Prediction model training method, pricing prediction method, device, equipment and medium

Abstract

The invention provides a prediction model training method and device, a pricing prediction method and device, equipment and a medium, the training method is applied to floating pricing of an online car-hailing car in various scenes, and the training method comprises the following steps: acquiring single mileage pricing information in various scenes; obtaining text features of the training data, wherein the text features are text features of the training data of the market demand curve task constructed based on the data set; pre-training a large model according to the single mileage pricing information and the text features of the training data to obtain a first feature vector satisfying the single mileage pricing information, the second feature vector being obtained by fitting the data set; fusing the first feature vector and the second feature vector, and constructing a basic vector of the market demand curve task; and the basic vector of the market demand curve task is sent to each order task for demand curve correction training, and a trained prediction model is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and particularly to a method for training a prediction model, a pricing prediction method, a device, an electronic device, and a computer-readable storage medium. Background Art

[0002] In the related art, online car-hailing services have become an indispensable means of transportation. With the innovation of network technology, the change of market demand, and the evolution of business models, the rise of the online car-hailing aggregation model has become a profound change in modern urban travel. In the online car-hailing aggregation model, the interest relationships among the platform, drivers, and passengers are intertwined. A reasonable pricing of online car-hailing can attract more travel users. In this way, not only can the travel satisfaction of users be improved, but also the overall income of online car-hailing drivers and the platform can be increased. Therefore, in the highly competitive aggregation model, a single online car-hailing charging standard can no longer meet the current market business needs.

[0003] Therefore, how to perform differential pricing (i.e., floating pricing) according to different product configurations, channels, types of travel users, and time, etc. in different situations of online car-hailing is a problem to be solved currently. Summary of the Invention

[0004] The present invention provides a method for training a prediction model, a pricing prediction method, a device, an electronic device, and a computer-readable storage medium, so as to at least solve the problem in the related art that due to the inability of online car-hailing to perform differential pricing according to different product configurations, channels, types of travel users, and time, etc. in different situations, the efficiency of passengers choosing to take a car is low and the satisfaction is poor. The technical solution of the present invention is as follows:

[0005] According to the first aspect of the embodiments of the present invention, a method for training a prediction model is provided. The method is applied to predicting the floating pricing of online car-hailing in multiple scenarios, and includes:

[0006] Obtain the single-mileage pricing information in multiple scenarios;

[0007] Obtain the text features of the training data, where the training data is the training data for constructing the market demand curve task based on the data set;

[0008] Pre-train a large model according to the single-mileage pricing information and the text features of the training data to obtain a first feature vector, and the first feature vector meets the single-mileage pricing information;

[0009] Obtain a second feature vector, where the second feature vector is obtained by performing a fitting process on the data set;

[0010] Fuse the first feature vector and the second feature vector to construct a basis vector for the market demand curve task;

[0011] Send the basis vector of the market demand curve task to each order task respectively for deviation correction training of the market demand curve, and obtain a trained prediction model.

[0012] Optionally, the obtaining of the single-mileage pricing information in multiple scenarios includes:

[0013] Obtain the pricing configuration information in multiple scenarios of online car-hailing, as well as the text data of multiple order mileages;

[0014] Based on the pricing configuration information, determine the single-mileage pricing information corresponding to the text data of each order mileage, and obtain the single-mileage pricing information in multiple scenarios.

[0015] Optionally, for the obtaining of the text features of the training data, the training data is the training data for constructing the market demand curve task based on the data set, and includes:

[0016] Obtain the data set;

[0017] Process the data set to obtain dialogue table-type data matching the template;

[0018] Convert the dialogue table-type data into text data;

[0019] Construct the training data for the market demand curve task based on the text data;

[0020] Extract the text features of the training data.

[0021] Optionally, for the pre-training of the large model according to the single-mileage pricing information and the text features of the training data, so that the first feature vector output by the large model meets the single-mileage pricing information, it includes:

[0022] Input the text features of the training data into the large model for pre-training, extract the first feature vector, and after performing single-layer perception processing on the first feature vector, output the training result;

[0023] Calculate the loss value based on the difference between the single-mileage pricing information and the training result;

[0024] Based on the loss value, update the parameters of the large model through the backpropagation mechanism. After multiple iterations, until the loss value is the smallest, determine that the extracted first feature vector meets the single-mileage pricing information, and obtain a trained large model.

[0025] Optionally, the task of fusing the first feature vector and the second feature vector to construct the basis vectors of the market demand curve includes:

[0026] Fusing the first feature vector and the second feature vector by using a gating network or an attention mechanism to obtain a third feature vector;

[0027] Performing different multi-layer perceptron processes on the third feature vector respectively to predict the first basis vector and the second basis vector for the task of constructing the market demand curve.

[0028] Optionally, the task of sending the basis vectors of the market demand curve task to each order task respectively for deviation correction training of the demand curve to obtain a trained prediction model includes:

[0029] Sending the first basis vector and the second basis vector of the market demand curve task to each order task, and using the cross-entropy loss function to perform supervised training on the deviation correction of the demand curve to obtain a trained prediction model.

[0030] According to the second aspect of the embodiments of the present invention, there is provided a method for predicting the price of online car-hailing, including:

[0031] Obtaining the feature data of the market log data;

[0032] Converting the feature data into text data;

[0033] Based on inputting the text data into the trained prediction model, determining the order assignment rate, where the prediction model is trained by using a large model based on the single-mileage pricing information and the text features of the training data in multiple scenarios to obtain a first feature vector that meets the single-mileage pricing information, and fusing and training the first feature vector and the obtained second feature vector;

[0034] Predicting the floating price of the online car-hailing in multiple scenarios based on the order assignment rate and the obtained business requirements.

[0035] According to the third aspect of the embodiments of the present invention, there is provided a prediction model training device, which is applied to predicting the floating price of the online car-hailing in multiple scenarios, including:

[0036] A first acquisition module, configured to acquire the single-mileage pricing information in multiple scenarios;

[0037] A second acquisition module, configured to acquire the text features of the training data, where the training data is the training data for constructing the market demand curve task based on the data set;

[0038] A training module for pre-training a large model based on the single-mileage pricing information and the text features of the training data to obtain a first feature vector that satisfies the single-mileage pricing information.

[0039] A third acquisition module for acquiring a second feature vector of the data set, which is obtained by fitting the data set.

[0040] A fusion module for fusing the first feature vector and the second feature vector to construct a basis vector for the market demand curve task.

[0041] A correction training module for sending the basis vector of the market demand curve task to each order task respectively for correction training of the market demand curve to obtain a trained prediction model.

[0042] Optionally, the first acquisition module includes:

[0043] A configuration information acquisition module for acquiring pricing configuration information in various scenarios of online car-hailing.

[0044] A data acquisition module for text data of multiple order mileages.

[0045] A mileage pricing determination module for determining the single-mileage pricing information corresponding to the text data of each order mileage based on the pricing configuration information to obtain the single-mileage pricing information in various scenarios.

[0046] Optionally, the second acquisition module includes:

[0047] A data set acquisition module for acquiring a data set.

[0048] A processing module for processing the data set to obtain dialogue table-type data matching a template.

[0049] A conversion module for converting the dialogue table-type data into text data.

[0050] A construction module for constructing training data for the market demand curve task based on the text data.

[0051] An extraction module for extracting text features of the training data.

[0052] Optionally, the training module includes:

[0053] A pre-training module for inputting the text features of the training data into a large model for pre-training to extract a first feature vector.

[0054] A single-layer perception processing module, which is used to perform single-layer perception processing on the first feature vector and then output a training result;

[0055] A loss calculation module, which is used to calculate a loss value based on the difference between the single-mileage pricing information and the training result;

[0056] An iteration module, which is used to update the parameters of the large model based on the loss value through a backpropagation mechanism. After multiple iterations, until the loss value is minimized, it is determined that the extracted first feature vector satisfies the single-mileage pricing information, and a trained large model is obtained.

[0057] Optionally, the fusion module includes:

[0058] A feature fusion module, which is used to fuse the first feature vector and the second feature vector by using a gated network or an attention mechanism to obtain a third feature vector;

[0059] A prediction module, which is used to perform different multi-layer perception processing on the third feature vector respectively to predict a first basis vector and a second basis vector for the task of constructing a market demand curve.

[0060] Optionally, the deviation correction training module includes:

[0061] A sending module, which is used to send the first basis vector and the second basis vector of the market demand curve task to each order task;

[0062] A supervised training module, which is used to perform supervised training on the deviation correction of the demand curve based on each order task by using a cross-entropy loss function to obtain a trained prediction model.

[0063] According to the fourth aspect of the embodiments of the present invention, a ride-hailing pricing prediction device is provided, including:

[0064] An acquisition module, which is used to acquire feature data of market log data;

[0065] A conversion module, which is used to convert the feature data into text data;

[0066] A determination module, which is used to input the text data into a trained prediction model to determine the order assignment rate. The prediction model is trained by using a large model based on the single-mileage pricing information and the text features of training data in multiple scenarios to obtain a first feature vector that satisfies the single-mileage pricing information, and is a model obtained by performing fusion training on the first feature vector and the acquired second feature vector;

[0067] A prediction module, which is used to predict the floating pricing of ride-hailing in multiple scenarios based on the order assignment rate and the acquired business requirements.

[0068] According to a fifth aspect of an embodiment of the present invention, there is provided an electronic device, including:

[0069] A processor;

[0070] A memory for storing executable instructions of the processor;

[0071] Wherein, the processor is configured to execute the instructions to implement the prediction model training method as described above or the online car-hailing pricing prediction method as described above.

[0072] According to a sixth aspect of an embodiment of the present invention, there is provided a computer-readable storage medium, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enabling the electronic device to execute the prediction model training method as described above or the online car-hailing pricing prediction method as described above.

[0073] According to a seventh aspect of an embodiment of the present invention, there is provided a computer program product, including a computer program or instructions, which implement the prediction model training method as described above or the online car-hailing pricing prediction method as described above when executed by a processor of an electronic device.

[0074] The technical solutions provided by the embodiments of the present invention at least bring the following beneficial effects:

[0075] In the embodiments of the present invention, single-mile pricing information in multiple scenarios is obtained; text features of training data are obtained, and the training data is training data for constructing a market demand curve task based on a data set; a large model is pre-trained according to the single-mile pricing information and the text features of the training data to obtain a first feature vector, and the first feature vector meets the single-mile pricing information; a second feature vector is obtained, and the second feature vector is obtained by fitting the data set; the first feature vector and the second feature vector are fused to construct a basic vector for the market demand curve task; the basic vector for the market demand curve task is sent to each order task respectively for deviation correction training of the demand curve, and a trained prediction model is obtained. That is to say, in the embodiments of the present invention, the information understanding and extraction capabilities of the large model are utilized to extract information of main features, the extracted features are fused, a basic vector for the market demand curve task is constructed, and deviation correction training of the demand curve is performed based on the basic vector, increasing the complexity of the large model, changing the data input mode of the traditional estimation model, and enabling order assignment on demand at a lower cost, improving the efficiency and satisfaction of passengers choosing to take a ride.

[0076] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. Description of the Drawings

[0077] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments in accordance with the present invention, and are used together with the specification to explain the principles of the present invention, and do not constitute an improper limitation of the present invention. In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings required for use in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other accompanying drawings can be obtained based on these drawings without creative efforts.

[0078] Figure 1 It is a schematic diagram of a logit demand curve provided by an embodiment of the present invention.

[0079] Figure 2 It is a flowchart of a prediction model training method provided by an embodiment of the present invention.

[0080] Figure 3 It is a specific flowchart of a prediction model training method provided by an embodiment of the present invention.

[0081] Figure 4 It is a flowchart of a ride-hailing pricing prediction method provided by an embodiment of the present invention.

[0082] Figure 5 It is a block diagram of a prediction model training device provided by an embodiment of the present invention.

[0083] Figure 6 It is a block diagram of a first acquisition module provided by an embodiment of the present invention.

[0084] Figure 7 It is a block diagram of a ride-hailing pricing prediction device provided by an embodiment of the present invention.

[0085] Figure 8 It is a block diagram of an electronic device provided by an embodiment of the present invention.

[0086] Figure 9 It is a block diagram of a device for a prediction model training method or ride-hailing pricing prediction provided by an embodiment of the present invention. Detailed implementation manners

[0087] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0088] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0089] Before introducing the embodiments of the present invention, first understand the following content:

[0090] The purpose of the embodiments of the present invention is to give a dynamic bidding scheme for online car-hailing in different situations, and achieve the maximization of the target by combining the budget and traffic characteristics. To achieve this purpose, it is necessary to first understand the following problem definition and analysis:

[0091] Problem definition: For the online car-hailing bidding scenario, assume that the total traffic (bubble number) is N, which is distributed throughout the day according to demand. The total subsidy amount set is B, which is distributed to passengers for subsidies, and the goal is to maximize the daily income.

[0092] Problem analysis: Assume that the cost and market return conform to the logit demand curve. In this embodiment, the market is defined as the cross-binning of time periods and mileage (for example, the 7-9 time period is binned according to 3-5 kilometers or 6-9 kilometers, etc.). The specific binning logic is divided according to the demand distribution of different cities. Currently, the equal-frequency binning method is used for division, with a total of 36 bins.

[0093] Among them, the form of the logit demand curve is defined as follows:

[0094]

[0095] Among them, in the above formula, qi(c) represents the market share (rate) when the cost is c; di(c) represents the absolute value of the market share when the cost is c; Di represents the absolute value of the total market share; ai and bi are hyperparameters in the function form that control the shape (slope) of the function graph, where the symbol := represents the assignment operation.

[0096] Among them, the graph of the logit demand curve corresponding to the basic function is as Figure 1 shown, Figure 1 is a schematic diagram of a logit demand curve provided by the embodiments of the present invention, that is, a schematic diagram of the logit response model. From Figure 1It can be seen that the abscissa represents the pricing (cost), the ordinate (sales) represents the order giving rate, and Market cost represents the market cost. After obtaining the logit demand curve of the market, in this embodiment, according to the principle of operations research planning, the business problem is abstracted into a mathematical problem, as shown in the following formula:

[0097]

[0098] Among them, di(ci) represents the absolute value of the i-th market share when the cost is c, and i is a variable.

[0099] After that, based on the above-defined formula qi(c), the following is obtained by solving the inverse function:

[0100]

[0101] Then, the original business problem is transformed into a Lagrangian dual problem, and its formula is as follows:

[0102]

[0103] And for the above formula, through the KKT condition constraint and the complementary slackness principle, it is deduced to:

[0104]

[0105] Finally, substitute the above formulas into formula c i and q i , and finally obtain the definition formula of each market share and the bidding formula (that is, the bidding amount formula for each market). The bidding amount formula for each market is as follows:

[0106]

[0107] Among them, c i represents the subsidy amount for the i-th market, q i is the corresponding return value (income rate), a i and b i are hyperparameters in the form of functions, λ (lambda) is a hyperparameter used to control the subsidy rate, and W is the Lambert function.

[0108] Based on the above analysis process, the embodiment of the present invention needs to model the market demand curve and predict a i and b i , a i and b i are the hyperparameters in the above formula and can be predicted through the above formula. After that, use the large model to correct and train the demand curve, so as to obtain a trained prediction model for predicting the floating pricing of online car-hailing in various scenarios. For the specific implementation process, please refer to the following embodiments.

[0109] Please refer to the figure, Figure 2 which is a flowchart of a prediction model training method provided by an embodiment of the present invention. The method is applied to predicting the floating pricing of online car-hailing in multiple scenarios, such as Figure 2 shown, the prediction model training method includes the following steps:

[0110] Step 201: Obtain the single-mileage pricing information in multiple scenarios.

[0111] Step 202: Obtain the text features of the training data, where the training data is the training data for constructing the market demand curve task based on the data set.

[0112] Step 203: Pre-train the large model according to the single-mileage pricing information and the text features of the training data to obtain a first feature vector, and the first feature vector satisfies the single-mileage pricing information.

[0113] Step 204: Obtain a second feature vector, which is obtained by fitting the data set.

[0114] Step 205: Fuse the first feature vector and the second feature vector to construct a basic vector for the market demand curve task.

[0115] Step 206: Send the basic vector of the market demand curve task to each order task respectively for deviation correction training of the demand curve to obtain a trained prediction model.

[0116] The prediction model training method described in the present invention can be applied to terminals, servers, etc., which are not limited here. The terminal implementation devices can be electronic devices such as smart phones, laptop computers, tablet computers, desktop computers, personal digital assistants (PDAs), and wearable devices. The server can be an independent server, or a server cluster, or a server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, intermediate services, domain name services, security services, content delivery networks, or a big data and artificial intelligence platform, etc., which are not limited here.

[0117] Next, in combination with Figure 2 , the specific implementation steps of a prediction model training method provided by an embodiment of the present invention will be described in detail.

[0118] In step 201, obtain the single-mileage pricing information in multiple scenarios.

[0119] In this step, first obtain the pricing configuration information under various scenarios of online car-hailing and the text data of various order mileage; then, based on the pricing configuration information, determine the single-mileage pricing information corresponding to the text data of each order mileage to obtain the single-mileage pricing information (y_ture) under various scenarios.

[0120] That is to say, in this step, the pricing configuration information is usually selected from the pre-configured rule strategy combinations, that is, the single-mileage pricing rules under different scenarios are preset as the current base version, and this data is called empirical data.

[0121] After that, obtain various order mileage and parse the various order mileage into text data.

[0122] Finally, based on the pricing configuration information, determine the single-mileage pricing information corresponding to the text data of each order mileage to obtain the single-mileage pricing information (y_ture) under various scenarios; for example, for an order with a mileage segment of 3 - 5 km, the single-mileage pricing is 3 yuan, etc.

[0123] In this embodiment, the pricing configuration information and the text data of various order mileage can be input into a model (such as the Chatgpt model, etc.), and the model will automatically parse the rules and match the custom pricing configuration information according to the provided text data of each order mileage. For example, the rule (i.e., the pricing configuration information) input into the Chatgpt model is that the single-mileage pricing for 3 - 5 km is 3 yuan; when the order mileage is 4 km, the corresponding single-mileage pricing information output by the Chatgpt model is 3 yuan, which is y_true.

[0124] In step 202, obtain the text features of the training data, where the training data is the training data for constructing the market demand curve based on the data set.

[0125] In this step, first, obtain the data set, which may include: market granularity data and feature data. Of course, in specific implementation, it is not limited to this.

[0126] Among them, market granularity data: in the online car-hailing industry, take the geographical location (such as a seven-level hexagonal grid) and time period (such as a 10-minute cycle, etc.) as a granularity, count the unit mileage cost C and the order giving rate (i.e., the number of given orders / the number of bubbles) therein, and construct the training data of the demand curve based on the unit mileage cost C and the order giving rate.

[0127] Among them, the feature data includes: real-time features, trend features, and deviation correction features. The real-time features include: data related to statistics and business indicators, including but not limited to: the number of cruising vehicles, call numbers, bubble numbers, cancellation numbers, etc. The trend features include: real-time data in a certain recent time period (such as two hours) as sequence features. For example, obtain the statistical values of features such as bubble numbers and call numbers every 15 minutes within the recent two hours, and construct a time series feature based on the statistical values. The deviation correction feature is a continuous curve at the current granularity.

[0128] In this embodiment, the text features of the training data can also be referred to as the text features of the constructed corpus.

[0129] Secondly, process the data set to obtain dialogue tabular data matching the template.

[0130] Specifically, in this step, a Chatgpt assistant generator can be used to process the data set to obtain dialogue tabular data matching the template.

[0131] In this embodiment, the process of using a Chatgpt assistant generator to process the data set is already well-known to those skilled in the art and will not be elaborated here. Among them, the matching template is a pre-set pricing template. Its Chatgpt assistant generator mainly functions to provide a generator matching template, using the absolute value of the single loss as the Reward (incentive function) for the reinforcement learning idea to help Chatgpt learn samples.

[0132] Thirdly, convert the dialogue tabular data into text data.

[0133] The conversion format in this step is already well-known to those skilled in the art and will not be elaborated here.

[0134] Finally, based on the text data, construct the training data for the market demand curve task and extract the text features of the training data.

[0135] In step 203, pre-train the large model according to the single-mileage pricing information and the text features of the training data to obtain a first feature vector, and the first feature vector satisfies the single-mileage pricing information.

[0136] In this step, first, input the text features of the training data and the single-mileage pricing information into a large model for pre-training, extract feature vectors, and perform single-layer perception (that is, if the activation function of each neuron is a linear function, then any number of layers of MLP can be reduced to an equivalent single-layer perceptron) processing on the extracted feature vectors, and then output the training result (y_pre); second, calculate the loss value based on the difference between the single-mileage pricing information (y_ture) and the training result (y_pre). Specifically, the loss value can be calculated through the mean squared error loss function (MSE, Mean Squared Error); finally, based on the loss value, update the parameters of the large model through the backpropagation mechanism. After multiple iterations, until the loss value is minimized. That is to say, the purpose of calculating the loss value is to evaluate whether the current training result is close to the single-mileage pricing information. The smaller the calculated loss value, the closer the current training result is to the single-mileage pricing information. When the loss value is minimized, it is determined that the feature vector extracted this time meets the single-mileage pricing information, and the feature vector extracted this time is called the first feature vector (H_c). At this time, the trained large model is obtained.

[0137] Among them, in this embodiment, y_pre represents the output value of the self-training model, that is, the training result each time. The purpose of pre-training the large model in this way is to make it fully understand the existing custom rule part.

[0138] That is to say, in this embodiment, the large model is pre-trained using the constructed real market Prompt data and the rule output part of Chatgpt, so that the first feature vector (i.e., the first hidden layer vector H_c) extracted by the large model conforms to the custom objective single-mileage pricing information. During the training process, the mean squared error loss function (MSE) is used for supervised training.

[0139] Among them, Prompt (prompt) refers to a piece of text or instruction added to the input when using a machine learning model. Its purpose is to guide the large model to generate more accurate and targeted outputs. A Prompt can be a question, a description, a formatted input, or even some keywords. By adding a Prompt to the input of the large model, the behavior of the large model can be customized and controlled to better meet the custom needs.

[0140] Mean squared error loss function (MSE): Backpropagation, train the model parameters, use gradient descent on all parameters to minimize the loss function of the neural network (NN, Neural Network) model on the training data. MSE is a commonly used loss function in regression tasks, which measures the average squared error between the model prediction value and the actual value. The calculation formula of MSE is as follows:

[0141]

[0142] Among them, n represents the number of samples, and y i represents the actual value of the i-th sample, and represents the predicted value of the model for the i-th sample. The smaller the value of MSE, the smaller the difference between the predicted value and the true value of the model, and the better the performance of the model.

[0143] Loss function (loss): The calculated predicted value and the actual value (known answer) y i gap.

[0144]

[0145] It should be noted that the meanings of the parameters in this formula are as described above and will not be elaborated here.

[0146] In step 204, a second feature vector is obtained, and the second feature vector is obtained by fitting the data set.

[0147] In this step, the data set may include: market granularity data and feature data. Of course, in specific implementations, it is not limited to this.

[0148] Among them, market granularity data: In the online car-hailing industry, geographical location (such as a seven-level hexagonal grid) and time period (such as a 10-minute cycle, etc.) are used as a granularity, and the unit mileage cost C and order assignment rate (i.e., the number of assigned orders / the number of bubbles) are statistically calculated. Based on the unit mileage cost C and order assignment rate, training data for the demand curve can be constructed.

[0149] Among them, the feature data may include: real-time features, trend features, and correction features. The real-time features include: data related to statistics and business indicators, including but not limited to: the number of cruising vehicles, the number of calls, the number of bubbles, the number of cancellations, etc. The trend features include: real-time data in a recent time period (such as two hours) as sequence features. For example, obtain the statistical values of features such as the number of bubbles and the number of calls every 15 minutes within the recent two hours, and construct a time series feature based on the statistical values, etc. The correction feature is a continuous curve at the current granularity.

[0150] In this step, a multi-layer MLP (Multi-Layer Perceptron) structure can be used to fit the information of tabular data in the data set and extract the second feature vector of the key part (i.e., the second hidden layer vector H_a, the same below). Here, fitting means matching a model or function with actual data to obtain an optimal model or function that can describe or predict these data. It can also be understood as connecting a series of points on a plane with a smooth curve. Since there are countless possibilities for this curve, there are various fitting methods. The fitting curve can generally be represented by a function, and according to the differences of this function, there are different fitting names. For example, common fitting methods include: the least squares curve fitting method, etc. These are already well-known techniques to those skilled in the art and will not be elaborated here.

[0151] In step 205, the first feature vector and the second feature vector are fused to construct the basic vector for the market demand curve task.

[0152] In this step, the first feature vector (i.e., the first hidden layer vector H_c) and the second feature vector (i.e., the second hidden layer vector H_a) can be first fused (i.e., combined) using a gating network or an attention mechanism to obtain the third feature vector H_s. Then, different multi-layer perceptron processes are performed on the third feature vector H_s respectively to predict the first basic vector a_i and the second basic vector b_i for the market demand curve task.

[0153] Among them, when using a gating network to fuse H_c and H_a, the weights are mostly 0 or 1. The attention mechanism can also be used to fuse H_c and H_a, and the formula for its fusion using the attention mechanism is as follows:

[0154] α c = softmax(W c ·H c + b c )

[0155] α a = softmax(W a ·H a + b a )

[0156] H s = α c ·H c + α a ·H a

[0157] Among them, W c , bc, W aω and β are respectively represented as learnable weight matrices, and Hs represents the third feature vector obtained by fusing the first hidden layer vector Hc and the second hidden layer vector Ha in the bilateral information extraction of the model structure, usually an embedding vector; α c and α α are respectively self-learning parameters in the attention mechanism. It should be noted that Hs, Hc, and Ha in this formula are all feature vectors, which are the third feature vector H_s, the first hidden layer vector H_c, and the second hidden layer vector H_a respectively. Here, the concept of weight is emphasized. The process of fusing using the above fusion formula is already well-known in the art and will not be elaborated here.

[0158] In step 206, the base vectors of the market demand curve task are respectively sent to each order task for deviation correction training of the demand curve, and a trained prediction model is obtained.

[0159] In this step, the first base vector and the second base vector of the market demand curve task are sent to each order task, and the cross-entropy loss function is used for supervised training of the deviation correction of the demand curve, and a trained prediction model is obtained. In the training process of this embodiment, cross-entropy loss is used for supervised training. Its specific training is already well-known in the art and will not be elaborated here.

[0160] In the embodiment of the present invention, single-mileage pricing information in multiple scenarios is obtained; text features of training data are obtained, and the training data is training data for constructing a market demand curve task based on a data set; the large model is pre-trained according to the single-mileage pricing information and the text features of the training data to obtain a first feature vector, and the first feature vector satisfies the single-mileage pricing information; a second feature vector is obtained, and the second feature vector is obtained by fitting the data set; the first feature vector and the second feature vector are fused to construct a base vector of the market demand curve task; the base vector of the market demand curve task is respectively sent to each order task for deviation correction training of the demand curve, and a trained prediction model is obtained. That is to say, in the embodiment of the present invention, the information understanding and extraction capabilities of the large model are utilized to extract information of main features, the extracted features are fused, a base vector of the market demand curve task is constructed, and deviation correction training of the demand curve is performed based on the base vector, increasing the complexity of the large model, changing the data input method of the traditional estimation model, enabling order assignment on demand at a lower cost, and improving the efficiency and satisfaction of passengers choosing to take a ride.

[0161] Please also refer to Figure 3 , which is a specific flowchart of a prediction model training method provided by an embodiment of the present invention. The method specifically includes:

[0162] Step 301: Obtain a custom configuration template (structured data).

[0163] In this embodiment, the custom configuration template can be configured based on experience or determined in the following manner: Obtain the pricing configuration information in various scenarios of online car-hailing and the text data of various order mileage.

[0164] Step 302: Based on the pricing configuration information and the text data of each order mileage, the single-mileage pricing information y_ture in various scenarios can be determined by using Chatgpt for assisted training.

[0165] Step 303: Obtain a data set.

[0166] In this step, the data set may include: market granularity data, feature data, experience data, etc. The specific content of various data is as described above and will not be elaborated here.

[0167] Step 304: Based on the data set, use the Chatgpt assisted generator to generate Chatgpt assisted information, and the Chatgpt assisted information includes: dialogue table-type data matching the template.

[0168] In this step, the main function of the Chatgpt assisted information is to provide a template for the generator to match. For the reinforcement learning idea, the absolute value of the single loss is used as the Reward (incentive function) to help Chatgpt learn samples. The specific method of using the Chatgpt assisted generator to generate Chatgpt assisted information is already well-known to those skilled in the art and will not be elaborated here.

[0169] Step 305: Based on the Chatgpt assisted information, use the Prompt generator to generate the text features of the training data, and the training data is the training data for constructing the market demand curve task based on the data set.

[0170] In this step, the Prompt generator is used to convert the dialogue table-type data into text data for large model training. At the same time, attention is paid to the introduction of focus information, such as emphasizing the tight supply and demand situation or the small order volume, to help the large model find the focus of data features for extraction. Usually, multiple templates are configured for automatic data matching and generation.

[0171] Step 306: Construct corpus text features based on the text data.

[0172] Step 307: Input the corpus text features into the large model for pre-training so that the first hidden layer vector H_c output by the large model conforms to the objective cognition of users.

[0173] That is to say, the first hidden layer vector H_c output by the large model satisfies the single-mileage pricing information in multiple scenarios.

[0174] Step 308: Perform single-layer MLP processing on the first hidden layer vector H_c to obtain y_pre.

[0175] Step 309: Based on the difference between the single-mileage pricing information y_ture and y_pre, calculate the loss value through the mean square error loss function (MSE), return the loss value through the incentive function (Reward) to Step 304; and based on the loss value, update the parameters of the large model through the backpropagation mechanism. After multiple iterations, until the loss value is minimized, it is determined that the extracted first hidden layer vector H_c satisfies the single-mileage pricing information.

[0176] Step 310: Perform fitting processing on the data set to obtain the second hidden layer vector H_a.

[0177] Step 311: Perform adaptive fusion on the first hidden layer vector H_c and the second hidden layer vector H_a to obtain the third feature vector H_s.

[0178] Specifically, an adaptive fusion module can be used for fusion.

[0179] Step 312: Perform different multi-layer perceptron (i.e., MLP1 and MLP2) processing on the third feature vector H_s respectively to predict the first basis vector a_i and the second basis vector b_i for the task of constructing the market demand curve.

[0180] Step 313: Perform pre-processing on the first basis vector a_i and the second basis vector b_i.

[0181] Step 314: Perform corrective training on the pre-processing result using the cross-entropy loss function (CrossEntropyLoss)_ to predict the floating pricing of the online car-hailing in multiple scenarios (i.e., predict the market demand curve).

[0182] Among them, in this embodiment, the cross-entropy loss function CEL, the size of which represents the difference between two probability distributions, can obtain an approximate distribution of the target probability distribution by minimizing the cross-entropy. In binary classification problem models: such as Logistic Regression, Neural Network, etc., the labels of real samples are [0, 1], representing the negative class and the positive class respectively. Usually, the model will go through a Sigmoid function at the end and output a probability value, which reflects the possibility of predicting the positive class: the larger the probability, the greater the possibility.

[0183] Specifically, the embodiment of the present invention can be divided into three parts:

[0184] The first part is to understand the base version strategy. The base version strategy is a combination of configured rule strategies, and the combination of its rule strategies is to set pricing rules under different scenarios.

[0185] In this step, the empirical data is input into models such as Chatgpt (or parsed item by item, which is difficult, but if the capabilities of large models are used for generalization, the data effect will be better), so that it can fully understand (i.e., parse into text form) to obtain the single-mileage pricing information y_ture under various scenarios. For example, for an order with a mileage segment of 3 - 5 km, the single-mileage pricing is 3 yuan, etc.

[0186] The second part: Pre-train the large model, that is Figure 3 The left half of the middle model structure.

[0187] In this step, the constructed real market Prompt data and the rule output part of Chatgpt are used for pre-training the large model, so that the first hidden layer vector H_c extracted by it conforms to the objective cognition of humans, and the mean squared error (MSE) loss function is used for supervised training.

[0188] The third part: Train the main body structure, combine numerical features, and depict the revenue curve, that is, the right half of the model structure

[0189] In this step, a multi-layer MLP (multi-layer perceptron) structure is used to fit the information of tabular data, extract the second hidden layer vector H_a of the key part and fuse it with the first hidden layer vector H_c extracted by the large model to form the basic vector part of the task, and send the information to each task respectively. The cross-entropy loss is used for supervised training during the training process.

[0190] In this embodiment, for the fusion of H_c and H_a, a gated network or an attention mechanism can be used. In most cases, the gated network will have weights of 0 or 1. It is recommended to use the attention mechanism, and the fusion method is, where W and b are learnable weight matrices.

[0191] In the embodiment of the present invention, the information understanding and extraction capabilities of the large model are used to extract the information of the main features, fuse the extracted features, construct the basic vector of the market demand curve task, and perform deviation correction training on the demand curve based on the basic vector, increasing the complexity of the large model, changing the data input method of the traditional prediction model, enabling order giving on demand at a lower cost, and improving the efficiency and satisfaction of passengers' choice of taking a ride.

[0192] For easy understanding, please refer to the following specific application embodiments.

[0193] The current market environment is described as follows: the number of surrounding vehicles {.f}, the average surrounding subsidy rate {.m}, within the last 30 minutes, the actual subsidy amount and pick-up situation: {.s}. Within the last 30 minutes, the predicted subsidy amount and pick-up situation: {.h}. Pay attention to the difference information, and vectorize the above information into 64-dimensional features. It should be noted that {.f}, {.m}, {.s}, and {.h} are all placeholders.

[0194] Among them, an example of empirical rule data (structured data) is shown in Table 1, but in actual applications, it is not limited to this.

[0195] Table 1

[0196]

[0197] Please also refer to Figure 4 , which is a flowchart of a method for predicting the pricing of online car-hailing provided by an embodiment of the present invention. The method includes:

[0198] Step 401: Obtain the feature data of market log data.

[0199] In this step, the feature data of the market log data may include: real-time features, trend features, correction features, etc. The specific acquisition process is detailed in the corresponding embodiments above and will not be elaborated here.

[0200] Step 402: Convert the feature data into text data;

[0201] In this step, there are various ways of conversion, and this embodiment does not limit them.

[0202] Step 403: Input the text data into a trained prediction model to determine the order assignment rate; among them, the prediction model is trained using a large model based on the single-mileage pricing information and text features of training data in multiple scenarios to obtain a first feature vector that meets the single-mileage pricing information, and the model is obtained by fusing and training the first feature vector and the obtained second feature vector.

[0203] In this step, the prediction model statistically analyzes the input text data to determine the number of orders assigned and the number of driver bubbles in the text data within a set time period. Then, divide the number of orders assigned by the number of bubbles to determine the order assignment rate.

[0204] Among them, the text data in this embodiment is the Prompt data.

[0205] Step 404: Based on the order assignment rate and the obtained business requirements, predict the floating pricing of online car-hailing in multiple scenarios.

[0206] In this step, obtaining the business requirements means controlling the subsidy consumption speed according to time, which can usually be achieved through PID (Proportional Integral Derivative) algorithm control. Among them, the PID is a feedback control algorithm, and its implementation process is already well-known to those skilled in the art and will not be elaborated here.

[0207] In this step, the given order rate and the obtained business requirements are input into the market bidding amount formula for calculation to predict the floating pricing of online car-hailing in various scenarios. Among them, the market bidding formula includes:

[0208]

[0209] Among them, c i represents the subsidy amount for the i-th market, q i represents the return value (income rate) of the corresponding market, a i and b i are hyperparameters in the form of functions. λ (lambda) represents the hyperparameter used to control the subsidy rate, and W is the Lambert function.

[0210] In the embodiment of the present invention, a trained prediction model is used to determine the given order rate, and based on the given order rate and the obtained business requirements, the floating pricing of online car-hailing in various scenarios is predicted. In this embodiment, by means of text plus numerical values, the data input method of the traditional estimation model is changed, and the given order rate is promoted as needed at a lower cost, improving the efficiency and satisfaction of passengers choosing to take a ride.

[0211] It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the present invention.

[0212] Please also refer to the figure, Figure 5 which is a block diagram of a prediction model training device provided by an embodiment of the present invention. The device is applied to predicting the floating pricing of online car-hailing in various scenarios, and includes: a first acquisition module 501, a second acquisition module 502, a training module 503, a third acquisition module 504, a fusion module 505, and a deviation correction training module 506.

[0213] The first acquisition module 501 is used to acquire the single-mile pricing information in various scenarios;

[0214] The second acquisition module 502 is configured to acquire the text features of the training data, where the training data is the training data for constructing the market demand curve task based on the data set;

[0215] The training module 503 is configured to pre-train the large model according to the single-mileage pricing information and the text features of the training data to obtain a first feature vector, where the first feature vector satisfies the single-mileage pricing information;

[0216] The third acquisition module 504 is configured to acquire a second feature vector of the data set, where the second feature vector is obtained by performing a fitting process on the data set;

[0217] The fusion module 505 is configured to fuse the first feature vector and the second feature vector to construct a basis vector for the market demand curve task;

[0218] The deviation correction training module 506 is configured to send the basis vector of the market demand curve task to each order task respectively for deviation correction training of the market demand curve to obtain a trained prediction model.

[0219] Optionally, in another embodiment, based on the above embodiment, the first acquisition module 501 includes: a configuration information acquisition module 601, a data acquisition module 602, and a mileage pricing determination module 603. The structural block diagram is as Figure 6 shown, where

[0220] The configuration information acquisition module 601 is configured to acquire the pricing configuration information in various scenarios of online car-hailing;

[0221] The data acquisition module 602 is configured to acquire text data of multiple order mileages;

[0222] The mileage pricing determination module 603 is configured to determine the single-mileage pricing information corresponding to the text data of each order mileage based on the pricing configuration information to obtain the single-mileage pricing information in various scenarios.

[0223] Optionally, in another embodiment, based on the above embodiment, the second acquisition module includes: a data set acquisition module, a processing module, a conversion module, a construction module, and an extraction module. Among them,

[0224] The data set acquisition module is configured to acquire a data set;

[0225] The processing module is configured to process the data set to obtain dialogue table-type data matching the template;

[0226] The conversion module is configured to convert the dialogue table-type data into text data;

[0227] A construction module for constructing training data for the task of building a market demand curve based on the text data;

[0228] An extraction module for extracting text features of the training data.

[0229] Optionally, in another embodiment, based on the above embodiment, the training module includes:

[0230] A pre-training module for inputting the text features of the training data into a large model for pre-training to extract a first feature vector;

[0231] A single-layer perceptron processing module for performing single-layer perceptron processing on the first feature vector and then outputting a training result;

[0232] A loss calculation module for calculating a loss value based on the difference between the single-mileage pricing information and the training result;

[0233] An iteration module for updating the parameters of the large model through a backpropagation mechanism based on the loss value. After multiple iterations, until the loss value is minimized and it is determined that the extracted first feature vector satisfies the single-mileage pricing information, a trained large model is obtained.

[0234] Optionally, in another embodiment, based on the above embodiment, the fusion module includes:

[0235] A feature fusion module for fusing the first feature vector and the second feature vector using a gated network or an attention mechanism to obtain a third feature vector;

[0236] A prediction module for performing different multi-layer perceptron processing on the third feature vector respectively to predict a first basis vector and a second basis vector for the task of building a market demand curve.

[0237] Optionally, in another embodiment, based on the above embodiment, the deviation correction training module includes:

[0238] A sending module for sending the first basis vector and the second basis vector of the market demand curve task to each order task;

[0239] A supervised training module for performing supervised training on the deviation correction of the demand curve based on each order task using a cross-entropy loss function to obtain a trained prediction model.

[0240] Please also refer to Figure 7 , which is a block diagram of a ride-hailing pricing prediction device provided by an embodiment of the present invention. The device includes: an acquisition module 701, a conversion module 702, a determination module 703, and a prediction module 704, where

[0241] An acquisition module 701 for acquiring feature data of market log data;

[0242] A conversion module 702 for converting the feature data into text data;

[0243] A determination module 703 for determining the order assignment rate based on inputting the text data into a trained prediction model, where the prediction model is trained by using a large model based on single-mileage pricing information and text features of training data in multiple scenarios to obtain a first feature vector that meets the single-mileage pricing information, and fusing and training the first feature vector and a obtained second feature vector;

[0244] A prediction module 704 for predicting the floating pricing of online car-hailing in multiple scenarios based on the order assignment rate and the obtained business requirements.

[0245] Optionally, an embodiment of the present invention further provides an electronic device, including:

[0246] A processor;

[0247] A memory for storing executable instructions of the processor;

[0248] Wherein, the processor is configured to execute the instructions to implement the prediction model training method as described above or the online car-hailing pricing prediction method as described above.

[0249] Optionally, an embodiment of the present invention further provides a computer-readable storage medium, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enabling the electronic device to execute the prediction model training method as described above or the online car-hailing pricing prediction method as described above.

[0250] Optionally, an embodiment of the present invention further provides a computer program product, including a computer program or instructions, when the computer program or instructions are executed by a processor of an electronic device, implementing the prediction model training method as described above or the online car-hailing pricing prediction method as described above.

[0251] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0252] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative effort.

[0253] Figure 8 FIG. is a block diagram of an electronic device 800 provided by an embodiment of the present invention. For example, the electronic device 800 can be a mobile terminal or a server. In this embodiment of the present invention, the electronic device is taken as a mobile terminal as an example for illustration. For example, the electronic device 800 can be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0254] Referring to Figure 8 FIG.

[0253] , the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0255] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communication, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0256] The memory 804 is configured to store various types of data to support the operation of the device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disk.

[0257] The power supply component 806 provides power for various components of the electronic device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

[0258] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can not only sense the boundaries of the touch or swipe actions, but also detect the duration and pressure associated with the touch or swipe operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera can receive external multimedia data. Each of the front camera and the rear camera can be a fixed optical lens system or have a focal length and optical zoom capabilities.

[0259] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC) that is configured to receive external audio signals when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 further includes a speaker for outputting audio signals.

[0260] The I / O interface 812 provides an interface between the processing component 802 and a peripheral interface module, and the peripheral interface module can be a keyboard, a click wheel, buttons, etc. These buttons can include, but are not limited to: a home button, a volume button, a power-on button, and a lock button.

[0261] The sensor assembly 814 includes one or more sensors for providing status assessments of various aspects for the electronic device 800. For example, the sensor assembly 814 can detect the on / off state of the device 800, the relative positioning of components, such as the display and keypad of the electronic device 800. The sensor assembly 814 can also detect a change in the position of the electronic device 800 or a component of the electronic device 800, the presence or absence of user contact with the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and a change in the temperature of the electronic device 800. The sensor assembly 814 can include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 can also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 can also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0262] The communication component 816 is configured to facilitate communication, in a wired or wireless manner, between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on communication standards, such as WiFi, a carrier network (such as 2G, 3G, 4G, or 5G), or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0263] In an embodiment, the electronic device 800 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components for performing the prediction model training method or the online car-hailing pricing prediction method shown above.

[0264] In an embodiment, a computer-readable storage medium is also provided. When the instructions in the computer-readable storage medium are executed by a processor of the electronic device, the electronic device 800 can perform the prediction model training method or the online car-hailing pricing prediction method shown above. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0265] In an embodiment, a computer program product is further provided, including a computer program or instructions. When the computer program or instructions are executed by a processor 820 of an electronic device 800, the electronic device 800 is caused to execute the prediction model training method or the online car-hailing pricing prediction method shown above.

[0266] In the above embodiment, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0267] Figure 9 It is a block diagram of a device 900 for prediction model training or online car-hailing pricing prediction provided by an embodiment of the present invention. For example, the device 900 can be provided as a server. Referring to Figure 9 , the device 900 includes a processing component 922, which further includes one or more processors, and memory resources represented by a memory 932 for storing instructions executable by the processing component 922, such as application programs. The application programs stored in the memory 932 can include one or more modules each corresponding to a set of instructions. In addition, the processing component 922 is configured to execute instructions to perform the above method.

[0268] Device 900 may further include a power supply component 926 configured to perform power management of device 900, a wired or wireless network interface 950 configured to connect device 900 to a network, and an input / output (I / O) interface 958. Device 900 may operate based on an operating system stored in memory 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.

[0269] Those skilled in the art will readily conceive of other embodiments of the present invention upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following the general principles of the invention and including known common general knowledge or conventional technical means in the technical field not disclosed herein. The specification and examples are only exemplary, and the true scope and spirit of the invention are pointed out by the following claims.

[0270] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A prediction model training method, characterized in that: The method is applied to predict the floating pricing of online ride-hailing services in various scenarios, including: Get single-mileage pricing information in a variety of scenarios; Acquire text features of training data, where the training data is training data for a task of constructing a market demand curve based on a data set; Pre-training a large model according to the single mileage pricing information and text features of the training data to obtain a first feature vector, where the first feature vector satisfies the single mileage pricing information; Obtaining a second eigenvector, where the second eigenvector is obtained by fitting the data set; The first feature vector and the second feature vector are merged to construct a basic vector of a market demand curve task; The basic vector of the market demand curve task is sent to each order task respectively to perform deviation correction training of the market demand curve to obtain a trained prediction model.

2. The prediction model training method according to claim 1, characterized in that: The acquisition of single mileage pricing information in various scenarios includes: Obtain pricing configuration information for online ride-hailing services in various scenarios, as well as text data on mileage for various orders; Based on the pricing configuration information, the single mileage pricing information corresponding to the text data of each order mileage is determined to obtain the single mileage pricing information in multiple scenarios.

3. The prediction model training method according to claim 1, characterized in that: The acquiring of text features of training data, wherein the training data is training data for a task of constructing a market demand curve based on a data set, includes: Get the data set; Processing the data set to obtain dialog table data matching the template; Converting the dialog table data into text data; Constructing training data for a market demand curve task based on the text data; Extract text features of the training data.

4. The prediction model training method according to claim 1, characterized in that: The pre-training of the large model according to the single mileage pricing information and the text features of the training data so that the first feature vector output by the large model satisfies the single mileage pricing information includes: Inputting the text features of the training data into the large model for pre-training, extracting the first feature vector, and performing single-layer perception processing on the first feature vector, and then outputting the training result; Calculating a loss value based on a difference between the single mileage pricing information and the training result; Based on the loss value, the parameters of the large model are updated through a back-propagation mechanism. After multiple iterations, until the loss value is minimized, it is determined that the extracted first feature vector satisfies the single-mile pricing information, thereby obtaining a trained large model.

5. The prediction model training method according to claim 1, characterized in that: The step of fusing the first feature vector and the second feature vector to construct a basic vector for the market demand curve task includes: Using a gating network or an attention mechanism to fuse the first feature vector and the second feature vector to obtain a third feature vector; Different multi-layer perception processes are performed on the third eigenvectors to predict the first basic vector and the second basic vector for the task of constructing the market demand curve.

6. The prediction model training method according to claim 1, characterized in that: The step of sending the basic vector of the market demand curve task to each order task to perform demand curve deviation correction training to obtain a trained prediction model includes: The first basis vector and the second basis vector of the market demand curve task are sent to each order task, and the cross entropy loss function is used to perform supervised training on the correction of the demand curve to obtain a trained prediction model.

7. A method for predicting online car-hailing pricing, characterized in that: include: Get feature data of market log data; Converting the feature data into text data; Based on the text data input into the trained prediction model, the order rate is determined, wherein the prediction model is based on the unit mileage pricing information in multiple scenarios and the text features of the training data, and is trained using a large model to obtain a first feature vector that satisfies the unit mileage pricing information, and the first feature vector and the obtained second feature vector are fused and trained to obtain a model; Based on the order rate and acquired business needs, the floating pricing of online ride-hailing services in various scenarios is predicted.

8. A prediction model training device, characterized in that: The device is used to predict the floating pricing of online ride-hailing vehicles in various scenarios, including: A first acquisition module is used to obtain single mileage pricing information in various scenarios; A second acquisition module is used to acquire text features of training data, where the training data is training data for a task of constructing a market demand curve based on a data set; A training module, configured to pre-train a large model according to the single mileage pricing information and text features of the training data to obtain a first feature vector, wherein the first feature vector satisfies the single mileage pricing information; A third acquisition module is used to acquire a second eigenvector of the data set, where the second eigenvector is obtained by fitting the data set; A fusion module, used for fusing the first feature vector and the second feature vector to construct a basic vector for a market demand curve task; The deviation correction training module is used to send the basic vector of the market demand curve task to each order task respectively to perform deviation correction training of the market demand curve to obtain a trained prediction model.

9. A device for predicting online car-hailing pricing, characterized in that: include: An acquisition module, used to acquire characteristic data of market log data; A conversion module, used for converting the feature data into text data; a determination module, configured to determine the order rate based on the input of the text data into a trained prediction model, wherein the prediction model is based on the unit mileage pricing information in various scenarios and the text features of the training data, is trained using a large model, obtains a first feature vector that satisfies the unit mileage pricing information, and fuses the first feature vector with the obtained second feature vector to obtain a model; The prediction module is used to predict the floating pricing of online car-hailing in various scenarios based on the order rate and the obtained business needs.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the prediction model training method as described in any one of claims 1 to 6 or the online car-hailing pricing prediction method as described in claim 7.

11. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the prediction model training method as described in any one of claims 1 to 6 or the online car-hailing pricing prediction method as described in claim 7.

Images