Transaction duration prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a transaction duration prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: predicting the transaction duration of a first goods source; performing feature extraction, feature vector fusion, feature vector mapping and other operations on the related information of the first goods source through a target transaction duration prediction model to predict a first transaction result of the first goods source within a first duration after the first goods source is put on a shelf and predict a second transaction result of the first goods source within a second duration after the first goods source is put on the shelf; and jointly determining the transaction duration of the first goods source by using the first transaction result and the second transaction result. In the application, when the second transaction result is predicted, the influence of the first transaction result on the second transaction result is considered, and on the basis of taking the related information of the first goods source as the input data for predicting the second transaction result, the first feature vector capable of representing the first transaction result is added as the input data; and the accuracy of the second transaction result is improved, so that the accuracy of the transaction duration of the first goods source is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a transaction duration prediction method, device, electronic device and storage medium. Background Art

[0002] With the development of the Internet, transportation service platforms have emerged. The transportation service platform is a medium for matching users' transportation needs with the carrier's transportation resources. Users can send the cargo information they need to transport to the transportation service platform through the user client. The transportation service platform will display the cargo information to the carrier through the carrier's client. The carrier will match the cargo based on its own transportation resources to meet the user's transportation needs.

[0003] When users use the transportation service platform, the transportation service platform will provide users with a reference transportation price based on the supply information provided by the users. Users can bid based on the reference transportation price, or bid after adding a price to the reference transportation price to ensure that the supply of goods is transmitted in a timely manner. Accordingly, the transportation service platform will predict the transaction time of the supply of goods based on the user's bid and feedback the prediction results to the customer.

[0004] However, in the related art, the transaction time predicted by the transportation service platform is not accurate enough and cannot meet the user's usage needs. Summary of the invention

[0005] The present application provides a transaction duration prediction method, device, electronic device and storage medium. On the basis of using relevant information of a first source of goods as input data for predicting a second transaction result, a first feature vector capable of representing the first transaction result is added as input data, thereby improving the accuracy of the second transaction result, thereby improving the accuracy of the transaction duration of the first source of goods.

[0006] In a first aspect, the present application provides a transaction duration prediction method, comprising: receiving first information; wherein the first information includes source information of a first source of goods to be traded and an expected bid of a user for the first source of goods; inputting the first information into a feature extraction network in a target transaction duration prediction model, using the feature extraction network to perform feature extraction on the first information, and obtaining a first feature vector corresponding to the first information and a second feature vector corresponding to the first information; wherein the first feature vector is used to predict a first transaction result of the first source of goods within a first time period after going online under the expected bid; the second feature vector is used to predict a second transaction result of the first source of goods within a second time period after going online under the expected bid; the second time period is greater than the first time period. a duration; inputting the first feature vector into the first network of the target transaction duration prediction model, using the first network to map the first feature vector to a specific numerical range, and obtaining a first confidence level; wherein the first confidence level is used to indicate the first transaction result; inputting the first feature vector and the second feature vector into the second network of the target transaction duration prediction model, using the second network to aggregate the first feature vector and the second feature vector to obtain an aggregated vector, and mapping the aggregated vector to a specific numerical range to obtain a second confidence level; the second confidence level is used to indicate the second transaction result; based on the first transaction result and the second transaction result, determining the transaction duration of the first source of goods under the expected bid.

[0007] In some embodiments, determining the transaction time of the first source of goods at the expected bid based on the first transaction result and the second transaction result includes: if the first transaction result indicates that the first source of goods is traded within the first time period at the expected bid, and the second transaction result indicates that the first source of goods is traded within the second time period at the expected bid, then determining that the transaction time of the first source of goods is less than the first time period; if the first transaction result indicates that the first source of goods is not traded within the first time period at the expected bid, and the second transaction result indicates that the first source of goods is traded within the second time period at the expected bid, then the transaction time of the first source of goods is greater than the first time period and less than the second time period; if the first transaction result indicates that the first source of goods is not traded within the first time period at the expected bid, and the second transaction result indicates that the first source of goods is not traded within the second time period at the expected bid, then the transaction time of the first source of goods is greater than the second time period.

[0008] In some embodiments, the second network includes an aggregation module and a first normalization layer connected in sequence; wherein the aggregation module is used to receive the first feature vector and the second feature vector, concatenate the first feature vector and the second feature vector to obtain a first concatenated vector, and perform forward propagation and nonlinear processing on the first concatenated vector to obtain the aggregated vector, and input the aggregated vector to the first normalization layer; the first normalization layer is used to map the aggregated vector to the specific numerical range to obtain the second confidence.

[0009] In some embodiments, the feature extraction network includes a feature fusion subnetwork, a first neural network layer, and a second neural network layer; the feature fusion subnetwork is connected to the first neural network layer and the second neural network layer respectively; the number of the source of supply information is multiple; the feature fusion subnetwork is used to receive the expected bid and multiple source of supply information, perform feature extraction and aggregation operations on the expected bid and the multiple source of supply information, and obtain a third feature vector corresponding to the multiple source of supply information and a fourth feature vector corresponding to the multiple source of supply information; the first neural network layer is used to perform forward propagation and nonlinear processing on the third feature vector to obtain the first feature vector; the second neural network layer is used to perform forward propagation and nonlinear processing on the fourth feature vector to obtain the second feature vector.

[0010] In some embodiments, the feature fusion subnetwork includes multiple embedding layers, multiple third neural network layers and a fourth neural network layer; wherein the fourth neural network layer includes multiple input neurons, a first output neuron and a second output neuron; the number of the supply information is multiple; the multiple embedding layers are respectively connected to the multiple third neural network layers, and the multiple third neural network layers are respectively connected to the multiple input neurons of the fourth neural network layer; the multiple embedding layers are used to receive the expected bid and multiple supply information, convert the expected bid and the multiple supply information into corresponding embedding vectors, and input the embedding vectors into the corresponding third neural network layers; wherein one embedding layer corresponds to the expected bid or one supply information; the multiple third neural network layers are respectively used to forward propagate the received embedding vectors to obtain multiple fifth feature vectors, and input the multiple fifth feature vectors into the fourth neural network layer; the fourth neural network layer receives the multiple fifth feature vectors through the multiple input neurons, and after forward propagation and feature aggregation of the multiple fifth feature vectors, outputs the third feature vector through the first output neuron, and outputs the fourth feature vector through the second output neuron.

[0011] In some embodiments, the number of the second feature vectors is N, the number of the second networks is N, and the N second feature vectors are respectively used to predict the transaction results of the first source of goods within N second time periods under the expected bid; the N second feature vectors, the N second networks, and the N second time periods correspond to each other; wherein N is greater than 1, and the N second time periods increase in sequence; the method further includes: outputting the first feature vector and the first second feature vector to the first second network, using the first second network to obtain the first aggregation vector and the first second confidence; the first confidence is used to indicate the expected bid, the the second transaction result of the first source of goods within the 1st second time period; inputting the first feature vector and the 1st aggregation vector to the i-1th aggregation vector into the ith second network, and using the ith second network to obtain the ith aggregation vector and the ith second confidence; the ith second confidence is used to indicate the second transaction result of the first source of goods within the ith second time period under the expected bid; wherein i is greater than 1 and less than or equal to N; based on the first transaction result, the second transaction result of the first source of goods within the 1st second time period to the second transaction result of the first source of goods within the Nth second time period, determine the transaction time period of the first source of goods.

[0012] In some embodiments, before receiving the first information, the method further includes: obtaining sample data, the sample data including the first bid of the second source of goods that has been historically traded and the source information of the second source of goods; the sample data carries a first label and a second label, wherein the first label is used to characterize whether the second source of goods is traded within the first time period under the first bid, and the second label is used to characterize whether the second source of goods is traded within the second time period under the first bid; inputting the sample data into the feature extraction network in the initial transaction time prediction model to obtain a sixth feature vector corresponding to the sample data and a seventh feature vector corresponding to the sample data; inputting the sixth feature vector into the first network in the initial transaction time prediction model to obtain a third confidence; inputting the seventh feature vector and the sixth feature vector into the second network in the initial transaction time prediction model to obtain a fourth confidence; based on a first loss function between the third confidence and the first label and a second loss function between the fourth confidence and the second label, iteratively training the initial transaction time prediction model until an iteration stop condition is met to obtain the target transaction time prediction model.

[0013] In a second aspect, the present application provides a transaction duration prediction device, comprising: a module for executing the method described in the first aspect.

[0014] In a third aspect, the present application provides an electronic device, comprising a memory and a processor; the memory is configured to store computer program instructions, and the processor is configured to execute the computer program instructions, so that the electronic device implements the method of the first aspect.

[0015] In a fourth aspect, the present application provides a chip comprising an interface circuit and a logic circuit, wherein the interface circuit is used to receive signals from other chips outside the chip and transmit them to the logic circuit, and the signal from the logic circuit is sent to other chips outside the chip, and the logic circuit is used to implement the method of any embodiment of the first aspect.

[0016] In a fifth aspect, the present application provides a storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0017] In a sixth aspect, the present application provides a computer program product comprising instructions, and when a computer runs the computer program product, the computer implements the method of any embodiment of the first aspect.

[0018] The present application provides a transaction duration prediction method, device, electronic device and readable storage medium, wherein, in the scenario of predicting the transaction duration of the first source of goods, the relevant information of the first source of goods is subjected to feature extraction, feature vector fusion, feature vector mapping and other operations through the target transaction duration prediction model to predict the first transaction result of the first source of goods within the first time after being put on the shelf and the second transaction result of the first source of goods within the second time after being put on the shelf, and then the first transaction result and the second transaction result are used to jointly determine the transaction duration of the first source of goods. In the present application, when predicting the second transaction result of the first source of goods within the second time, considering the influence of the first transaction result on the second transaction result, on the basis of using the relevant information of the first source of goods as the input data for predicting the second transaction result, a first feature vector capable of characterizing the first transaction result is added as input data to improve the accuracy of the second transaction result. Based on this, the transaction duration of the first source of goods after being put on the shelf is determined using the first transaction result and the second transaction result to improve the accuracy of the transaction duration of the first source of goods. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of transaction duration prediction based on multiple binary classification models;

[0020] Figure 2 A schematic diagram of the implementation flow of the transaction duration prediction method provided in the embodiment of the present application;

[0021] Figure 3 A schematic diagram of the structure of the target transaction duration prediction model provided in the embodiment of the present application;

[0022] Figure 4 A schematic diagram of the structure of the second network provided in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of the implementation flow of the transaction duration prediction method provided in the embodiment of the present application;

[0024] Figure 6 A schematic diagram of the implementation process of the feature extraction network provided in the embodiment of the present application;

[0025] Figure 7 A schematic diagram of the structure of a feature fusion subnetwork provided in an embodiment of the present application;

[0026] Figure 8 A schematic diagram of the structure of the target transaction duration prediction model provided in the embodiment of the present application;

[0027] Fig. 9 A schematic diagram of the implementation flow of the transaction duration prediction method provided in the embodiment of the present application;

[0028] Fig.10 A schematic diagram of the structure of a transaction duration prediction device provided in an embodiment of the present application;

[0029] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In the scenario where the transportation service platform matches the user's transportation needs with the carrier's transportation resources, the transportation service platform will provide the user with a reference transportation price based on the source of goods information provided by the user. The user can bid based on the reference transportation price, or bid after adding a price to the reference price to ensure that the source of goods is transmitted in a timely manner; accordingly, the transportation service platform will predict the transaction time of the source of goods based on the bid price, and feedback the predicted transaction time to the customer. This allows the customer to determine the transportation order based on the predicted transaction time, or increase or reduce the price based on the current bid until the transportation order is determined.

[0031] In related technologies, the transportation service platform can be based on Figure 1 The transaction duration is determined in the manner shown.

[0032] Figure 1 This is a schematic diagram of transaction time prediction based on multiple binary classification models.

[0033] like Figure 1 As shown, the multiple binary classification models are binary classification model 1, binary classification model 2, ..., binary classification model K, and the multiple binary classification models are independent of each other.

[0034] Multiple binary classification models are used to predict the transaction results of users at their expected bids within multiple different time periods after the supply goes online.

[0035] Among them, binary classification model 1 is used to predict the transaction results within time 1 after the supply is online, binary classification model 2 is used to predict the transaction results within time 2 after the supply is online,..., binary classification model K is used to predict the transaction results within time K after the supply is online.

[0036] Duration 1, duration 2, and duration K increase in sequence.

[0037] Before using multiple binary classification models, they are trained separately. Different binary classification models have different corresponding labels for the same training samples.

[0038] Exemplarily, binary classification model 1 and binary classification model 2 are used to illustrate different labels corresponding to the same training sample.

[0039] When the input sample of the binary classification model 1 is the price of the goods obtained by statistics within duration 1, the corresponding label indicates that the goods are sold within duration 1; when the input sample of the binary classification model 1 is the price of the goods obtained by statistics when the duration is greater than 1 and less than duration 2, the corresponding label indicates that the goods are not sold within duration 1.

[0040] The input sample of binary classification model 2 is the price of the goods obtained by statistics within duration 1, and the corresponding label indicates that the transaction is completed within duration 2; when the input sample of binary classification model 2 is the price of the goods obtained by statistics that is greater than duration 1 and less than duration 2, the corresponding label indicates that the goods are sold within duration 2.

[0041] In this way, different binary classification models have different labels for the same training samples, and different binary classification models have different convergence conditions during training, and different binary classification models have different model parameters after training. In this way, multiple binary classification models are trained independently, which may lead to conflicts in the output results of multiple binary classification models.

[0042] For example, for a certain source of goods, its price is A, binary classification model 1 predicts that it will be sold within duration 1 after going online, and binary classification model 2 predicts that it will not be sold within duration 2 after going online. However, the actual situation is: duration 2 includes duration 1, if source 1 is sold within duration 1, it can definitely be sold within duration 2.

[0043] It can be seen that the use of Figure 1 The method of predicting the transaction time has the defects of inaccurate prediction and the prediction result not being consistent with the actual situation.

[0044] Based on this, an embodiment of the present application provides a transaction time prediction method. On the basis of using the relevant information of the first source of goods as input data for predicting the second transaction result, a first feature vector that can characterize the first transaction result is added as input data, thereby improving the accuracy of the second transaction result, thereby improving the accuracy of the transaction time of the first source of goods.

[0045] Figure 2 The following is a schematic diagram of the implementation process of the transaction duration prediction method provided in the embodiment of the present application. Figure 2 As shown, the method includes:

[0046] S201. Receive first information.

[0047] The first information includes the user's expected bid and supply information for the first supply source.

[0048] The executor of the embodiment of the present application may be a transportation service platform that provides matching services between the user's transportation needs and the carrier's transportation resources.

[0049] The expected bid is provided by the user, and the source of goods information is provided by the user and obtained from the statistical data of the transportation service platform.

[0050] The expected bid is determined by the user based on the reference bid determined by the transportation service platform. Specifically, the transportation service platform provides service access to the user through the user program (Application, APP) of the user client, and the user enters the attribute information of the first source of goods to be transported on the APP; the transportation service platform determines a reference bid for the user based on the attribute information of the first source of goods, and displays the reference bid to the user through the APP. The user increases or decreases the price based on the reference bid to obtain the expected bid, and sends the expected bid to the first source of goods transportation platform through the APP.

[0051] There may be multiple sources of supply information, including: attribute information of the first source of supply, historical statistical information of the transportation service platform, etc.

[0052] Among them, the attribute information of the first source of cargo includes: delivery time, weight of the first source of cargo, size of the truck required to transport the first source of cargo, transportation distance, latitude of the detailed address of the transportation starting point, latitude of the detailed address of the transportation destination, longitude of the detailed address of the transportation starting point, longitude of the detailed address of the transportation destination, identification of the secondary category, maximum volume of the first source of cargo, city where the transportation starting point is located, city where the transportation destination is located, whether it is a fixed-price first source of cargo, whether the entrusted cargo is transmitted by default, whether it is a preferred car, whether it is negotiated by phone, loading and unloading methods, the position of the current date in the week (for example, the current date is the first day of the week), the number of vehicle models, identification of the truck type, identification of the first-level category, identification of the departure place, identification of the destination, load rate, heavy cargo mark of the same consignor (real-time feature), heavy cargo of different consignees (real-time feature), number of vehicle lengths, and time difference between loading and shipping.

[0053] Historical statistical information includes: the transaction rate of the secondary category of the transmission line of the first source of cargo to be traded, the transaction rate of the route in the historical time, the ton-kilometer price of the departure city, the kilometer price of the departure city, the supply-demand ratio of the route in the historical time, the telephone negotiation transaction rate of the city-level route in the historical time, the historical city-level route entrustment transaction rate, the historical provincial route entrustment transaction rate, the historical transaction rate of the city where the starting place is located, the historical vehicle length telephone negotiation transaction rate, the historical transaction rate of the minimum vehicle length, the average exposure of the first source of cargo with the minimum vehicle length in the historical time, the average clicks on the first source of cargo with the minimum vehicle length in the historical time, the supply-demand ratio of the vehicle length of the route in the historical time: the number of telephone drivers / the total amount of the first source of cargo, the supply-demand ratio of the vehicle type of the route in the historical time, the number of telephone drivers / the total amount of the first source of cargo, the historical online and offline transaction rate of route + vehicle length, the historical transaction rate of packaging method, the number of shipments, the number of entrustments, the historical number of shipments, the number of transactions, the transaction rate, the user's exposure click rate in the historical time, etc.

[0054] It should be noted that the above historical statistical information is only part of the data collected on the transportation service platform. In actual use, more statistical data can be used to predict the transaction time.

[0055] The transportation service platform predicts the transaction time of the first source of goods under the expected bid based on the first information. If the user is satisfied with the transaction time, the order information can be generated through the APP, and the order information of the first source of goods to be traded can be published on the carrier client. If a carrier accepts the order through the carrier client, the first source of goods is traded and the user's transportation needs are met.

[0056] S202: input the first information into a feature extraction network in a target transaction duration prediction model, and use the feature extraction network to extract features from the first information to obtain a first feature vector corresponding to the first information and a second feature vector corresponding to the first information.

[0057] Next, through Figure 3The structure of the target transaction time prediction model is explained. Figure 3 A schematic diagram of the structure of the target transaction duration prediction model provided in the embodiment of the present application; Figure 3 As shown, the target transaction duration prediction model 30 includes a feature extraction network 31, a first network 32 and a second network 33.

[0058] Among them, the feature extraction network 31 is used to extract features of the source of goods information, and its internal output layer includes a first output neuron and a second output neuron. The feature extraction network 31 outputs a first feature vector and a second feature vector through the first output neuron and the second output neuron respectively.

[0059] The specific structure of the feature extraction network 31 is as follows: Figure 6 The explanation is not repeated here.

[0060] The first eigenvector is used to predict the first transaction result of the first source of goods to be traded within the first period of time after going online under the expected bid.

[0061] The second feature vector is used to predict the second transaction result of the first source of goods to be traded within the second time period after going online under the expected bid. In other words, the transaction result of the first source of goods within the second time period can also be obtained based only on the second feature vector.

[0062] Among them, going online means that, assuming that the user generates a transaction order through the APP at the currently determined expected price, and the transaction order is published on the APP, that is, the starting time of the first period and the second period are both the moment when the transaction order corresponding to the first source of goods is published on the APP.

[0063] The transaction results include completed and unsuccessful. If the order is completed, it means that a carrier has accepted the order and the carrier can transport the user's first source of goods based on its own transmission resources; if the order is unsuccessful, it means that no carrier has accepted the order and the user's transportation needs have not been met.

[0064] S203, inputting the first feature vector into the first network 32 of the target transaction duration prediction model 30, and using the first network 32 to map the first feature vector to a specific numerical range to obtain a first confidence level; wherein the first confidence level is used to indicate a first transaction result.

[0065] The first network 32 may include a normalization layer. In some embodiments, the first network 32 may map the first feature vector to a range of 0 to 1 through a softmax function or a logit function to obtain the first confidence level, which represents the probability of a transaction of the first source of goods within the first time period under the expected bid.

[0066] The first source of goods transportation platform sets a probability threshold. If the first confidence level is greater than the probability threshold, it is determined that the first source of goods is traded within the first time period. If the first confidence level is greater than the probability threshold, it is determined that the first source of goods is not traded within the first time period. It can be seen that the first confidence level can indicate the first transaction result.

[0067] S204, inputting the first feature vector and the second feature vector into the second network 33 of the target transaction duration prediction model 30, using the second network 33 to aggregate the first feature vector and the second feature vector to obtain an aggregated vector, and mapping the aggregated vector to a specific numerical range to obtain a second confidence level that the first source of goods is traded within the second duration.

[0068] The first eigenvector and the second eigenvector are both high-dimensional vectors.

[0069] Specifically, Figure 4 A schematic diagram of the structure of the second network 33 provided in the embodiment of the present application is shown in FIG. Figure 4 As shown, the second network 33 includes an aggregation module 331 and a first normalization layer 332 connected in sequence; wherein the aggregation module 331 specifically includes a merging module, a multilayer perceptron (MLP) and a forward propagation layer.

[0070] Among them, MLP is a kind of feedforward neural network. According to the position of different layers, MLP can be divided into input layer, hidden layer and output layer. Starting from the neurons in the first hidden layer, each layer of neurons receives the input of the previous layer, performs nonlinear transformation on the input of the previous layer, and outputs the value of the next layer until it reaches the output layer. Among them, the layers are fully connected, that is, any neuron in any layer must be connected to any neuron in the next layer. In this way, MLP can fully aggregate the input vector elements to extract the input source information.

[0071] For each hidden layer, the input vector can be processed by the following formula:

[0072] y1=α1(W*x1+b1) Formula 1;

[0073] Among them, y1 is the output vector, x1 is the input vector, b1 is the offset vector, W is the weight matrix, and α1() is the activation function.

[0074] Since there are more hidden layers and more W and b1, the model has a higher complexity and can complete more complex learning tasks with higher performance.

[0075] exist Figure 4In the embodiment, the merging module is used to receive the first eigenvector and the second eigenvector, concatenate the first eigenvector and the second eigenvector to obtain a concatenated vector, and send the concatenated vector to the MLP. The concatenated vector is forward propagated and nonlinearly transformed by the MLP to fully aggregate the vectors received by each neuron and add nonlinear features. The vector output by the MLP is further feature extracted by the forward propagation layer to obtain an aggregated vector, and the aggregated vector is input to the first normalization layer 332.

[0076] The first normalization layer 332 is used to map the aggregated vector to the specific numerical range to obtain a second confidence level.

[0077] Specifically, the first normalization layer 332 can map the aggregate vector to the interval of 0 to 1 through a softmax function or a logit function to obtain the second confidence, which represents the probability of the first source of goods being traded within the second time period under the expected bid.

[0078] The first source of goods transportation platform is set with a probability threshold. If the second confidence level is greater than the probability threshold, it is determined that the first source of goods is sold within the second time period. If the second confidence level is less than the probability threshold, it is determined that the first source of goods is not sold within the second time period. It can be seen that the first confidence level can indicate the second transaction result.

[0079] S205: Determine the transaction duration of the first source of goods based on the first transaction result and the second transaction result.

[0080] As described above, the second duration is greater than the first duration, the second duration and the first duration start at the same time, and the second duration includes the first duration. It can be seen that the first transaction result depends on the second transaction result; specifically, if the first transaction result is a successful transaction, the second transaction result must also be a successful transaction.

[0081] In this application, when predicting the second transaction result of the first source of goods within the second time period, considering the impact of the first transaction result on the second transaction result, on the basis of using the second feature vector obtained based on the first information of the first source of goods as the input data for predicting the second transaction result, the first feature vector that can characterize the first transaction result is added as input data to improve the accuracy of the second transaction result. Based on this, the first transaction result and the second transaction result are used to determine the transaction time of the first source of goods after being put on the shelf, so as to improve the accuracy of the transaction time of the first source of goods.

[0082] Since the dependency between the first transaction result within the first time period and the second transaction result within the second time period is taken into account when determining the transaction duration, the situation in the related art where the first transaction result is a successful transaction but the second transaction result is a failed transaction is avoided. Based on this, the first transaction result and the second transaction result include the following situations:

[0083] Case 1: The first transaction result is a successful transaction, and the second transaction result is a successful transaction.

[0084] Case 2: The first transaction result is a failed transaction, and the second transaction result is a successful transaction.

[0085] Case 3: The first transaction result is transaction failure, and the second transaction result is transaction failure.

[0086] Next, through Figure 5 The embodiment specifically describes how to determine the transaction duration of the first source of goods based on the first transaction result and the second transaction result.

[0087] Figure 5 A schematic diagram of the implementation flow of the transaction duration prediction method provided in the embodiment of the present application is shown in FIG. Figure 5 As shown, S205 can be implemented by the following steps:

[0088] Step a1, determine which situation the first transaction result and the second transaction result belong to; if it is situation 1, execute step a21; if it is situation 2, execute step a22; if it is situation 3, execute step a23.

[0089] Step a21: Determine whether the transaction duration of the first source of goods is less than the first duration.

[0090] It can be understood that in case 1, the second transaction result indicates that the first source of goods is expected to be sold within the second time period; the first transaction result indicates that the first source of goods is expected to be sold within the first time period. Then the transportation service platform can determine that the first source of goods is specifically sold within the first time period included in the second time period.

[0091] In this case, the transportation service platform can display the maximum transaction duration to the user through the APP. In case 1, the transportation service platform can display the maximum transaction duration to the user through the APP as the first duration.

[0092] Step a22: Determine that the transaction time of the first source of goods is after the first time period and before the second time period.

[0093] Specifically, the second transaction result indicates that the first source of goods is expected to be sold within the second time period, but the first transaction result indicates that the first source of goods is expected to not be sold within the first time period. Then the transportation service platform can determine that the first source of goods is sold after the first time period and before the second time period.

[0094] In this case, the transportation service platform can display the maximum transaction time and the minimum transaction time to the user through the APP. In case 1, the transportation service platform can display to the user through the APP that the minimum transaction time is the first time and the maximum transaction time is the second time.

[0095] Step a23, determining that the transaction time of the first source of goods is later than the second time.

[0096] Specifically, the first transaction result indicates that the first source of goods is expected to not be sold within the first time period, and the second transaction result indicates that the first source of goods is expected to not be sold within the second time period; based on this, the transportation service platform can determine that the first source of goods is expected to be sold after the second time period.

[0097] In this case, the transportation service platform can display the minimum transaction time to the user through the APP. Specifically, the transportation service platform displays the minimum transaction time to the user as the second time.

[0098] In this embodiment, the specific transaction time of the first source of goods can be determined through different transaction results of the first source of goods in different time periods, and the corresponding transaction time can be displayed to the user, so that the user can increase the price / reduce the price / place an order to release the first source of goods based on their own needs, thereby meeting the user's transportation needs.

[0099] As described above, the target transaction time prediction model 30 includes a feature extraction network 31. Figure 6 The embodiment describes the structure of the feature extraction network 31 and how to obtain the first feature vector and the second feature vector based on the feature extraction network 31 .

[0100] Figure 6 Schematic diagram of the structure of the feature extraction network 31 provided in the embodiment of the present application. Figure 6 As shown, the feature extraction network 31 includes a feature fusion subnetwork 311, a first neural network layer 312 and a second neural network layer 313; wherein the feature fusion subnetwork 311 is connected to the first neural network layer 312 and the second neural network layer 313 respectively.

[0101] As mentioned above, the first information includes the expected bid and multiple sources of supply information.

[0102] In some embodiments, the multiple sources of supply information specifically include: attribute information of the first source of supply and historical statistical information of the transportation service platform. The feature fusion subnetwork 311 performs feature extraction and aggregation operations on the expected bid, the attribute information of the first source of supply, and the historical statistical information, so as to obtain a third feature vector containing the expected bid and features of each source of supply information and a fourth feature vector containing the expected bid and features of each source of supply information.

[0103] Among them, the first neural network layer 312 is composed of at least one group of MLP and forward propagation layers, each group of MLP and forward propagation layers are connected in series, and the first neural network layer 312 is used to perform forward propagation and nonlinear processing on the third eigenvector to obtain the first eigenvector.

[0104] The second neural network layer 313 also includes at least one set of MLP and forward propagation layers. The second neural network layer 313 is used to perform forward propagation and nonlinear processing on the fourth eigenvector to obtain the second eigenvector.

[0105] The structure of the first neural network layer 312 is similar to the structure and function of the second neural network layer 313. Next, taking the first neural network layer 312 as an example, it is explained how the first neural network layer 312 processes the third eigenvector.

[0106] Among them, an MLP includes at least one activation layer, and the first neural network layer 312 includes multiple activation layers. For each activation layer, the activation function of the following formula 2 is used for the input feature vector, so that the influence of the price feature on the result value tends to be monotonic, that is, the higher the price, the higher the first confidence and the second confidence.

[0107]

[0108] Wherein, c is the output of the activation layer, x2 is the input of the activation layer, when the activation layer is the first layer of the first neural network layer 312, x2 is the third eigenvector obtained through forward propagation; ReLu() is a nonlinear activation function, a is a hyperparameter, b is a parameter fitted based on a, and b satisfies the following formula three.

[0109]

[0110] Where b[k]≤b[k+1]; K is an integer greater than 2, and α[k] is a parameter that ensures that the output of the activation function in the next MLP is greater than the input through interpolation.

[0111] Before training, the first neural network layer 312 fixes the input range of each activation layer to [amin, amax] and trains the output parameter b of the activation layer. For monotonic input, a linear inequality constraint is added to constrain the activation layer parameter b to be between 0 and 1 to constrain the nonlinear activation function to be monotonic.

[0112] The first neural network layer 312 includes multiple activation layers, wherein the input of the nonlinear activation function of each activation layer is based on the output of the nonlinear activation function of the previous activation layer, and each nonlinear activation function is monotonically increasing. In this way, the monotonicity of the target transaction time prediction network can be guaranteed, so that when the bid input into the target transaction time prediction network is higher, the transaction rate is higher, which is in line with the actual situation.

[0113] Next, through Figure 7 The embodiment explains how to obtain the third feature vector and the fourth feature vector through the feature fusion sub-network 311.

[0114] Figure 7 This is a schematic diagram of the structure of the feature fusion sub-network 311 provided in the embodiment of the present application. Figure 7 As shown, the feature fusion subnetwork 311 includes multiple embedding layers, multiple third neural network layers and one fourth neural network layer.

[0115] The fourth neural network layer includes a plurality of input neurons, a first output neuron and a second output neuron.

[0116] The multiple embedding layers are respectively connected to the multiple third neural network layers, and the multiple third neural network layers are respectively connected to the multiple input neurons of the fourth neural network layer. The multiple embedding layers correspond one to one to the multiple third neural network layers.

[0117] Multiple embedding layers are used to receive the expected bid and multiple supply information, convert the expected bid and multiple supply information into corresponding embedding vectors, and input the embedding vectors into the corresponding third neural network layer.

[0118] The multiple third neural network layers are respectively used to forward propagate the received embedding vectors to obtain multiple fifth eigenvectors, and input the multiple fifth eigenvectors into the fourth neural network layer.

[0119] The fourth neural network layer receives multiple fifth eigenvectors through multiple input neurons respectively, performs forward propagation and feature aggregation on the multiple fifth eigenvectors, outputs the third eigenvector through the first output neuron, and outputs the fourth eigenvector through the second output neuron.

[0120] Among them, one embedding layer corresponds to an expected bid or a source of supply information.

[0121] It should be noted that the above-mentioned source of goods information includes the attribute information of the first source of goods, and the historical statistical information of the transportation service platform for the first source of goods is only an example; in actual use, the number of source of goods information is not limited to the above two types, and may also include more than two types.

[0122] Next, the structure and function of the feature fusion sub-network 311 are explained by way of example.

[0123] For example, in Figure 7 In the example, the number of embedded layers is Q, the number of supply information is Q-1, embedded layer 1 corresponds to the expected bid, and embedded layers 2 to Q correspond to Q-1 supply information respectively; then embedded layer 1 receives the expected bid, embedded layer 2 receives supply information 1, embedded layer 3 receives supply information 2, and embedded layer Q receives supply information Q-1. Q is greater than 2.

[0124] Embedding layer 1 converts the expected bid into embedding vector 1, embedding layer 2 converts the supply information 1 into embedding vector 2, embedding layer 3 converts the supply information 2 into embedding vector 3, and embedding layer Q converts the supply information Q-1 into embedding vector Q.

[0125] The embedding layer can use different conversion methods for different first information. For attribute information, it can be converted into corresponding identifiers first, and then the identifiers can be converted into embedding vectors; for historical statistical information, due to the large amount of data, the data can be bucketed before conversion.

[0126] The third neural network layer 1 performs forward propagation on the received embedding vector 1 to obtain the fifth eigenvector 1; the third neural network layer 2 performs forward propagation on the received embedding vector 2 to obtain the fifth eigenvector 2; the third neural network layer Q performs forward propagation on the received embedding vector Q to obtain the fifth eigenvector Q.

[0127] The fifth eigenvector 1 to the fifth eigenvector Q are all transmitted to the fourth neural network layer and are received by input neurons 1 to input neurons Q of the fourth neural network layer, respectively.

[0128] In this embodiment, there are multiple hidden layers between the input neurons and the output neurons of the fourth neural network layer. Since the hidden layers of the fourth neural network layer are fully connected, after the multiple hidden layers perform feature fusion and forward propagation on the fifth feature vector 1 to the fifth feature vector Q, the third feature vector output by the first output neuron carries the features of the expected bid, supply information 1 to supply information Q; the fourth feature vector output by the second output neuron also carries the features of the expected bid, supply information 1 to supply information Q. In this way, the function of aggregating the features of the expected bid and multiple supply information is realized.

[0129] It should be noted that Figure 3 The target duration prediction model of the embodiment is only an example of the embodiment of the present application. In some embodiments, the fourth neural network layer may include multiple second output neurons; the target transaction duration prediction model may include multiple second neural network layers and multiple second networks. In this way, the fourth neural network layer outputs multiple fourth feature vectors through multiple second output neurons, and the multiple fourth feature vectors are respectively input to multiple second neural network layers, and the multiple second neural networks respectively output multiple second feature vectors.

[0130] Among them, the multiple second output neurons, the multiple fourth eigenvectors, the multiple second neural network layers, the multiple second eigenvectors, and the multiple second networks correspond one to one.

[0131] The multiple second feature vectors are respectively used to predict the second transaction results within multiple second time periods after the first source of goods goes online under the expected bid.

[0132] Assume that the number of the second output neuron, the fourth eigenvector, the second neural network layer, the second eigenvector, and the second network is N.

[0133] The N second durations include the first second duration, the second second duration, ..., and the Nth second duration, which are respectively represented by second duration 1, second duration 2, ..., and second duration N. N is greater than 1.

[0134] The N second eigenvectors include the first second eigenvector, the second second eigenvector, ..., the Nth second eigenvector; which are represented by second eigenvector 1, second eigenvector 2, ..., second eigenvector N, respectively.

[0135] The N second networks include a first second network, a second second network, ..., and an Nth second network; they are represented by second network 1, second network 2, ..., and second network N, respectively.

[0136] The N second networks correspond to N aggregation vectors, and the N aggregation vectors include a first aggregation vector, a second aggregation vector, ..., and an Nth aggregation vector; which are represented by aggregation vector 1, aggregation vector 2, ..., and aggregation vector N, respectively.

[0137] The N aggregation vectors correspond to N second confidences, which include the first second confidence, the second second confidence, ..., the Nth second confidence; which are represented by second confidence 1, second confidence 2, ..., second confidence N, respectively.

[0138] The N second confidences, the N second time durations, and the N second transaction results correspond one to one; the N second time durations increase sequentially, and the N second confidences respectively represent the second transaction results of the first source of goods in the corresponding second time duration under the expected bid.

[0139] The N second durations may be represented as second duration 1, second duration 2, ..., second duration N.

[0140] The N second transaction results can be expressed as second transaction result 1, second transaction result 2, ..., second transaction result N.

[0141] Figure 8 This is an example of a target transaction duration prediction model in the case of the above-mentioned multiple second networks provided in an embodiment of the present application. Figure 8 The target transaction duration prediction model is implemented by following steps b1 to b2 to implement S204:

[0142] Step b1, for the second network 1, such as Figure 2As shown in the embodiment, the first feature vector and the second feature vector 1 are input into the second network 1, and the second feature vector and the first feature vector are aggregated by the second network 1 to obtain an aggregated vector 1, and the aggregated vector 1 is mapped to a specific numerical range to obtain a second confidence 1.

[0143] Step b2: for the second network i, the first feature vector, the aggregated vector 1 to the aggregated vector i-1, and the second feature vector i are input into the second network i, and the first feature vector, the aggregated vector 1 to the aggregated vector i-1, and the second feature vector i are aggregated by the second network i to obtain the aggregated vector i, and the aggregated vector i is mapped to a specific numerical range to obtain the second confidence i.

[0144] The second confidence i is used to characterize the transaction result within the i-th second period after the first source of goods goes online under the expected bid.

[0145] Specifically, the first eigenvector, the aggregate vectors 1 to i-1, and the second eigenvector i are concatenated through the second network i to obtain a corresponding first concatenated vector, and the corresponding first concatenated vector is forward propagated and nonlinearly processed to obtain an aggregate vector i.

[0146] The aggregation vector i is used to represent the transaction result within the i-th second time period after the first source of goods goes online under the expected bid.

[0147] Wherein, i is greater than 1 and less than or equal to N.

[0148] When executing step a2, i can be taken from 2 to N in sequence, and the second transaction result 2 within the 2nd second time period to the second transaction result N within the Nth second time period can be calculated in sequence.

[0149] Specifically, the first feature vector, the aggregation vector 1 and the second feature vector 2 are first aggregated to obtain the aggregation vector 2, and the aggregation vector 2 is mapped to a specific numerical range to obtain the second transaction result 2.

[0150] Next, the first feature vector, the aggregated vector 1, the aggregated vector 2, and the second feature vector 3 are aggregated to obtain the aggregated vector 3, and the aggregated vector 3 is mapped to a specific numerical range to obtain the second transaction result 3.

[0151] The method for calculating the second transaction results 4 to N is similar to the method for calculating the second transaction results 2 and the second transaction results 3 described above, and will not be repeated here.

[0152] After obtaining the first transaction result and the second transaction result 1 to the second transaction result N, the transaction time of the first source of goods can be calculated based on the first transaction result and the second transaction result 1 to the second transaction result N.

[0153] Specifically, the first transaction result, the second transaction result 1 to the second transaction result N are arranged in order from small to large according to the duration. If the first M transaction results (including the first transaction result and the second transaction result) indicate a transaction, and the remaining transaction results indicate an unsuccessful transaction, then it is determined that the transaction duration of the first source of goods is less than the second duration corresponding to the Mth transaction result. Where M is greater than 0 and less than N.

[0154] If the first M transaction results (including the first transaction result and the second transaction result) indicate that the transaction was not completed, and the remaining transaction results indicate that the transaction was completed, it is determined that the transaction duration of the first source of goods is greater than the second duration corresponding to the Mth transaction result.

[0155] If all transaction results (including the first transaction result and the second transaction result) indicate that the transaction has not been completed, the transaction result of the first source of goods is greater than the second time length corresponding to the Nth transaction result.

[0156] exist Figure 8 In the figure, the second transaction result of the current second time length is aggregated based on the first feature vector and the second feature vector corresponding to each transaction time length before the second transaction time length. It can be seen that when determining the second transaction result of the current second time length, the dependency between the transaction results corresponding to the first time length and the previous second time length and the transaction result of the current second time length is taken into account. In this way, the determined second transaction result is more accurate, which is conducive to more accurate determination of the transaction time length.

[0157] The target transaction time prediction model is obtained by training the initial transaction time prediction model. Fig. 9 The embodiment illustrates a training method for an initial transaction duration prediction model.

[0158] Fig. 9 A schematic diagram of the implementation flow of the transaction time prediction method provided in the embodiment of the present application; Fig. 9 As shown, before S201, it also includes S101 to S105:

[0159] S101. Obtain sample data, where the sample data includes a first bid of a second source of goods that has been traded historically and source information of the second source of goods.

[0160] The sample data carries a first label and a second label. The first label is used to indicate whether the second source of goods is sold within a first period of time under a first bid. The second label is used to indicate whether the second source of goods is sold within a second period of time under the first bid.

[0161] In some embodiments, the transportation service platform may obtain the transaction time of the second source of goods, and generate a first label and a second label based on the transaction time.

[0162] For example, if the first duration is 5 minutes, the second duration is 10 minutes, and the actual transaction duration of the second source of goods is 8 minutes, then the first tag indicates that the second source of goods was not traded within 5 minutes, and the second tag indicates that the second source of goods was traded within 10 minutes.

[0163] The label can be represented by the numbers 0 and 1. 1 means the transaction is completed, and 0 means it is not completed.

[0164] S102: Input the sample data into a feature extraction network in an initial transaction duration prediction model to obtain a sixth feature vector corresponding to the sample data and a seventh feature vector corresponding to the sample data.

[0165] It should be noted that the model structure of the initial transaction time prediction model is exactly the same as the model structure of the target transaction time prediction model. The purpose of training the initial transaction time prediction model is to optimize the model parameters in the initial transaction time prediction model.

[0166] Specifically, a feature extraction network is used to perform feature aggregation on the price information and each source of supply information in the sample data, and the obtained sixth feature vector and seventh feature vector both contain the features of the price information and each source of supply information.

[0167] S103: Input the sixth feature vector into the first network in the initial transaction duration prediction model to obtain a third confidence level.

[0168] Specifically, the first network may map the sixth eigenvector to a specific numerical range to obtain a third confidence level.

[0169] The third confidence level is used to characterize the probability of a transaction being completed within the first period of time after the second source of goods is launched under the first bid.

[0170] S104: Input the sixth eigenvector and the seventh eigenvector into the second network in the initial transaction duration prediction model to obtain a fourth confidence level.

[0171] Specifically, after the second network aggregates the sixth eigenvector and the seventh eigenvector, the aggregated vectors are mapped to a specific numerical range to obtain a fourth confidence level.

[0172] Specifically, the fourth confidence level is used to characterize the probability of a transaction being completed with the second source of goods within the second period of time after the launch of the product under the first bid.

[0173] S105. Based on the first loss function between the third confidence and the first label and the second loss function between the fourth confidence and the second label, the initial transaction time prediction model is iteratively trained until the iteration stop condition is met to obtain the target transaction time prediction model.

[0174] It should be noted that the above is a step of training the initial transaction duration prediction model when the number of second durations is 1. When the number of second durations is greater than 1, each second duration corresponds to a second label, and the initial transaction duration prediction model is as follows: Figure 8 As shown, for Figure 8 The training method of the initial duration prediction model is similar to that of the present embodiment and will not be described in detail here.

[0175] In this embodiment, during training, the iteration stop condition of the initial transaction duration prediction model is determined based on the loss function corresponding to the first label and the second label, taking into account the dependency relationship between the first label and the second label. In this way, the transaction duration predicted by the trained target transaction duration prediction model is more accurate.

[0176] Based on the foregoing embodiments, an embodiment of the present application provides a transaction duration prediction device, which includes the modules included and the units included in the modules, which can be implemented by a processor; of course, it can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0177] Fig.10 A schematic diagram of the structure of the transaction duration prediction device provided in the embodiment of the present application is shown as follows: Fig.10 As shown, the transaction duration prediction device 100 includes a receiving module 1001, an input module 1002 and a determination module 1003, wherein:

[0178] The receiving module 1001 is used to receive first information; wherein the first information includes the supply information of the first supply source to be traded and the user's expected bid for the first supply source.

[0179] Input module 1002 is used to input the first information into a feature extraction network in a target transaction duration prediction model, use the feature extraction network to extract features from the first information, and obtain a first feature vector corresponding to the first information and a second feature vector corresponding to the first information; wherein the first feature vector is used to predict a first transaction result within a first duration after the first source of goods is launched at the expected bid; the second feature vector is used to predict a second transaction result within a second duration after the first source of goods is launched at the expected bid; the second duration is greater than the first duration; input the first feature vector into the first network of the target transaction duration prediction model, use the first network to map the first feature vector to a specific numerical range, and obtain a first confidence; wherein the first confidence is used to indicate the first transaction result; input the first feature vector and the second feature vector into the second network of the target transaction duration prediction model, use the second network to aggregate the first feature vector and the second feature vector to obtain an aggregate vector, and map the aggregate vector to a specific numerical range to obtain a second confidence; the second confidence is used to indicate the second transaction result.

[0180] The determination module 1003 is used to determine the transaction time of the first source of goods at the expected bid based on the first transaction result and the second transaction result.

[0181] In some embodiments, the determination module 1003 is specifically used to determine that the transaction time of the first source of goods is less than the first time period if the first transaction result indicates that the first source of goods is sold within the first time period under the expected bid, and the second transaction result indicates that the first source of goods is sold within the second time period under the expected bid; if the first transaction result indicates that the first source of goods is not sold within the first time period under the expected bid, and the second transaction result indicates that the first source of goods is sold within the second time period under the expected bid, then the transaction time of the first source of goods is greater than the first time period and less than the second time period; if the first transaction result indicates that the first source of goods is not sold within the first time period under the expected bid, and the second transaction result indicates that the first source of goods is not sold within the second time period under the expected bid, then the transaction time of the first source of goods is greater than the second time period.

[0182] In some embodiments, the second network includes an aggregation module and a first normalization layer connected in sequence; wherein the aggregation module is used to receive the first feature vector and the second feature vector, concatenate the first feature vector and the second feature vector to obtain a first concatenated vector, and perform forward propagation and nonlinear processing on the first concatenated vector to obtain the aggregated vector, and input the aggregated vector to the first normalization layer; the first normalization layer is used to map the aggregated vector to the specific numerical range to obtain the second confidence.

[0183] In some embodiments, the feature extraction network includes a feature fusion subnetwork, a first neural network layer, and a second neural network layer; the feature fusion subnetwork is connected to the first neural network layer and the second neural network layer respectively; the number of the source of supply information is multiple; the feature fusion subnetwork is used to receive the expected bid and multiple source of supply information, perform feature extraction and aggregation operations on the expected bid and the multiple source of supply information, and obtain a third feature vector corresponding to the multiple source of supply information and a fourth feature vector corresponding to the multiple source of supply information; the first neural network layer is used to perform forward propagation and nonlinear processing on the third feature vector to obtain the first feature vector; the second neural network layer is used to perform forward propagation and nonlinear processing on the fourth feature vector to obtain the second feature vector.

[0184] In some embodiments, the feature fusion subnetwork includes multiple embedding layers, multiple third neural network layers and a fourth neural network layer; wherein the fourth neural network layer includes multiple input neurons, a first output neuron and a second output neuron; the number of the supply information is multiple; the multiple embedding layers are respectively connected to the multiple third neural network layers, and the multiple third neural network layers are respectively connected to the multiple input neurons of the fourth neural network layer; the multiple embedding layers are used to receive the expected bid and multiple supply information, convert the expected bid and the multiple supply information into corresponding embedding vectors, and input the embedding vectors into the corresponding third neural network layers; wherein one embedding layer corresponds to the expected bid or one supply information; the multiple third neural network layers are respectively used to forward propagate the received embedding vectors to obtain multiple fifth feature vectors, and input the multiple fifth feature vectors into the fourth neural network layer; the fourth neural network layer receives the multiple fifth feature vectors through the multiple input neurons, and after forward propagation and feature aggregation of the multiple fifth feature vectors, outputs the third feature vector through the first output neuron, and outputs the fourth feature vector through the second output neuron.

[0185] In some embodiments, the number of the second feature vectors is N, the number of the second networks is N, and the N second feature vectors are respectively used to predict the transaction results of the first source of goods within N second time periods under the expected bid; the N second feature vectors, the N second networks, and the N second time periods correspond to each other; wherein N is greater than 1, and the N second time periods increase in sequence; the input module 1002 is further used to output the first feature vector and the first second feature vector to the first second network for the first first feature vector, and use the first second network to obtain the first aggregation vector and the first second confidence; the first confidence is used to indicate the first source of goods under the expected bid. The second transaction result of the source of goods within the 1st second time period; for the i-th first feature vector, the first feature vector, the 1st aggregation vector to the i-1th aggregation vector are input into the i-th second network, and the i-th second network is used to obtain the i-th aggregation vector and the i-th second confidence; the i-th second confidence is used to indicate the second transaction result of the first source of goods within the i-th second time period under the expected bid; wherein i is greater than 1 and less than or equal to N; a determination module 1003 is used to determine the transaction time period of the first source of goods based on the first transaction result, the second transaction result of the first source of goods within the 1st second time period to the second transaction result of the first source of goods within the Nth second time period.

[0186] It should be noted that in the embodiments of the present application, if the above-mentioned transaction duration prediction method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application can be essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiments of the present application are not limited to any specific combination of hardware and software.

[0187] An embodiment of the present application provides an electronic device, Fig.11 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application, such as Fig.11 As shown, the electronic device 110 includes a memory 1101 and a processor 1102. The memory 1101 stores a computer program that can be run on the processor 1102. When the processor 1102 executes the program, the steps in the transaction duration prediction method provided in the above embodiment are implemented.

[0188] It should be noted that the memory 1101 is configured to store instructions and applications executable by the processor 1102, and can also cache data to be processed or already processed by the processor 1102 and various modules in the electronic device 110 (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory 1101 (Random Access Memory, RAM).

[0189] An embodiment of the present application provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps in the transaction duration prediction method provided in the above method embodiment.

[0190] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium, storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0191] It should be understood that "one embodiment" or "an embodiment" or "some embodiments" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in one embodiment" or "in some embodiments" appearing throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments. The above description of each embodiment tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. For the sake of brevity, this article will not repeat them.

[0192] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0193] In this application, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c alone can represent: a, b, c, a and b, a and c, or a, b, and c, where a, b, and c can be single or multiple. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0194] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms.

[0195] The modules described above as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules; they may be located in one place or distributed on multiple system units; some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0196] In addition, all functional modules in the embodiments of the present application may be integrated into one processing unit, or each module may be a separate unit, or two or more modules may be integrated into one unit; the above-mentioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0197] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0198] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0199] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0200] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0201] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A transaction duration prediction method, characterized in that: The method comprises: Receive first information; wherein the first information includes supply information of a first supply source to be traded and a user's expected bid for the first supply source; The first information is input into a feature extraction network in a target transaction duration prediction model, and the feature extraction network is used to extract features from the first information to obtain a first feature vector corresponding to the first information and a second feature vector corresponding to the first information; wherein the first feature vector is used to predict a first transaction result within a first duration after the first source of goods is online at the expected bid; and the second feature vector is used to predict a second transaction result within a second duration after the first source of goods is online at the expected bid; and the second duration is greater than the first duration; Inputting the first feature vector into the first network of the target transaction duration prediction model, and using the first network to map the first feature vector into a specific numerical range to obtain a first confidence level; wherein the first confidence level is used to indicate the first transaction result; Inputting the first feature vector and the second feature vector into the second network of the target transaction duration prediction model, using the second network to aggregate the first feature vector and the second feature vector to obtain an aggregate vector, and mapping the aggregate vector to a specific numerical range to obtain a second confidence level; the second confidence level is used to indicate a second transaction result of the first source of goods within a second duration after the first source of goods goes online; Based on the first transaction result and the second transaction result, determine the transaction time of the first source of goods at the expected bid.

2. The method according to claim 1, characterized in that The determining, based on the first transaction result and the second transaction result, the transaction duration of the first source of goods at the expected bid price includes: If the first transaction result indicates that the first source of goods is traded within the first time period at the expected bid, and the second transaction result indicates that the first source of goods is traded within the second time period at the expected bid, it is determined that the transaction time period of the first source of goods is less than the first time period; If the first transaction result indicates that the first source of goods was not traded within the first time period at the expected bid, and the second transaction result indicates that the first source of goods was traded within the second time period at the expected bid, then the transaction time of the first source of goods is greater than the first time period and less than the second time period; If the first transaction result indicates that the first source of goods has not been traded within the first time period under the expected bid, and the second transaction result indicates that the first source of goods has not been traded within the second time period under the expected bid, then the transaction time of the first source of goods is greater than the second time period.

3. The method according to claim 1, characterized in that The second network includes an aggregation module and a first normalization layer connected in sequence; wherein, The aggregation module is used to receive the first feature vector and the second feature vector, perform splicing processing on the first feature vector and the second feature vector to obtain a first splicing vector, perform forward propagation and nonlinear processing on the first splicing vector to obtain the aggregated vector, and input the aggregated vector to the first normalization layer; The first normalization layer is used to map the aggregated vector to the specific numerical range to obtain the second confidence.

4. The method according to claim 1, characterized in that: The feature extraction network includes a feature fusion subnetwork, a first neural network layer, and a second neural network layer; the feature fusion subnetwork is connected to the first neural network layer and the second neural network layer respectively; the number of the supply source information is multiple; The feature fusion subnetwork is used to receive the expected bid and the multiple supply source information, perform feature extraction and aggregation operations on the expected bid and the multiple supply source information, and obtain a third feature vector corresponding to the multiple supply source information and a fourth feature vector corresponding to the multiple supply source information; The first neural network layer is used to perform forward propagation and nonlinear processing on the third eigenvector to obtain the first eigenvector; The second neural network layer is used to perform forward propagation and nonlinear processing on the fourth eigenvector to obtain the second eigenvector.

5. The method according to claim 4, characterized in that The feature fusion subnetwork includes multiple embedding layers, multiple third neural network layers and a fourth neural network layer; wherein the fourth neural network layer includes multiple input neurons, a first output neuron and a second output neuron; The multiple embedding layers are respectively connected to the multiple third neural network layers, and the multiple third neural network layers are respectively connected to the multiple input neurons of the fourth neural network layer; The multiple embedding layers are used to receive the expected bid and the multiple supply information, convert the expected bid and the multiple supply information into corresponding embedding vectors respectively, and input the embedding vectors into the corresponding third neural network layer; wherein one embedding layer corresponds to the expected bid or one supply information; The multiple third neural network layers are respectively used to perform forward propagation on the received embedding vectors to obtain multiple fifth eigenvectors, and input the multiple fifth eigenvectors into the fourth neural network layer; The fourth neural network layer receives the multiple fifth eigenvectors through the multiple input neurons respectively, performs forward propagation and feature aggregation on the multiple fifth eigenvectors, and then outputs the third eigenvector through the first output neuron and outputs the fourth eigenvector through the second output neuron.

6. The method according to claim 1, characterized in that The number of the second feature vectors is N, the number of the second networks is N, and the N second feature vectors are respectively used to predict the transaction results of the first source of goods within N second time periods under the expected bid; the N second feature vectors, the N second networks, and the N second time periods correspond one to one; wherein N is greater than 1, and the N second time periods increase in sequence; the method further includes: For the first second feature vector, the first feature vector and the first second feature vector are input into the first second network to obtain a first aggregation vector and a first second confidence; the first confidence is used to indicate the second transaction result of the first source of goods within the first second time period under the expected bid; For the i-th second feature vector, the first feature vector, the 1st aggregation vector to the i-1th aggregation vector are input into the i-th second network to obtain the i-th aggregation vector and the i-th second confidence; the i-th second confidence is used to indicate the second transaction result of the first source of goods within the i-th second time period under the expected bid; wherein i is greater than 1 and less than or equal to N; The transaction duration of the first source of goods is determined based on the first transaction result, the second transaction result of the first source of goods within the 1st second time period to the second transaction result of the first source of goods within the Nth second time period.

7. The method according to claim 1, characterized in that Before receiving the first information, the method further includes: Acquire sample data, the sample data including a first bid of a second source of goods that has been traded historically and source information of the second source of goods; the sample data carries a first tag and a second tag, wherein the first tag is used to indicate whether the second source of goods has been traded within a first time period under the first bid, and the second tag is used to indicate whether the second source of goods has been traded within a second time period under the first bid; Inputting the sample data into a feature extraction network in an initial transaction duration prediction model to obtain a sixth feature vector corresponding to the sample data and a seventh feature vector corresponding to the sample data; Inputting the sixth feature vector into the first network in the initial transaction duration prediction model to obtain a third confidence level; Inputting the seventh feature vector and the sixth feature vector into the second network in the initial transaction duration prediction model to obtain a fourth confidence level; Based on a first loss function between the third confidence level and the first label and a second loss function between the fourth confidence level and the second label, the initial transaction time prediction model is iteratively trained until an iteration stop condition is met to obtain the target transaction time prediction model.

8. A transaction duration prediction device, characterized in that: include: A module for executing the method according to any one of claims 1 to 7.

9. An electronic device, characterized in that: include: Memory and processor; The memory is configured to store computer program instructions; The processor is configured to execute the computer program instructions so that the electronic device implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: include: Computer program instructions; The electronic device runs the computer program instructions so that the electronic device implements the method according to any one of claims 1 to 7.

Images