with a high ranking will not have the problem of frequently receiving orders and affecting safe driving due to fatigue driving. By generating the prediction result of the driver's order acceptance willingness through the order acceptance willingness prediction model, it is not only beneficial to improve the driver's order acceptance efficiency but also beneficial to improve the accuracy of order allocation for drivers.

[0099] In this embodiment, a prediction method for a driver's order assignment matching strategy is provided, which can be used in computer devices such as mobile phones, tablet computers, desktop computers, portable notebooks, servers, mobile terminals, etc. Figure 6 It is a flowchart of the prediction method for a driver's order assignment matching strategy according to an embodiment of the present invention, as Figure 6 shown, and the process includes the following steps:

[0100] Step S601: Obtain target influence factor data affecting drivers' order acceptance, where the target influence factor data includes various influence factor data.

[0101] Specifically, the driver is one or more drivers participating in receiving orders. The target impact factor data is the impact factor data that the driver is about to be predicted. Since the regions where the drivers are located are different, the corresponding target impact factor data of drivers in different regions will also vary. For example, assuming that the region where the target driver is located is the Beijing region, then the driver receives orders in the Beijing region, and the target impact factor data is the relevant data of the Beijing region; assuming that the region where the target driver is located is the Tianjin region, then the driver receives orders in the Tianjin region, and the target impact factor data is the relevant data of the Tianjin region.

[0102] In a specific example, the various impact factor data that affect the target impact factor data of the driver's order receiving also include: passenger praise rate, today's order receiving volume, cumulative order receiving volume, cumulative visit volume, cumulative income situation, current on-duty duration on the day, cumulative working duration, cumulative order rejection times, and cumulative itinerary on the day.

[0103] For the specific implementation manner of obtaining the target impact factor data that affects the driver's order receiving, refer to the above steps a1 - a5, which will not be elaborated here.

[0104] Step S602: Obtain the current satisfaction level of the target passenger with respect to the driver. The current satisfaction level of the target passenger with respect to the driver is determined based on the praise rate of the target passenger for the driver.

[0105] Specifically, based on the praise rate given by the target passenger in the historical orders, a simple group division of the target passengers can be carried out. Target passengers with a higher number of praise rate times may correspond to a higher current satisfaction level, that is, the service of the driver by this target passenger is easily satisfied. On the contrary, target passengers with a lower number of praise rate times will correspondingly have a lower current satisfaction level, that is, the service requirements of the target passenger for the driver are relatively strict. Assume that the target passengers are roughly divided into three groups {c1, c2, c3} according to the ascending order of the current satisfaction level with respect to the driver.

[0106] Step S603: Extract the target impact factor features from the target impact factor data, and input the target impact factor features into the order receiving willingness prediction model in the above embodiment for prediction. After the target impact factor features are calculated through multiple layers of networks in the order receiving willingness prediction model to obtain the weight parameter values of the target impact factor features, the order receiving willingness prediction result corresponding to the target impact factor data is obtained.

[0107] Specifically, for the prediction method of the target impact factor data, refer to the above step S102, which will not be elaborated here.

[0108] Step S604: Generate a dispatching matching strategy between the driver and the target passenger according to the order receiving willingness prediction result of the driver and the current satisfaction level of the target passenger with respect to the driver.

[0109] Specifically, during the process of dispatching orders to drivers, the order acceptance willingness prediction model is used to predict the order acceptance willingness prediction result corresponding to the target influencing factor data, find the prediction result with the highest order acceptance willingness, and then further query the driver corresponding to the prediction result with the highest order acceptance willingness to generate the order dispatching matching strategy between the driver and the target passenger. As Figure 7 shown, it is a schematic diagram of the order acceptance willingness of matching drivers. In Figure 7 , considering both the driver and the target passenger comprehensively, the final intelligent order dispatching strategy is obtained. In Figure 7 , the order acceptance willingness of the driver increases successively, and the satisfaction of the target passenger decreases successively. For the driver with the highest order acceptance willingness, it corresponds to matching the target passengers who are relatively easy to be satisfied and the target passengers who are least likely to be satisfied; for the driver with good order acceptance willingness, it corresponds to matching the target passengers who are relatively easy to be satisfied and the target passengers who are most likely to be satisfied; for the driver with qualified order acceptance willingness, it corresponds to matching the target passengers who are relatively easy to be satisfied and the target passengers who are most likely to be satisfied; for the driver with the lowest order acceptance willingness, it corresponds to matching the target passengers who are relatively easy to be satisfied and the target passengers who are most likely to be satisfied.

[0110] For example, for the target passengers in the c3 group, that is, the target passengers who are least likely to be satisfied, it should be considered to give priority to dispatching orders to the drivers corresponding to the data with the highest order acceptance willingness.

[0111] For the target passengers in the c2 group, that is, the target passengers who are relatively easy to be satisfied, it can be considered to give priority to dispatching orders to the drivers corresponding to the highest and good levels of the order acceptance willingness data.

[0112] For the target passengers in the c1 group, that is, the target passengers who are most likely to be satisfied, it can be considered to give priority to dispatching orders to the drivers corresponding to the good and qualified levels of the order acceptance willingness data.

[0113] For the drivers corresponding to the lowest order acceptance willingness data, that is, the drivers classified as the poor level, a certain number of them should be randomly selected from the c1 or c2 target passenger groups for dispatching orders, and the target passenger orders in the c3 group cannot be dispatched, so as to avoid the drivers being unable to receive orders normally due to their low ranking, resulting in a significant reduction in the drivers' order acceptance enthusiasm.

[0114] Therefore, the embodiments of the present disclosure use the historical influencing factor data affecting the drivers' order acceptance to input the trained order acceptance willingness prediction model to obtain the order acceptance willingness prediction result corresponding to the target influencing factor data, and according to the order acceptance willingness prediction result and the current satisfaction level of the target passenger with the driver, for the target passengers in different satisfaction groups, match them with appropriate order acceptance willingness, and further determine the order dispatching matching strategy between the driver and the target passenger, so as to achieve a win-win situation for both the passenger side and the driver side, and retain as many active users as possible for the order dispatching platform.

[0115] AsFigure 8 As shown, it is a simple schematic diagram for generating the order assignment matching strategy for drivers. In Figure 8 , first, basic data construction is carried out. When carrying out basic data construction, the following steps are included in sequence: collecting user data, filtering and cleaning data, and locally storing data. On this basis, based on the BPNN model (order acceptance willingness prediction model), the order acceptance willingness corresponding to the data (target influencing factor data) on the driver side is predicted, and then the current satisfaction level of the target passenger with the driver is determined according to the positive comment rate of the target passenger on the driver (grouping the target passengers according to the frequency of good reviews), and finally, an intelligent order assignment strategy is generated.

[0116] In this embodiment, an order acceptance willingness training device for drivers is also provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0117] This embodiment provides an order acceptance willingness training device for drivers. As Figure 9 shown, it includes:

[0118] A data acquisition module 901, which is used to acquire historical influencing factor data that affects drivers' order acceptance. The historical influencing factor data includes various influencing factor data;

[0119] A feature extraction module 902, which is used to extract historical influencing factor features from the historical influencing factor data and input the historical influencing factor features into the training of the order acceptance willingness prediction model. After the historical influencing factor features pass through multiple layers of networks in the order acceptance willingness prediction model to calculate the weight parameter values of the historical influencing factor features, an order acceptance willingness prediction result corresponding to the historical influencing factor data is generated, as well as a loss value between the order acceptance willingness prediction result and the true label of the historical influencing factor data. The loss value is used to update the weight parameter values.

[0120] In some optional implementation manners, the various influencing factor data includes: passenger positive comment rate, number of orders received today, cumulative number of orders received, cumulative number of visits, cumulative income situation, on-duty duration today, cumulative working hours, cumulative number of order rejections, and cumulative itinerary today.

[0121] In some optional implementation manners, the feature extraction module 902 includes:

[0122] An encoding sub-module, which is used to encode the historical influencing factor features by using the feature encoding network in the order acceptance willingness prediction model. The feature encoding network is an XLNet encoding network;

[0123] A decoding sub-module, configured to decode historical impact factor features by using a feature decoding network in an order-taking willingness prediction model to generate an order-taking willingness prediction result for a driver, where the feature encoding network is a conditional random field network.

[0124] In some alternative embodiments, the data acquisition module 901 includes:

[0125] A data collection sub-module, configured to collect original sample data of historical impact factor data affecting a driver's order-taking. The original sample data includes various original impact factor data;

[0126] A data processing sub-module, configured to perform filtering and cleaning processing on the original sample data;

[0127] A data storage sub-module, configured to store the original sample data after the filtering and cleaning processing;

[0128] A first selection sub-module, configured to select a part of the data from the original sample data after the filtering and cleaning processing as a training sample set for the order-taking willingness prediction model. The training sample set is historical impact factor data affecting a driver's order-taking.

[0129] In some alternative embodiments, it further includes: a data selection module, configured to select another part of the data from the original sample data after the filtering and cleaning processing as a test sample set for the order-taking willingness prediction model. The test sample set is used to test the accuracy of the order-taking willingness prediction model.

[0130] In some alternative embodiments, the multi-layer network of the order-taking willingness prediction model is respectively: an input layer, a hidden layer, an activation layer, and an output layer. Among them, the input layer generates a feature matrix according to historical impact factor features, the hidden layer generates a first weight matrix according to the feature matrix, the activation layer is configured to generate an activation function according to the output parameters and input parameters of the order-taking willingness prediction model, and the output layer generates a second weight matrix according to the activation function and the first weight matrix.

[0131] An embodiment of the present disclosure provides a dispatching matching strategy prediction device for a driver. The device includes:

[0132] A first acquisition module, configured to acquire target impact factor data affecting a driver's order-taking. The target impact factor data includes various impact factor data;

[0133] A second acquisition module, configured to acquire the current satisfaction level of a target passenger with respect to a driver. The current satisfaction level of the target passenger with respect to the driver is determined according to the positive review rate of the target passenger with respect to the driver;

[0134] A feature extraction module is configured to extract target impact factor features from target impact factor data, and input the target impact factor features into the above-mentioned order acceptance willingness prediction model for prediction. After the target impact factor features pass through multiple layers of networks in the order acceptance willingness prediction model to calculate the weight parameter values of the target impact factor features, an order acceptance willingness prediction result corresponding to the target impact factor data is obtained;

[0135] A strategy generation module is configured to generate a dispatching matching strategy between the driver and the target passenger according to the order acceptance willingness prediction result and the current satisfaction level of the target passenger with the driver.

[0136] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.

[0137] The order acceptance willingness training device for the driver in this embodiment, or the dispatching matching strategy prediction device for the driver is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0138] The embodiment of the present invention further provides a computer device having the above-mentioned order acceptance willingness training device for the driver, or the dispatching matching strategy prediction device for the driver.

[0139] Please refer to Figure 10 , Figure 10 which is a schematic structural diagram of a computer device provided by an optional embodiment of the present invention. As shown in Figure 10 , the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 10 In

[0140] The processor 10 may be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device may be a complex programmable logic device, a field-programmable gate array, a generic array logic, or any combination thereof.

[0141] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0142] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may further include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely disposed relative to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0143] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may further include a combination of the above types of memories.

[0144] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0145] The embodiments of the present invention further provide a computer-readable storage medium. The method according to the embodiments of the present invention may be implemented in hardware, firmware, or may be implemented as computer code recorded on a storage medium, or may be implemented as computer code originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and to be stored in a local storage medium, so that the method described herein may be stored in such software processed on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium may be a magnetic disk, an optical disc, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium may further include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0146] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A training method for a driver's willingness to accept orders, characterized in that, The method includes: Obtaining historical impact factor data that affects a driver's order acceptance, where the historical impact factor data includes various impact factor data; Extracting historical impact factor features from the historical impact factor data and inputting the historical impact factor features into the order acceptance willingness prediction model for training. After the historical impact factor features are calculated for the weight parameter values of the historical impact factor features through multiple layers of networks in the order acceptance willingness prediction model, an order acceptance willingness prediction result corresponding to the historical impact factor data is generated, as well as a loss value between the order acceptance willingness prediction result and the true label of the historical impact factor data, and the loss value is used to update the weight parameter values.

2. The driver's order acceptance willingness training method according to claim 1, characterized in that The various impact factor data includes: passenger positive review rate, current day's order acceptance volume, cumulative order acceptance volume, cumulative visit volume, cumulative income situation, current day's on-duty duration, cumulative working duration, cumulative order rejection times, current day's cumulative itinerary.

3. The method for training the order acceptance willingness of a driver according to claim 1, wherein Inputting the historical impact factor features into the order acceptance willingness prediction model for training. After the historical impact factor features are calculated for the weight parameter values of the historical impact factor features through multiple layers of networks in the order acceptance willingness prediction model, the order acceptance willingness prediction result of the driver is generated, including: Encoding the historical impact factor features using the feature encoding network in the order acceptance willingness prediction model, where the feature encoding network is an XLNet encoding network; Decoding the historical impact factor features using the feature decoding network in the order acceptance willingness prediction model to generate the order acceptance willingness prediction result of the driver, where the feature encoding network is a conditional random field network.

4. The method for training the order acceptance willingness of a driver according to any one of claims 1 to 3, characterized in that, Obtaining historical impact factor data that affects a driver's order acceptance, including: Collecting the original sample data of the historical impact factor data that affects a driver's order acceptance, where the original sample data includes various original impact factor data; Performing filtering and cleaning processing on the original sample data; Storing the original sample data after the filtering and cleaning processing; Selecting a part of the data from the original sample data after the filtering and cleaning processing as the training sample set of the order acceptance willingness prediction model, and the training sample set is the historical impact factor data that affects a driver's order acceptance.

5. The driver's order acceptance willingness training method according to claim 4, characterized in that, Selecting another part of the data from the original sample data after the filtering and cleaning processing as the test sample set of the order acceptance willingness prediction model, and the test sample set is used to test the accuracy of the order acceptance willingness prediction model.

6. The method for training the order acceptance willingness of a driver according to claim 1, characterized in that, The multiple layers of networks in the order acceptance willingness prediction model are respectively: an input layer, a hidden layer, an activation layer, and an output layer. Among them, the input layer generates a feature matrix based on the historical impact factor features, the hidden layer generates a first weight matrix based on the feature matrix, the activation layer is used to generate an activation function according to the output parameters and input parameters of the order acceptance willingness prediction model, and the output layer generates a second weight matrix based on the activation function and the first weight matrix.

7. A method for predicting the dispatching matching strategy of a driver, characterized in that, The method includes: Obtaining target impact factor data that affects a driver's order acceptance, where the target impact factor data includes various impact factor data; Obtain the current satisfaction level of the target passenger with respect to the driver, where the current satisfaction level of the target passenger with respect to the driver is determined based on the favorable comment rate of the target passenger for the driver; Extract target influence factor features from the target influence factor data, and input the target influence factor features into the order acceptance willingness prediction model described in any one of claims 1 to 6 for prediction. After the target influence factor features are calculated for the weight parameter values of the target influence factor features through multiple layers of networks in the order acceptance willingness prediction model, obtain the order acceptance willingness prediction result corresponding to the target influence factor data; Generate a dispatching matching strategy between the driver and the target passenger according to the order acceptance willingness prediction result and the current satisfaction level of the target passenger with respect to the driver.

8. A training device for a driver's willingness to accept orders, characterized in that, The device includes: A data acquisition module, configured to acquire historical influence factor data that affects a driver's order acceptance, where the historical influence factor data includes various influence factor data; A feature extraction module, configured to extract historical influence factor features from the historical influence factor data, and input the historical influence factor features into the order acceptance willingness prediction model for training. After the historical influence factor features are calculated for the weight parameter values of the historical influence factor features through multiple layers of networks in the order acceptance willingness prediction model, generate the order acceptance willingness prediction result corresponding to the historical influence factor data, and the loss value between the order acceptance willingness prediction result and the true label of the historical influence factor data, where the loss value is used to update the weight parameter values.

9. A computer device, characterized in that, Including: A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the order acceptance willingness training method for the driver described in any one of claims 1 to 6, or the dispatching matching strategy prediction method for the driver described in claim 7.

10. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the order acceptance willingness training method for the driver described in any one of claims 1 to 6, or the dispatching matching strategy prediction method for the driver described in claim 7.

Images