Goods source sorting factor determination method, model training method and electronic equipment

Abstract

The invention provides a goods source sorting factor determination method, a model training method and electronic equipment, and relates to the technical field of goods transportation. The training method of the model comprises the following steps: determining sample bottom line price features based on obtained sample goods source historical transaction orders; obtaining sample goods source price competitiveness characteristics and sample goods source real-time pricing characteristics, and training the initial goods source sorting factor determination model based on the sample bottom line price characteristics, the sample goods source price competitiveness characteristics and the sample goods source real-time pricing characteristics to obtain a trained goods source sorting factor determination model; wherein the goods source sorting factor determination model is used for determining a goods source sorting factor of the goods source, and the goods source sorting factor comprises a click rate and / or a transaction rate of the goods source. According to the scheme, the goods source sorting matching effect of the model can be improved, and the monotonicity of the model on the goods source price is guaranteed.

Description

Technical Field

[0001] This application relates to the field of cargo transportation technology, and in particular to a method for determining cargo source ranking factors, a model training method, and an electronic device. Background Technology

[0002] In logistics and transportation service platforms, cargo ranking is a crucial element. For example, the platform can recommend cargo lists to drivers based on cargo ranking. Drivers can then access these lists through search engines or by clicking on web pages, making it easier for them to accept orders. The cargo ranking results are typically determined by ranking factors, such as click-through rates. Therefore, identifying these ranking factors is critical to cargo ranking.

[0003] Deep learning-based neural network models play a crucial role in freight ranking. They can be used to determine freight ranking factors, and then use these factors to determine the freight ranking. Theoretically, the higher the price of a freight, the greater the probability of drivers clicking and completing a transaction, and consequently, the higher the matching degree of that type of freight. However, in related technologies, because the model is primarily trained using freight costs as the price feature, there are few price-related features. This leads to situations where, when using the trained model for freight ranking, increasing the freight price actually lowers the ranking score. When maintaining price monotonicity is required, the model's matching performance significantly declines. Summary of the Invention

[0004] This application provides a method for determining the source ranking factor, a model training method, and an electronic device to improve the source ranking matching effect of the model.

[0005] Firstly, this application provides a training method for a source ranking factor determination model, comprising: Obtain historical transaction orders for sample goods, and determine sample floor price features based on the historical transaction orders for sample goods. The sample floor price features are used to characterize the sample floor price of multiple target dimensions and the relationship between the sample goods price and the sample floor price. The price competitiveness characteristics and real-time price increase characteristics of sample goods are obtained, and the initial goods ranking factor determination model is trained based on the sample bottom price characteristics, the sample goods price competitiveness characteristics, and the sample goods real-time price increase characteristics to obtain the trained goods ranking factor determination model; the goods ranking factor determination model is used to determine the goods ranking factor of the goods, and the goods ranking factor includes the click-through rate and / or transaction rate of the goods.

[0006] Optionally, the plurality of target dimensions includes at least two of the following dimensions: driver dimension, driver-route intersection dimension, and route-registered vehicle length intersection dimension; the step of determining the sample floor price feature based on the historical transaction orders of the sample cargo source includes: The minimum standardized price of the sample cargo in the driver dimension of the historical transaction orders of the sample cargo is determined to obtain the first sample bottom line price; The minimum standardized price of the sample cargo in the driver-route intersection dimension of the historical transaction orders of the sample cargo is determined to obtain the second sample bottom line price; The minimum standardized price of the sample cargo in the historical transaction orders of the sample cargo, based on the intersection dimension of the route and the registered vehicle length, is determined to obtain the third sample bottom line price; At least two of the first sample baseline price, the second sample baseline price, and the third sample baseline price are determined as the sample baseline prices for the plurality of target dimensions; Determine the target difference between the sample floor price and the sample source price; the target difference includes the absolute difference and / or the relative difference. The difference between the sample floor price and the sample target price is defined as the sample floor price feature.

[0007] Optionally, obtaining the price competitiveness characteristics and real-time price markup characteristics of the sample goods includes: Obtain the price percentile of the first sample source within a preset time period in each of the divided sample sub-markets to obtain the price competitiveness characteristics of the sample source. Obtain the sample price increase characteristics of the second sample source within a preset time period of the second sample, and obtain the real-time price increase characteristics of the sample source; wherein, the sample price increase characteristics include at least one of the sample price increase amount, the first sample time interval between the price increase time and the release time of the second sample source, and the second sample time interval between the last price increase and the current time.

[0008] Optionally, the initial source ranking factor determination model includes an initial price feature extraction layer, an initial feature encoding layer, an initial scenario extraction layer, an initial task extraction layer, and an initial task pyramid layer; training the initial source ranking factor determination model based on the sample floor price features, the sample source price competitiveness features, and the sample source real-time price increase features includes: The sample floor price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature are used as sample price sharing features, and sample scenario-related features are obtained. The sample price sharing feature and the sample scenario related feature are input into the initial feature encoding layer. The initial feature encoding layer encodes the sample price sharing feature and the sample scenario related feature to obtain the sample price encoding feature and sample scenario encoding feature output by the initial feature encoding layer. The sample price encoding features are input into the initial price feature extraction layer, and the initial price feature extraction layer extracts price features from the sample price encoding features to obtain the sample price features output by the initial price feature extraction layer. The sample price encoding feature and the sample scene encoding feature are input into the initial scene extraction layer. The initial scene extraction layer extracts scene features from the sample price encoding feature and the sample scene encoding feature to obtain the sample scene features output by the initial scene extraction layer. The sample scene features are input into the initial task extraction layer, and the initial task extraction layer extracts task features from the sample scene features to obtain the sample task features output by the initial task extraction layer. The sample task features and sample price features are input into the initial task tower layer, and task identification is performed through the initial task tower layer to obtain the sample source ranking factor output by the initial task tower layer; the sample source ranking factor includes sample click rate and / or sample transaction rate. The model parameters of the initial source ranking factor determination model are adjusted based on the sample source ranking factor and sample label data until the initial source ranking factor determination model converges, thus obtaining the trained source ranking factor determination model.

[0009] Optionally, the step of inputting the sample price sharing feature and the sample scene-related feature into the initial feature encoding layer, and performing feature encoding on the sample price sharing feature and the sample scene-related feature through the initial feature encoding layer to obtain the sample price encoding feature and sample scene encoding feature output by the initial feature encoding layer includes: Each type of sample price sharing feature is used as a sample bucketing feature. The non-default value price features in the sample bucketing feature are bucketed by equal frequency to obtain the first sample bucketing. The default value price feature in the sample to be binned is taken as a new bin and added to the boundary of the first sample bin, together with the first sample bin to form the second sample bin of the sample to be binned feature; Based on a preset gain coefficient, the bucket boundary value of the second sample bucket is increased to obtain the target sample bucket feature corresponding to the sample bucket feature to be bucketed. The target sample binning features and the sample scene-related features corresponding to the shared features of all types of sample prices are input into the initial feature encoding layer. The initial feature encoding layer encodes all the target sample binning features and the sample scene-related features to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer.

[0010] Optionally, after obtaining the trained source ranking factor determination model, the training method of the source ranking factor determination model further includes: Collect real driver-cargo pairs samples, and increase the prices in the real driver-cargo pairs samples based on a preset price increase ratio to obtain virtual driver-cargo pairs samples; The real driver-cargo pair sample and the virtual driver-cargo pair sample are determined as test samples; For each test sample group in the test sample, the virtual driver-cargo pair samples in the test sample group are input into the cargo source ranking factor determination model to obtain the virtual cargo source ranking factor output by the cargo source ranking factor determination model; the virtual cargo source ranking factor includes virtual click-through rate and / or virtual transaction rate; The virtual ranking score of the virtual driver pair samples in the test sample group is determined based on the virtual cargo source ranking factor. Determine the Spearman correlation coefficient between the virtual ranking score of the test sample group and the source price of the real cargo pair sample in the test sample group; the Spearman correlation coefficient is used to characterize the price monotonicity of the source ranking factor determination model.

[0011] Secondly, this application provides a method for determining the source ranking factor, including: Retrieve historical transaction orders for goods within the first preset time period; The baseline price feature is determined based on the historical transaction orders of the goods. The baseline price feature is used to characterize the baseline price of multiple target dimensions and the relationship between the price of the goods and the baseline price. Obtain the price competitiveness characteristics of goods within the second preset time period and the real-time price increase characteristics of goods within the third preset time period; The bottom line price feature, the source price competitiveness feature, and the source real-time price increase feature are used as price sharing features and input into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model. The product ranking factor determination model is trained based on the training method of the product ranking factor determination model as described in any of the first aspects above; the product ranking factor includes the product's click-through rate and / or transaction rate.

[0012] Optionally, the source ranking factor determination model includes a price feature extraction layer, a feature encoding layer, a scenario extraction layer, a task extraction layer, and a task pyramid layer; the step of inputting the baseline price feature, the source price competitiveness feature, and the source real-time price increase feature as price sharing features into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model includes: Obtain scene-related features; The baseline price feature, the source price competitiveness feature, and the source real-time price increase feature are used as price sharing features and input together with the scenario-related features into the feature encoding layer. The feature encoding layer encodes the price sharing feature and the scenario-related feature to obtain the price encoding feature and scenario encoding feature output by the feature encoding layer. The price coding feature is input into the price feature extraction layer, and the price feature extraction layer extracts price features from the price coding feature to obtain the price features output by the price feature extraction layer. The price encoding feature and the scene encoding feature are input into the scene extraction layer. The scene extraction layer extracts scene features from the price encoding feature and the scene encoding feature to obtain the scene features output by the scene extraction layer. The scene features are input into the task extraction layer, and the task extraction layer extracts task features from the scene features to obtain the task features output by the task extraction layer. The task features and price features are input into the task tower layer, and task identification is performed through the task tower layer. The source ranking factor output by the task tower layer is determined as the source ranking factor output by the source ranking factor determination model.

[0013] Thirdly, this application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the training method for the source ranking factor determination model as described in any of the first aspects above, or to implement the steps of the source ranking factor determination method as described in any of the second aspects above.

[0014] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the training method for the source ranking factor determination model as described in any of the first aspects above, or implements the steps of the source ranking factor determination method as described in any of the second aspects above.

[0015] The method for determining the source ranking factor, the training method for the model, and the electronic equipment provided in this application, in the training process of the source ranking factor determination model, firstly determine the sample floor price characteristics based on the historical transaction orders of the sample source, then obtain the sample source price competitiveness characteristics and the sample source real-time price increase characteristics, and train the initial source ranking factor determination model based on the sample floor price characteristics, sample source price competitiveness characteristics, and sample real-time price increase characteristics to obtain the trained source ranking factor determination model. Among them, the sample bottom-line price feature is used to characterize the sample bottom-line price in multiple target dimensions and the relationship between the sample source price and the sample bottom-line price. It comes from the historical transaction orders of the sample source and can reflect the characteristics of the source price and its relationship with user behavior from multiple different dimensions. The sample source price competitiveness feature can reflect the value of the sample source, and the sample source real-time price increase feature can characterize the personalized price increase behavior of the sample source. In this way, when training the model based on the sample bottom-line price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature, the characteristics of price factors in multiple different dimensions and the correlation between price features and user behavior are comprehensively considered, which greatly enriches the price-related features, thereby improving the model's ability to capture prices, and thus improving the model's source ranking and matching effect, while ensuring the monotonicity of the model in terms of source prices as much as possible. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the training method for the source ranking factor determination model provided in this application embodiment; Figure 2 A schematic diagram illustrating the structure and working principle of the initial source ranking factor determination model provided in this application embodiment; Figure 3 A flowchart illustrating the method for determining the source ranking factor provided in this application embodiment; Figure 4 This is a schematic diagram illustrating the structure and working principle of the source ranking factor determination model provided in the embodiments of this application; Figure 5 A schematic diagram of the structure of the training device for the source ranking factor determination model provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of the source ranking factor determination device provided in the embodiments of this application. Detailed Implementation

[0017] In this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c alone can mean: a alone, b alone, c alone, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c. a, b, and c can be single or multiple. Furthermore, the terms "first" and "second" are used only to distinguish different descriptive objects and should not be construed as indicating or implying relative importance.

[0018] With the rapid development of internet technology, the logistics industry's transportation services are gradually transitioning from traditional offline operations to unified online logistics transportation service platforms. More and more drivers and shippers are relying on these comprehensive platforms to connect cargo with transportation needs. By integrating massive amounts of cargo information, these platforms provide shippers with a wide range of transportation options and offer drivers abundant freight opportunities, playing a vital role in the logistics transportation sector.

[0019] In logistics and transportation service platforms, cargo ranking is a crucial element. For example, the platform can recommend cargo lists to drivers based on cargo ranking. Drivers can then access these lists through search engines or by clicking on web pages, making it easier for them to accept orders. The cargo ranking results are typically determined by ranking factors, such as click-through rates. Therefore, identifying these ranking factors is critical to cargo ranking.

[0020] Deep learning-based neural network models play a crucial role in freight ranking. They can be used to determine freight ranking factors, and then use these factors to determine the freight ranking. Theoretically, the higher the price of a freight, the greater the probability of drivers clicking and completing a transaction, and correspondingly, the higher the matching degree of that freight. However, in related technologies, because the model is primarily trained using freight costs as the price feature, there are few price-related features. This leads to situations where, when using the trained model for freight ranking, increasing the freight price actually decreases the matching degree. When maintaining price monotonicity is required, the model's matching performance declines significantly.

[0021] On the other hand, in multi-task scenarios, different tasks may depend on price factors to varying degrees. For example, a product's click-through rate may be more affected by price, while its collection rate may be more related to the product's quality. Using a single product's shipping cost as a price feature to train a model makes it difficult to coordinate price factors across multiple tasks.

[0022] Based on this, embodiments of this application provide an improved training method for a source ranking factor determination model and a method for determining source ranking factors using the source ranking factor determination model trained by this method. The method extracts sample floor price features from historical transaction orders of sample sources. These sample floor price features, sample source price competitiveness features, and sample source real-time price increase features are used as price-related sample features for model training. The initial source ranking factor determination model is then trained to obtain a trained source ranking factor determination model. This model can learn price-related features from multiple dimensions, enhancing its ability to capture price information and thus improving the model's monotonicity and matching effect on source prices. In multi-task scenarios, it can also improve the model's ability to coordinate price factors across multiple tasks.

[0023] The training method for the source ranking factor determination model provided in this application can be applied to electronic devices or to a training device for the source ranking factor determination model set in such electronic devices. This device can be implemented through software, hardware, or a combination of both. The electronic devices may include at least one of the following: servers, mobile phones, computers, vehicle terminals, tablet computers, wearable devices, and smart home devices. Servers may include dedicated servers, virtual servers, and cluster servers.

[0024] The following example uses applications in electronic devices, combined with... Figures 1-2 The training method for the source ranking factor determination model provided in the embodiments of this application is described in detail.

[0025] Figure 1 This document illustrates a flowchart of the training method for the source ranking factor determination model provided in an embodiment of this application. (Refer to...) Figure 1 As shown, the training method for the source ranking factor determination model may include the following steps 110 to 120.

[0026] Step 110: Obtain historical transaction orders for the sample goods and determine the sample floor price characteristics based on the historical transaction orders for the sample goods.

[0027] The sample floor price feature is used to characterize the sample floor price across multiple target dimensions, as well as the relationship between the sample source price and the sample floor price. These multiple target dimensions may include at least two of the following: driver dimension, driver-route intersection dimension, and route-registered vehicle length intersection dimension, but are not limited to these.

[0028] Specifically, electronic devices can obtain historical transaction orders of goods within a preset historical time period from the logistics and transportation service platform, and obtain historical transaction orders of sample goods. These historical transaction orders of sample goods can reflect the behavioral information of sample drivers and the transaction information between sample drivers and sample goods within the preset historical time period.

[0029] After obtaining the historical transaction orders of the sample cargo, the sample price information in the historical transaction orders can be identified. Then, based on the sample price attribute information corresponding to this sample price information, statistical analysis is performed on these sample price information across multiple target dimensions to obtain the sample floor price characteristics. The sample price attribute information may include at least one of the following, but is not limited to: sample driver information, sample route information, sample vehicle length information, and sample cargo information.

[0030] The sample cargo has a large number of historical transaction orders, and the price units for the sample cargo may also be varied. For example, some sample cargo may be priced based on the number of transport trips, while others may be priced based on weight, and the same price unit may also have different pricing methods. Furthermore, the price distribution of LTL and FTL cargo is naturally not in the same space, with LTL prices generally being lower. Therefore, in one embodiment, before performing statistical analysis on the sample price information across multiple target dimensions, the sample price information can be standardized to obtain the corresponding standardized prices for the sample cargo. Then, statistical analysis across multiple target dimensions can be performed based on these standardized prices.

[0031] For example, the price of sample cargo can be standardized by using a normalization method that calculates the price per kilometer for full truckloads and the price per ton-kilometer for less-than-truckload shipments.

[0032] For example, determining the sample floor price characteristics based on historical transaction orders of sample goods may include the following steps 111 to 116.

[0033] Step 111: Determine the minimum standardized price of the sample cargo from the driver dimension in the historical transaction orders of the sample cargo, and obtain the first sample bottom line price.

[0034] For each sample driver in the historical transaction orders of the sample cargo, the electronic device can obtain the sample orders for that driver from the historical transaction orders. Then, it can obtain the sample cargo price in each sample order and standardize these sample cargo prices to obtain the standardized sample cargo price per kilometer for each sample order. Finally, it can determine the minimum value among these standardized sample cargo prices to obtain the first sample floor price for that driver. In this way, the first sample floor price corresponding to each sample driver in the historical transaction orders of the sample cargo can be obtained, that is, the first sample floor price at the driver level in the historical transaction orders of the sample cargo.

[0035] Step 112: Determine the minimum standardized price of the sample cargo in the driver-route intersection dimension of the historical transaction orders of the sample cargo, and obtain the second sample bottom line price.

[0036] For each sample driver in the historical transaction orders of the sample cargo, the electronic device can obtain the sample orders for that driver from the historical transaction orders. Then, it retrieves the first sample route information and the corresponding sample cargo price from each sample order, and standardizes the sample cargo price to obtain the corresponding standardized sample cargo price. Next, for each first sample route in the first sample route information, it determines the minimum value among the standardized sample cargo prices corresponding to that first sample route, obtaining the second sample floor price corresponding to that first sample route. In this way, the second sample floor price for each first sample route of each sample driver can be obtained, that is, the second sample floor price of the driver-route intersection dimension in the historical transaction orders of the sample cargo.

[0037] Step 113: Determine the minimum standardized price of the sample cargo in the historical transaction orders of the sample cargo, which is the intersection dimension of route and registered vehicle length, to obtain the third sample bottom line price.

[0038] The electronic device can obtain all route information from the historical transaction orders of the sample cargo to obtain the second sample route information. For each second sample route in the second sample route information, the electronic device can obtain all registered vehicle length information and the corresponding sample cargo price for that second sample route from the historical transaction orders of the sample cargo, and standardize the sample cargo price to obtain the sample registered vehicle length for the second sample route and the standardized price of the sample cargo for each sample registered vehicle length. Then, for each sample registered vehicle length for the second sample route, the minimum value of the standardized price of the sample cargo for that sample registered vehicle length is determined to obtain the third sample floor price for that sample registered vehicle length for the second sample route. In this way, the third sample floor price for each sample registered vehicle length for each sample route can be obtained, that is, the third sample floor price of the intersection dimension of route and registered vehicle length in the historical transaction orders of the sample cargo can be obtained.

[0039] Step 114: Determine at least two of the first sample bottom line price, the second sample bottom line price, and the third sample bottom line price as sample bottom line prices for multiple target dimensions.

[0040] After obtaining the first sample bottom line price, the second sample bottom line price, and the third sample bottom line price, at least two of these sample bottom line prices can be determined as sample bottom line prices for multiple target dimensions.

[0041] Step 115: Determine the target difference between the sample floor price and the sample source price.

[0042] The target difference value of the samples includes at least one of the absolute difference of the samples and the relative difference of the samples.

[0043] After obtaining the sample floor price, for each sample floor price, feature cross-validation is performed between the sample floor price and the corresponding sample source price to determine the sample absolute difference and / or sample relative difference between the sample floor price and the corresponding sample source price.

[0044] For example, the baseline price of the sample includes the first baseline price for each driver, i.e., the first baseline price for each sample driver. For each sample driver, the absolute difference between the first baseline price of that driver and the price of the sample cargo in each sample order corresponding to that driver can be determined, thus obtaining the absolute difference of each sample order corresponding to that driver. Furthermore, the ratio of this absolute difference to the corresponding sample cargo price can be determined, thus obtaining the relative difference of the sample driver.

[0045] In this way, by explicitly performing feature cross-referencing on the sample floor price and the sample source price, the absolute difference and / or relative difference between the sample floor price and the corresponding sample source price can be determined, thus explicitly characterizing the relationship between the sample floor price and the sample source price.

[0046] Step 116: Determine the difference between the sample floor price and the sample target price as the sample floor price feature.

[0047] Electronic devices can identify all determined sample floor prices and sample target differences as sample floor price features.

[0048] Step 120: Obtain the price competitiveness characteristics and real-time price increase characteristics of the sample goods, and train the initial goods ranking factor determination model based on the sample floor price characteristics, sample goods price competitiveness characteristics, and sample goods real-time price increase characteristics to obtain the trained goods ranking factor determination model.

[0049] The product ranking factor determination model is used to determine the product ranking factors, which can be used to rank products. These ranking factors can include the product's click-through rate and / or conversion rate.

[0050] For example, taking the product ranking factors, which include the product's click-through rate and conversion rate, as an example, after determining the product's click-through rate and conversion rate, the product's click-through rate and conversion rate can be summed by a power weight to obtain the product's ranking score. Then, the products can be sorted in descending order of their ranking scores.

[0051] The price competitiveness characteristic of sample goods can characterize the competitiveness or value of sample goods. The higher the competitiveness of the sample goods, the greater the probability and speed of transaction. The real-time price increment characteristic of sample goods is used to characterize the personalized price increment behavior of sample goods.

[0052] Specifically, in one embodiment, obtaining the price competitiveness characteristics of the sample goods and the real-time price increase characteristics of the sample goods may include the following steps 1211 to 1212.

[0053] Step 1211: Obtain the price quantile of the first sample source within the preset time period in each of the divided sample sub-markets to obtain the price competitiveness characteristics of the sample source.

[0054] The preset time period for the first sample can be set according to actual needs, such as 48 hours. This application embodiment does not impose any special limitation on this. The sample sub-market can be divided based on sample attribute information, and sample sources with the same sample attributes can be grouped into the same sample sub-market. The sample attribute information may include at least one of non-price information such as sample source route and sample vehicle length, but is not limited to this.

[0055] Specifically, when a sample shipment is published or its price is updated on a logistics and transportation service platform, it can be assigned to a corresponding sample sub-market based on its sample attribute information. For example, sample shipments in a sub-market can be ranked according to their prices, and the price quantile of the sample shipment's ranking within its corresponding sub-market can represent its competitiveness.

[0056] Electronic devices can obtain the price percentile of the first sample goods within a preset time period in their respective sample sub-markets, thereby obtaining the price competitiveness characteristics of the sample goods.

[0057] Step 1212: Obtain the sample price increase characteristics of the second sample source within the preset time period of the second sample, and obtain the real-time price increase characteristics of the sample source.

[0058] Among them, the sample price increase characteristics include at least one of the following: sample price increase amount, the time interval between the price increase and the release of the second sample source, and the time interval between the last price increase and the current time, but not limited to these.

[0059] The preset time period for the second sample can be set according to actual needs, such as 2 days or 5 days, and this application embodiment does not make any special limitation on this.

[0060] Specifically, for sample goods posted on logistics and transportation service platforms, if no transactions are completed within a certain period, the cargo owner may increase the price. Electronic devices can statistically analyze the price increase characteristics of the second sample goods within a preset time period, obtaining the real-time price increase characteristics of the sample goods. The second sample goods refer to those that have undergone price increases.

[0061] After obtaining the sample floor price characteristics, sample cargo price competitiveness characteristics, and sample cargo real-time price increment characteristics, electronic devices can perform supervised training on the initial cargo ranking factor determination model based on these characteristics, resulting in a trained cargo ranking factor determination model. Subsequently, this model can be used to determine the cargo ranking factors for real-time cargo published on the logistics and transportation service platform, and then the cargo can be ranked based on the determined ranking factors.

[0062] For example, the initial supply ranking factor determination model can be a model with a deep neural network architecture. Electronic devices can input the sample bottom price characteristics, sample supply price competitiveness characteristics, and sample supply real-time price increase characteristics into the initial supply ranking factor determination model to conduct supervised training on the initial supply ranking factor determination model. The sample label data during training includes label supply ranking factors, which include label click-through rate and / or label transaction rate.

[0063] For example, the initial source ranking factor determination model can be a neural network model with a multi-scenario, multi-task architecture. For instance, it can adopt an existing neural network model with a multi-scenario, multi-task architecture, such as the Hierarchical Information Extraction Network (HiNet) model. Alternatively, it can be a new model formed by adding an initial price feature extraction layer to an existing neural network model with a multi-scenario, multi-task architecture. The initial price feature extraction layer extracts price features from the sample price features encoded by the neural network model with a multi-scenario, multi-task architecture, and inputs the extracted price features into the final initial task layer of the neural network model with a multi-scenario, multi-task architecture to perform task recognition together with other features, thereby improving the shallow expression of price features.

[0064] In multi-scenario, multi-task application scenarios, electronic devices can also acquire sample scenario-related features. These features, along with sample floor price features, sample supply price competitiveness features, and sample supply real-time price markup features, are input into an initial supply ranking factor determination model. Supervised training is then performed on this initial model to obtain a trained supply ranking factor determination model. The sample scenario-related features can include shared features, specific features, identifying features, and attribute features. Shared features represent characteristics common to all scenarios; specific features represent characteristics unique to a particular scenario; identifying features distinguish different scenarios; and attribute features represent scenario information, such as the scenario name.

[0065] For example, the shared features of the sample scenario may include at least one of the following, but are not limited to: the sample distance between the sample driver and the loading point of the sample cargo, the number of sample cargoes clicked by the sample driver in the first preset sample time period, and the ratio of the weight of the sample cargoes to the load capacity of the sample driver's vehicle; the specific features of the sample scenario may include at least one of the following, but are not limited to: the number of times the sample driver clicked on cargoes on the sample routes in each scenario during the past second preset sample time period, the sample duration of the sample driver's subscription to the sample routes, and the number of times the sample driver searched for samples during the third preset sample time period; the sample scenario identification features may include the corresponding number of the scenario; the sample scenario attribute features may include the name of the scenario.

[0066] Figure 2 This illustration shows the structure and working principle of the initial source ranking factor determination model provided in an embodiment of this application. In one embodiment, referring to... Figure 2 As shown, the initial source ranking factor determination model may include an initial price feature extraction layer 201, an initial feature encoding layer 202, an initial scene extraction layer 203, an initial task extraction layer 204, and an initial task tower layer 205.

[0067] Accordingly, step 120 trains the initial source ranking factor determination model based on the sample bottom price characteristics, sample source price competitiveness characteristics, and sample source real-time price increase characteristics, which may include the following steps 1221 to 1227.

[0068] Step 1221: Use the sample floor price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature as sample price sharing features, and obtain sample scenario-related features.

[0069] Among them, the sample scene-related features can include sample scene shared features, sample scene-specific features, sample scene identification features, and sample scene attribute features.

[0070] Step 1222: Input the sample price sharing features and sample scene-related features into the initial feature encoding layer 202, and encode the sample price sharing features and sample scene-related features through the initial feature encoding layer 202 to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer 202.

[0071] The initial feature encoding layer 202 is used to encode the sample price sharing feature and the sample scene encoding feature.

[0072] In one embodiment, the sample floor price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature can be directly input into the initial feature encoding layer 202 as sample price sharing features. The initial feature encoding layer 202 performs feature encoding on the sample price sharing features to obtain the sample price encoding features corresponding to the sample price sharing features output by the initial feature encoding layer 202.

[0073] Considering that the shared price features of the samples may be continuous, these features can be binned before being input into the initial feature encoding layer 202 for feature encoding. This means discretizing the shared price features before feature encoding. However, the binning method for continuous features affects the model's sensitivity to price, potentially leading to a lower probability of change in feature bins when prices rise compared to when prices fall. Furthermore, in actual testing, it was found that approximately 30% of the samples' price features were assigned to the default value bins, while the remaining bins had coarser boundary granularity, further contributing to the model's lower sensitivity to price.

[0074] Based on this, in another embodiment, the binning method can be improved by first removing the default value samples, then re-binding the remaining samples according to price-related features to refine the granularity of binning, and then appending the price features of the default value samples to the boundaries of the already divided bins. Furthermore, after obtaining the new feature bin boundaries, the boundary values ​​of the bins can be increased by a preset percentage, such as 1%, which can solve the problem of unchanged binning after a price increase.

[0075] Specifically, the sample price sharing features and sample scenario-related features are input into the initial feature encoding layer 202. The initial feature encoding layer 202 encodes the sample price sharing features and sample scenario-related features to obtain the sample price encoded features and sample scenario encoded features output by the initial feature encoding layer 202. This can include: taking each type of sample price sharing feature as a sample to be binned feature; performing equal-frequency binning on the non-default value price features in the sample to be binned features to obtain the first sample bin; and appending the default value price features in the sample to be binned features as a new bin to the first bin. The boundary of the first sample bucket, together with the first sample bucket, forms the second sample bucket for the sample features to be bucketed. Based on a preset gain coefficient, the bucket boundary value of the second sample bucket is increased to obtain the target sample bucket features corresponding to the sample features to be bucketed. The target sample bucket features corresponding to the sample scene-related features and the price-sharing features of all types of samples are input into the initial feature encoding layer 202. The initial feature encoding layer 202 encodes the target sample bucket features and the sample scene-related features to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer 202.

[0076] The default price feature refers to the price feature corresponding to sample goods with no price, such as a price of 0. Another example is the real-time price increase feature for sample goods; some sample goods may not have price increases, and the price-related feature of these price increases is the default price feature. Non-default price features are the remaining price features excluding the default price feature. The preset gain coefficient can be set according to actual needs or through experimentation; for example, a 1% gain indicates a 1% increase.

[0077] For example, taking the real-time price increase feature of sample goods as an example, some sample goods do not have price increase behavior. The price-related features (i.e., default value price features) corresponding to these sample goods can be removed from the real-time price increase feature of sample goods. The remaining features are then divided into equal-frequency bins according to price values. For example, if 5 bins are formed, these 5 bins are determined as the first sample bin. Then, the previously removed features are added as a new bin before the first bin of the first sample bin, forming a second sample bin with the first sample bin to share the sample price feature, resulting in 6 bins. Next, the bin boundary value of each of these 6 bins can be increased by 1%, resulting in 6 new bins. These 6 new bins are determined as the target sample bin features. Then, the target sample bin features and sample scenario-related features are input into the initial feature encoding layer 202. The initial feature encoding layer 202 performs feature encoding to obtain the sample price encoded features and sample scenario encoded features output by the initial feature encoding layer 202.

[0078] In this way, by first binning the non-default price features and then adding the default price features as new bins to the boundaries of the existing bins, the granularity of binning can be refined, thereby improving the model's sensitivity to price. Simultaneously, by applying gain to the bin boundary values ​​of each bin, the sensitivity of the bins to price changes can be increased, further enhancing the model's overall price sensitivity.

[0079] For example, considering the potentially large variance of the price competitiveness feature of the sample goods, the feature can be divided into first and second equal-interval bins, such as 5 bins and 10 bins respectively. The binning results can then be used as the target sample binning features for the price competitiveness feature. This improves the binning effect of the price competitiveness feature and enhances the sensitivity and performance of the model.

[0080] Step 1223: Input the sample price coding features into the initial price feature extraction layer 201, and extract price features from the sample price coding features through the initial price feature extraction layer 201 to obtain the sample price features output by the initial price feature extraction layer 201.

[0081] The initial price feature extraction layer 201 is used to extract price features. For example, it can be implemented using a multilayer perceptron (MLP) neural network structure.

[0082] Step 1224: Input the sample price coding features and sample scene coding features into the initial scene extraction layer 203. The initial scene extraction layer 203 extracts scene features from the sample price coding features and sample scene coding features to obtain the sample scene features output by the initial scene extraction layer 203.

[0083] The initial scene extraction layer 203 can be used to extract scene-specific information and valuable information shared between scenes from the sample price coding features and sample scene coding features, thus obtaining sample scene features. Extracting scene features from the sample price coding features and sample scene coding features helps improve the information representation capabilities of the task layer.

[0084] For example, such as Figure 2As shown, the initial scene extraction layer 203 may include an initial scene common expert network and an initial scene specific expert network. The sample scene shared coding features in the sample price coding features and sample scene coding features can be input into the initial scene common expert network and the initial scene specific expert network of the initial scene extraction layer 203, respectively. The sample scene specific features in the sample scene coding features can be input into the initial scene specific expert network. The common information between different scenes is extracted through the initial scene common expert network, and the scene-specific information is learned through the initial scene specific expert network. The initial scene extraction layer 203 outputs the extracted common information between different scenes and the scene-specific information as sample scene features.

[0085] Considering the potential correlation between different scenarios, information from other scenarios may also contribute to the information representation of the current scenario. Furthermore, considering the varying representational capabilities of different scenarios, in one embodiment, the initial scenario extraction layer 203 may further include an initial scenario attention network. The electronic device may also input sample price encoding features and all sample scenario encoding features into the initial scenario attention network for scenario perception. Then, the sample scenario attention features output by the initial scenario attention network, along with the features output by the initial scenario common expert network and the initial scenario specific expert network, are used as the sample scenario features output by the initial scenario extraction layer 203. This measures the importance of the contribution of other scenario information to the information representation of the current scenario, thereby enhancing the information representation capability of the current scenario.

[0086] Step 1225: Input the sample scene features into the initial task extraction layer 204, and extract the task features of the sample scene features through the initial task extraction layer 204 to obtain the sample task features output by the initial task extraction layer 204.

[0087] The sample scene features output by the initial scene extraction layer 203 can be input into the initial task extraction layer 204 after vector concatenation. The initial task extraction layer 204 then extracts task features from the sample scene features. Specifically, the initial task extraction layer 204 can be used to extract the shared information of all tasks and the unique information of each task from the sample scene features to obtain the sample task features.

[0088] For example, such as Figure 2As shown, the initial task extraction layer 204 may include an initial task shared expert network and an initial task specific expert network. Sample scene features can be input into the initial task shared expert network and the initial task specific expert network respectively. The initial task shared expert network learns the shared information in all tasks in the current scene, and the initial task specific expert network extracts the specific information of each task in the current scene. The initial task extraction layer 204 outputs the extracted shared information in all tasks and the specific information of each task as sample task features.

[0089] Understandably, the initial task-specific expert network can be designed based on the number of tasks, with each task corresponding to one initial task-specific expert network, for example, such as... Figure 2 As shown, it can include two initial task-specific expert networks: a first initial task-specific expert network and a second initial task-specific expert network, which are assigned to two tasks, such as click-through rate and conversion rate, respectively.

[0090] Step 1226: Input the sample task features and sample price features into the initial task tower layer 205, perform task identification through the initial task tower layer 205, and obtain the sample source ranking factor output by the initial task tower layer 205.

[0091] Among them, the sample product ranking factors include sample click-through rate and / or sample transaction rate.

[0092] The sample task features output by the initial task extraction layer 204 and the sample price features output by the initial price feature extraction layer 201 can be vectorized separately according to different tasks and then input into the initial task tower layer 205. The initial task tower layer 205 performs task recognition and obtains the sample source ranking factor output by the initial task tower layer 205.

[0093] In this way, by concatenating the additional extracted sample price features into the input vector of the initial task tower layer 205 and performing task recognition together with the sample task features output by the initial task extraction layer 204, the shallow expression of price features can be improved, thereby improving the sensitivity and monotonicity of the initial task tower layer 205 and the model to price factors.

[0094] Step 1227: Adjust the model parameters of the initial source ranking factor determination model based on the sample source ranking factor and sample label data until the initial source ranking factor determination model converges, and obtain the trained source ranking factor determination model.

[0095] Specifically, electronic devices can determine the loss value of the loss function of the initial source ranking factor determination model based on the sample label data corresponding to the sample source ranking factor and sample price sharing features. Then, the model parameters of the initial source ranking factor determination model are adjusted based on the loss value until the loss value converges. At this point, the initial source ranking factor determination model is considered to have converged, and the trained source ranking factor determination model can be obtained.

[0096] Among them, the sample label data corresponding to the sample price sharing feature includes the label source ranking factor, which includes the label click-through rate and / or label transaction rate.

[0097] After obtaining the sample source ranking factor, the sample source ranking factor and the label source ranking factor can be compared. Based on the comparison result, the initial source ranking factor is determined, and the loss value of the model's loss function is determined. The loss function can be any existing loss function, and this application does not impose any special restrictions on it.

[0098] Then, the trained cargo ranking factor determination model can be used to determine the cargo ranking factors of real-time cargo published on the logistics and transportation service platform, and then the cargo can be ranked based on the determined cargo ranking factors.

[0099] The training method for the source ranking factor determination model provided in this application first determines the sample bottom line price characteristics based on the historical transaction orders of the sample source, then obtains the sample source price competitiveness characteristics and the sample source real-time price increase characteristics, and trains the initial source ranking factor determination model based on the sample bottom line price characteristics, sample source price competitiveness characteristics, and sample source real-time price increase characteristics to obtain the trained source ranking factor determination model. Among them, the sample bottom-line price feature is used to characterize the sample bottom-line price in multiple target dimensions and the relationship between the sample source price and the sample bottom-line price. It comes from the historical transaction orders of the sample source and can reflect the characteristics of the source price and its relationship with user behavior from multiple different dimensions. The sample source price competitiveness feature can reflect the value of the sample source, and the sample source real-time price increase feature can characterize the personalized price increase behavior of the sample source. In this way, when training the model based on the sample bottom-line price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature, the characteristics of price factors in multiple different dimensions and the correlation between price features and user behavior are comprehensively considered, which greatly enriches the price-related features, thereby improving the model's ability to capture prices, and thus improving the model's source ranking and matching effect, while ensuring the monotonicity of the model in terms of source prices as much as possible.

[0100] Price is a crucial factor in cargo ranking and matching scenarios. Other characteristics also significantly influence the matching effectiveness, such as whether the cargo is new and the degree of match between the cargo and driver selection criteria. Using real samples makes it difficult to measure the model's optimization performance. Furthermore, commonly used ranking models' AUC (Area Under Curve) and GAUC (Gross Area Under Curve) metrics are insufficient for optimizing the model's performance on price monotonicity.

[0101] Based on this, in one embodiment of this application, the training method for the source ranking factor determination model based on the above embodiments can employ adversarial testing to address interference issues caused by other features and factors. Real samples can be sampled online, and only the price-related features of these samples are increased to the same degree—that is, a virtual price increase—while other features remain unchanged. Each real sample can generate a preset number of virtual samples in this way, such as 30 virtual samples. The price increase can be evenly distributed within a preset range, such as 1% to 30%, thus forming test samples. The virtual samples are used for offline inference on the trained source ranking factor determination model to obtain virtual ranking scores. In each pair of samples, the Spearman correlation coefficient is used to calculate the consistency between the source price and the virtual ranking score in ranking, thereby quantitatively analyzing whether the monotonic relationship of price factors has strengthened.

[0102] Specifically, after obtaining the trained source ranking factor determination model, the training method of the source ranking factor determination model may also include the following steps 130 to 170.

[0103] Step 130: Collect real driver-cargo pairs, and increase the prices in the real driver-cargo pairs based on the preset price increase ratio to obtain virtual driver-cargo pairs.

[0104] The preset price increase percentage can be set according to actual needs, for example, it can be set in the range of 1% to 30%.

[0105] For each real driver-cargo pair sample, the price of that real driver-cargo pair sample can be increased a preset number of times to obtain a corresponding preset number of virtual driver-cargo pairs samples. The preset price increase percentage for each increase is different; for example, it can be evenly distributed within a preset range, such as between 1% and 30%. The preset number can be designed as needed, such as 30 or 50.

[0106] Step 140: Determine the real driver-cargo pair sample and the virtual driver-cargo pair sample as the test sample.

[0107] Understandably, in the test samples, each real driver-cargo pair and its multiple derived virtual driver-cargo pairs form a test sample group.

[0108] Step 150: For each test sample group in the test sample, input the virtual driver-cargo pair samples in the test sample group into the cargo source ranking factor determination model to obtain the virtual cargo source ranking factor output by the cargo source ranking factor determination model.

[0109] Among them, the ranking factors for virtual goods include virtual click-through rate and / or virtual transaction rate.

[0110] Step 160: Determine the virtual ranking score of the virtual driver pairs in the test sample group based on the virtual cargo source ranking factor.

[0111] Taking virtual cargo ranking factors, including virtual click-through rate and virtual transaction rate, as an example, the virtual click-through rate and virtual transaction rate can be weighted and summed or subjected to power-weighted summation to obtain the virtual ranking score. In this way, for each test sample group, the virtual ranking score of the virtual cargo samples in that test sample group can be determined.

[0112] Step 170: Determine the Spearman correlation coefficient between the virtual ranking score of the test sample group and the source price of the real cargo in the test sample group.

[0113] The Spearman correlation coefficient is used to characterize the price monotonicity of the source ranking factor determination model.

[0114] For each test sample group, after obtaining the virtual ranking score of the virtual driver-cargo pair in the test sample group, the Spearman coefficient between the virtual ranking score and the source price of the real driver-cargo pair in the test sample group can be determined. The Spearman coefficient is used to analyze the consistency between the source price and the virtual ranking score in the ranking, and to quantitatively analyze whether the monotonic relationship of the price factor is enhanced.

[0115] Experiments showed that, for the supply ranking factor determination model trained using the training method provided in this application, for supplies with price competitiveness ranging from 10 to 90, adversarial testing was conducted using a test set with price increases not exceeding 10%. The average Spearman coefficient increased by 2.85 pp, and the median increased by 2.15 pp. Using a test set with price increases within 20%, the average Spearman coefficient increased by 0.74 pp, and the median increased by 0.86 pp. In terms of model matching performance, the click-through rate (GAUC) increased by 0.22 pp, and the conversion rate (GAUC) increased by 0.24 pp. The supply ranking factor determination model significantly improved its supply ranking matching performance, and the monotonic relationship of price factors was strengthened. Here, "pp" represents the absolute value of the direct subtraction of percentages; for example, 3% - 2% = 1% = 1 pp. GAUC is a metric used to evaluate model performance. It is calculated by grouping samples and then calculating the AUC (Area Under Curve) within each group, assessing the probability that positive samples have a higher prediction score than negative samples. AUC is also a metric used to evaluate model performance, referring to the area under the Receiver Operating Characteristic (ROC) curve.

[0116] Based on the training methods for the source ranking factor determination model in the above embodiments, this application also provides a method for determining source ranking factors. This method can be applied to an electronic device or to a source ranking factor determination device installed in the electronic device. The device can be implemented through software, hardware, or a combination of both. The electronic device may include at least one of the following: a server, mobile phone, computer, vehicle terminal, tablet computer, wearable device, and smart home device, but is not limited thereto. The server may include a dedicated server, a virtual server, and a cluster server.

[0117] The following example uses applications in electronic devices, combined with... Figures 3-4 The method for determining the source ranking factor provided in the embodiments of this application will be described in detail.

[0118] Figure 3 This illustration shows a flowchart of the method for determining the source ranking factor provided in an embodiment of this application. (Refer to...) Figure 3 As shown in the embodiments of this application, the method for determining the source ranking factor may include the following steps 310 to 340.

[0119] Step 310: Obtain historical transaction orders for goods within the first preset time period.

[0120] The first preset time period can be set according to actual needs, such as 24 hours, 48 ​​hours, etc., and this application does not make any special restrictions on it.

[0121] Step 320: Determine the bottom price characteristics based on historical transaction orders of the goods.

[0122] The floor price feature is used to characterize the floor price across multiple target dimensions and the relationship between the source price and the floor price. These multiple target dimensions may include at least two of the following: driver dimension, driver-route intersection dimension, and route-registered vehicle length intersection dimension, but are not limited to these.

[0123] Electronic devices can identify price information in historical orders related to goods, and then perform statistical analysis on this price information across multiple target dimensions based on the corresponding price attribute information to obtain bottom-line price characteristics. The price attribute information may include at least one of the following, but is not limited to: driver information, route information, vehicle length information, and goods information.

[0124] For historical order data, some freight is priced based on the number of trips, while others are priced based on weight, and even the same unit of measurement may have different pricing methods. Furthermore, the price distribution for less-than-truckload (LTL) and full truckload (FTL) freight is naturally not in the same space, with LTL prices generally being lower. Therefore, in one embodiment, before performing statistical analysis on the price information in historical order data across multiple target dimensions, the price information can be standardized to obtain the corresponding standardized freight price. Then, statistical analysis across multiple target dimensions can be performed based on the standardized freight price.

[0125] Specifically, step 320, which determines the baseline price feature based on historical freight transaction orders, may include: determining the minimum standardized price of freight in the driver dimension of historical freight transaction orders to obtain the first baseline price; determining the minimum standardized price of freight in the driver-route intersection dimension of historical freight transaction orders to obtain the second baseline price; determining the minimum standardized price of freight in the route-registered vehicle length intersection dimension of historical freight transaction orders to obtain the third baseline price; determining at least two of the first, second, and third baseline prices as baseline prices for multiple target dimensions; determining the target difference value between the baseline price and the freight price; and defining the baseline price and the target difference value as the baseline price feature.

[0126] The specific implementation principle of determining the bottom price feature based on the historical transaction orders of the source of goods is the same as the principle of determining the sample bottom price feature based on the historical transaction orders of the sample source of goods in steps 111 to 116 above. The specific implementation principle can be referred to in steps 111 to 116, and will not be repeated here.

[0127] Step 330: Obtain the price competitiveness characteristics of the goods within the second preset time period and the real-time price increase characteristics of the goods within the third preset time period.

[0128] The second and third preset time periods can be set according to actual needs; they can be the same or different. The price competitiveness characteristic of the goods reflects the competitiveness or value of the goods; the higher the competitiveness of the goods, the greater the probability and speed of a transaction. The real-time price increase characteristic of the goods reflects the personalized price increase behavior of the goods.

[0129] For example, step 330, obtaining the price competitiveness characteristics of the goods within the second preset time period and the real-time price increase characteristics of the goods within the third preset time period, may include: obtaining the price percentile of the first goods within the second preset time period in each of the divided sub-markets to obtain the price competitiveness characteristics of the goods; and obtaining the price increase characteristics of the second goods within the third preset time period to obtain the real-time price increase characteristics of the goods. The specific implementation principle is the same as that of steps 1211-1212 above, which obtain the sample goods price competitiveness characteristics and sample goods real-time price increase characteristics. For details, please refer to steps 1211-1212 above; they will not be repeated here.

[0130] Step 340: Input the bottom line price feature, the source price competitiveness feature, and the source real-time price increase feature as price sharing features into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model.

[0131] The product ranking factor determination model is trained based on the training method of the product ranking factor determination model described in any of the above embodiments; the product ranking factor is used to rank the products and may include the product's click-through rate and / or transaction rate.

[0132] For example, taking the product ranking factors, including the product's click-through rate and conversion rate, as an example, after determining the product's click-through rate and conversion rate, the product's click-through rate and conversion rate can be summed by a power weight to obtain the product's ranking score. Then, the products can be sorted in descending order of ranking score, and the ranking result can be used for product list recommendation, etc.

[0133] For example, the source ranking factor determination model can be a single-scenario deep neural network architecture model. Electronic devices can input the bottom line price feature, the source price competitiveness feature, and the source real-time price increase feature into the source ranking factor determination model, and determine the source ranking factor of the source through the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model.

[0134] For example, the model for determining the source ranking factor can be a neural network model with a multi-scenario, multi-task architecture. For instance, it can use an existing multi-scenario, multi-task neural network model, such as the HiNet model; or, it can be a new model formed by adding a price feature extraction layer to an existing multi-scenario, multi-task neural network model. For example, Figure 4 This illustration shows the structure and working principle of the source ranking factor determination model provided in this application embodiment. (Refer to...) Figure 4 As shown, the source ranking factor determination model may include a price feature extraction layer 401, a feature encoding layer 402, a scene extraction layer 403, a task extraction layer 404, and a task tower layer 405.

[0135] Combination Figure 4 The source ranking factor determination model shown in step 340 inputs the bottom line price feature, source price competitiveness feature, and source real-time price increase feature as price sharing features into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model. This can be achieved through the following steps 341 to 346.

[0136] Step 341: Obtain scene-related features.

[0137] Scene-related features can include scene-shared features, scene-specific features, scene-identifying features, and scene-attribute features. Scene-shared features represent features shared across different scenes; scene-specific features represent features unique to a single scene; scene-identifying features identify different scenes; and scene-attribute features represent scene attribute information, such as the scene's name.

[0138] For example, scene-sharing features may include at least one of the following, but are not limited to: the distance between the driver and the cargo loading point, the number of cargoes clicked by the driver in the fourth preset time period, and the ratio of cargo weight to the driver's vehicle load; scene-specific features may include at least one of the following, but are not limited to: the number of times the driver clicked on cargoes on routes in each scene in the past fifth preset time period, the duration of the driver's subscribed routes, and the number of times the driver searched in the sixth preset time period; scene identification features may include the corresponding scene number; scene attribute features may include the scene name.

[0139] Step 342: Input the bottom line price feature, the source price competitiveness feature, and the source real-time price increase feature as price sharing features, and input them together with the scenario-related features into the feature encoding layer 402. The feature encoding layer 402 encodes the price sharing feature and the scenario-related feature to obtain the price encoding feature and scenario encoding feature output by the feature encoding layer 402.

[0140] In one embodiment, the bottom-line price feature, the source price competitiveness feature, and the source real-time price increase feature can be used as price sharing features and directly input into the feature encoding layer 402 together with the scenario-related features. The feature encoding layer 402 performs feature encoding on the price sharing feature and the scenario-related features to obtain the price encoding feature and scenario encoding feature output by the feature encoding layer 402.

[0141] In another embodiment, the price-sharing features can be binned and then input together with scenario-related features into the feature encoding layer 402 for feature encoding. That is, the price-sharing features are discretized before feature encoding. The binning method can refer to the binning method described in step 1222 above, and will not be repeated here. Through this binning method, the granularity of the binning can be refined, the sensitivity of the bins to price changes can be improved, and thus the sensitivity of the model to price can be improved.

[0142] For example, for the source price competitiveness feature, it can be divided into first and second equal-interval bins, such as bins of 5 and 10 respectively. The binning results can be used as the target binning feature for the source price competitiveness feature. This can improve the binning effect of the source price competitiveness feature and enhance the sensitivity and effectiveness of the model.

[0143] Step 343: Input the price coding features into the price feature extraction layer 401, and extract the price features from the price coding features through the price feature extraction layer 401 to obtain the price features output by the price feature extraction layer 401.

[0144] The price feature extraction layer 401 is used to extract price features, and for example, it can be an MLP neural network structure.

[0145] Step 344: Input the price coding features and scene coding features into the scene extraction layer 403, and extract scene features from the price coding features and scene coding features through the scene extraction layer 403 to obtain the scene features output by the scene extraction layer 403.

[0146] The scene extraction layer 403 can be used to extract scene-specific information representations and valuable information shared between scenes from price coding features and scene coding features, thus obtaining scene features. By extracting scene features from price coding features and scene coding features, the ability of task layer information representation can be improved.

[0147] For example, such as Figure 4As shown, the scene extraction layer 403 may include a scene-shared expert network and a scene-specific expert network. The scene-shared coding features in the price coding features and scene coding features can be input into the scene-shared expert network and the scene-specific expert network of the scene extraction layer 403, respectively. At the same time, the scene-specific features in the scene coding features are input into the scene-specific expert network. The scene-shared expert network extracts the common information between different scenes, and the scene-specific expert network learns the scene-specific information. The scene extraction layer 403 outputs the extracted common information between different scenes and the scene-specific information as scene features.

[0148] For example, such as Figure 4 As shown, the scene extraction layer 403 may also include a scene attention network, which can input price encoding features and all scene encoding features into the scene attention network for scene perception, thereby obtaining scene attention features. The obtained scene attention features, together with the features output by the scene common expert network and the scene specific expert network, are used as the scene features output by the scene extraction layer 403, thereby measuring the importance of the contribution of other scene information to the representation of the current scene information, and thus enhancing the information expression ability of the current scene.

[0149] Step 345: Input the scene features into the task extraction layer 404, and extract the task features from the scene features through the task extraction layer 404 to obtain the task features output by the task extraction layer 404.

[0150] The scene features output by scene extraction layer 403 can be input into task extraction layer 404 after vector concatenation. Task extraction layer 404 then extracts task features from the scene features. Specifically, task extraction layer 404 can be used to extract shared information of all tasks and task-specific information from the scene features to obtain task features.

[0151] For example, such as Figure 4 As shown, the task extraction layer 404 may include a task-sharing expert network and a task-specific expert network. Scene features can be input into the task-sharing expert network and the task-specific expert network, respectively. The task-sharing expert network learns the shared information of all tasks in the current scene, and the task-specific expert network extracts the specific information of each task in the current scene. The task extraction layer 404 outputs the extracted shared information of all tasks and the specific information of each task as task features. It can be understood that the task-specific expert network can be designed according to the number of tasks, with each task corresponding to one task-specific expert network.

[0152] Step 346: Input the task features and price features into the task tower layer 405, perform task identification through the task tower layer 405, and determine the source ranking factor output by the task tower layer 405 as the source ranking factor output by the source ranking factor determination model.

[0153] In this way, by concatenating the additional extracted price features into the input vector of the task tower layer 405 and performing task recognition together with the task features output by the task extraction layer 404, the shallow expression of price features can be improved, thereby improving the sensitivity and monotonicity of the task tower layer 405 and the model to price factors.

[0154] according to Figure 4 The model for determining the sourcing ranking factor, as shown, involves price-sharing features that, after input discretization and feature encoding, are fed into a scenario-sharing expert network and a scenario-specific expert network for information extraction. These are then processed by a task extraction layer 404 and a task pyramid layer 405 to learn the targets for clicks and / or transactions. However, the price feature, a crucial decision-making factor in the business scenario, faces information decay or gradient vanishing issues in the multi-layer neural network's information extraction process without a price feature extraction layer 401. As the network depth increases, the numerical information and gradient signals of the price feature may gradually weaken or even be overwhelmed by other features, leading to the partial loss of critical price decision information during transmission. This ultimately affects the model's pricing accuracy and business decision-making effectiveness. Figure 4 The supply ranking factor determination model shown is based on existing multi-scenario, multi-task neural network models. It additionally designs a unique information extraction network, namely the price feature extraction layer 401, for several price-related features, and retains this information as input to both the scenario-shared expert network and the scenario-specific expert network. The information extracted by the price feature extraction layer 401 is directly concatenated into the vector output by the task extraction layer 404, and together they enter the task tower layer 405 for task recognition. This improves the shallow representation of price features, thereby enhancing the sensitivity and monotonicity of the task tower layer 405 and the supply ranking factor determination model to price factors.

[0155] The method for determining the product ranking factors provided in this application comprehensively considers the characteristics of price factors in multiple different dimensions, as well as the correlation between price features and user behavior, which greatly enriches the price-related features, thereby improving the product ranking matching effect and ensuring the monotonicity of product prices.

[0156] It is understood that the training method and the method for determining the source ranking factor of the model provided in this application embodiment can be applied to a single scenario or multiple scenarios.

[0157] This application also provides a training device for a source ranking factor determination model. Figure 5 This diagram illustrates the structure of the training device for the source ranking factor determination model. Figure 5 As shown, the training device for this source ranking factor determination model may include: The first acquisition module 510 is used to acquire historical transaction orders of sample goods. The first determining module 520 is used to determine the sample floor price characteristics based on the historical transaction orders of the sample goods. The sample floor price characteristics are used to characterize the sample floor price of multiple target dimensions and the relationship between the sample goods price and the sample floor price. The second acquisition module 530 is used to acquire the price competitiveness characteristics of the sample goods and the real-time price increase characteristics of the sample goods. Training module 540 is used to train the initial source ranking factor determination model based on the sample bottom price characteristics, sample source price competitiveness characteristics, and sample source real-time price increase characteristics, to obtain the trained source ranking factor determination model; the source ranking factor determination model is used to determine the source ranking factor of the source, which includes the source click rate and / or transaction rate.

[0158] In one embodiment, the multiple target dimensions include at least two of the following dimensions: driver dimension, driver-route intersection dimension, and route-registered vehicle length intersection dimension. Correspondingly, the first determining module 520 is specifically used to: determine the minimum standardized price of the sample cargo in the driver dimension of historical transaction orders to obtain a first sample baseline price; determine the minimum standardized price of the sample cargo in the driver-route intersection dimension of historical transaction orders to obtain a second sample baseline price; determine the minimum standardized price of the sample cargo in the route-registered vehicle length intersection dimension of historical transaction orders to obtain a third sample baseline price; determine at least two of the first, second, and third sample baseline prices as sample baseline prices for the multiple target dimensions; determine the sample target difference value between the sample baseline price and the sample cargo price; the sample target difference value includes the sample absolute difference and / or sample relative difference; and determine the sample baseline price and the sample target difference value as sample baseline price features.

[0159] In one embodiment, the second acquisition module 530 is specifically used to: acquire the price percentile of the first sample source within a preset time period of the first sample in each of the divided sample sub-markets, and obtain the price competitiveness characteristics of the sample source; acquire the sample price increase characteristics of the second sample source within a preset time period of the second sample, and obtain the real-time price increase characteristics of the sample source; wherein, the sample price increase characteristics include at least one of the sample price increase amount, the first sample time interval between the price increase time and the release time of the second sample source, and the second sample time interval from the last price increase to the current time.

[0160] In one embodiment, the initial supply ranking factor determination model includes an initial price feature extraction layer, an initial feature encoding layer, an initial scenario extraction layer, an initial task extraction layer, and an initial task pyramid layer. Correspondingly, the training module 540 is specifically used to: use the sample floor price feature, the sample supply price competitiveness feature, and the sample supply real-time price increase feature as sample price sharing features, and obtain sample scenario-related features; input the sample price sharing features and the sample scenario-related features into the initial feature encoding layer, and perform feature encoding on the sample price sharing features and sample scenario-related features through the initial feature encoding layer to obtain the sample price encoding features and sample scenario encoding features output by the initial feature encoding layer; input the sample price encoding features into the initial price feature extraction layer, and perform price feature extraction on the sample price encoding features through the initial price feature extraction layer to obtain the sample price features output by the initial price feature extraction layer. The sample price encoding features and sample scene encoding features are input into the initial scene extraction layer. The initial scene extraction layer extracts scene features from the sample price encoding features and sample scene encoding features to obtain the sample scene features output by the initial scene extraction layer. The sample scene features are input into the initial task extraction layer. The initial task extraction layer extracts task features from the sample scene features to obtain the sample task features output by the initial task extraction layer. The sample task features and sample price features are input into the initial task pyramid layer. The initial task pyramid layer performs task recognition to obtain the sample supply ranking factor output by the initial task pyramid layer. The sample supply ranking factor includes the sample click-through rate and / or the sample transaction rate. Based on the sample supply ranking factor and sample label data, the model parameters of the initial supply ranking factor determination model are adjusted until the initial supply ranking factor determination model converges to obtain the trained supply ranking factor determination model.

[0161] In one embodiment, when the training module 540 inputs the sample price sharing features and sample scene-related features into the initial feature encoding layer, and performs feature encoding on the sample price sharing features and sample scene-related features through the initial feature encoding layer to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer, the specific steps are as follows: Each type of sample price sharing feature is used as a sample to be binned feature; the non-default value price features in the sample to be binned features are binned equally frequently to obtain a first sample bin; the default value price features in the sample to be binned features are used as a new bin and appended to the boundary of the first sample bin to form a second sample bin for the sample to be binned features together with the first sample bin; based on a preset gain coefficient, the bin boundary value of the second sample bin is increased to obtain the target sample binning features corresponding to the sample to be binned features; the target sample binning features and sample scene-related features corresponding to all types of sample price sharing features are input into the initial feature encoding layer, and the initial feature encoding layer performs feature encoding on all target sample binning features and sample scene-related features to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer.

[0162] In one embodiment, the training device for the source ranking factor determination model further includes a testing module, which is used to: collect real driver-cargo pairs samples and increase the prices of the real driver-cargo pairs samples based on a preset price increase ratio to obtain virtual driver-cargo pairs samples; determine the real driver-cargo pairs samples and virtual driver-cargo pairs samples as test samples; for each test sample group in the test samples, input the virtual driver-cargo pairs samples in the test sample group into the source ranking factor determination model to obtain the virtual source ranking factor output by the source ranking factor determination model, wherein the virtual source ranking factor includes virtual click-through rate and / or virtual transaction rate; determine the virtual ranking score of the virtual driver-cargo pairs samples in the test sample group based on the virtual source ranking factor; determine the Spearman correlation coefficient between the virtual ranking score of the test sample group and the source price of the real driver-cargo pairs samples in the test sample group; the Spearman correlation coefficient is used to characterize the price monotonicity of the source ranking factor determination model.

[0163] The training device for determining the source ranking factor model provided in this application embodiment has the same implementation principle and beneficial effects as the training method for determining the source ranking factor model provided in the above embodiment, and will not be repeated here.

[0164] This application also provides a device for determining the source ranking factor. Figure 6 A schematic diagram of the source ranking factor determination device is shown below. Figure 6 As shown, the source ranking factor determination device may include: The third acquisition module 610 is used to acquire historical transaction orders of goods within the first preset time period; The second determining module 620 is used to determine the bottom line price characteristics based on the historical transaction orders of the goods. The bottom line price characteristics are used to characterize the bottom line price of multiple target dimensions and the relationship between the goods price and the bottom line price. The fourth acquisition module 630 is used to acquire the price competitiveness characteristics of the goods within the second preset time period and the real-time price increase characteristics of the goods within the third preset time period; The third determining module 640 is used to input the bottom price feature, the source price competitiveness feature, and the source real-time price increase feature as price sharing features into the source ranking factor determining model to obtain the source ranking factor output by the source ranking factor determining model; wherein, the source ranking factor determining model is trained based on the training method of the source ranking factor determining model described in any of the above embodiments; the source ranking factor includes the source click rate and / or transaction rate.

[0165] In one embodiment, the source ranking factor determination model includes a price feature extraction layer, a feature encoding layer, a scenario extraction layer, a task extraction layer, and a task pyramid layer; correspondingly, the third determination module 640 is specifically used for: acquiring scenario-related features; using the bottom-line price feature, source price competitiveness feature, and source real-time price increase feature as price-sharing features, and inputting them together with the scenario-related features into the feature encoding layer, and performing feature encoding on the price-sharing features and scenario-related features through the feature encoding layer to obtain the price encoding features and scenario encoding features output by the feature encoding layer; inputting the price encoding features into the price feature extraction layer, and performing feature encoding on the price-sharing features through the price feature extraction layer... Price features are extracted from code features to obtain price features output by the price feature extraction layer. Price and scene coding features are input into the scene extraction layer, which extracts scene features from them to obtain scene features output by the scene extraction layer. Scene features are input into the task extraction layer, which extracts task features from them to obtain task features output by the task extraction layer. Task features and price features are input into the task pyramid layer, which performs task identification. The source ranking factor output by the task pyramid layer is determined as the source ranking factor output by the source ranking factor determination model.

[0166] The source ranking factor determination device provided in this application embodiment has the same implementation principle and beneficial effects as the source ranking factor determination method provided in the above embodiments, and will not be repeated here.

[0167] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0168] This application also provides an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the steps of the training method for the source ranking factor determination model described in any of the above method embodiments, or the steps of the source ranking factor determination method described in any of the above method embodiments, which will not be repeated here.

[0169] Based on the training method or the method for determining the source ranking factor described in any of the above embodiments, this application also provides a computer-readable storage medium. For example, a non-transitory computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, magnetic tape, a floppy disk, or an optical data storage device. This storage medium stores computer instructions for executing the training method or the method for determining the source ranking factor described in any of the above embodiments, which will not be elaborated further here.

[0170] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware to implement them. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0171] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations thereof that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

Claims

1. A training method for a source ranking factor determination model, characterized in that, include: Obtain historical transaction orders for sample goods, and determine sample floor price features based on the historical transaction orders for sample goods. The sample floor price features are used to characterize the sample floor price of multiple target dimensions and the relationship between the sample goods price and the sample floor price. The price competitiveness characteristics and real-time price increase characteristics of sample goods are obtained, and the initial goods ranking factor determination model is trained based on the sample bottom price characteristics, the sample goods price competitiveness characteristics, and the sample goods real-time price increase characteristics to obtain the trained goods ranking factor determination model; the goods ranking factor determination model is used to determine the goods ranking factor of the goods, and the goods ranking factor includes the click-through rate and / or transaction rate of the goods.

2. The training method for the source ranking factor determination model according to claim 1, characterized in that, The multiple target dimensions include at least two of the following: driver dimension, driver-route intersection dimension, and route-registered vehicle length intersection dimension; the determination of the sample floor price feature based on the historical transaction orders of the sample cargo source includes: The minimum standardized price of the sample cargo in the driver dimension of the historical transaction orders of the sample cargo is determined to obtain the first sample bottom line price; The minimum standardized price of the sample cargo in the driver-route intersection dimension of the historical transaction orders of the sample cargo is determined to obtain the second sample bottom line price; The minimum standardized price of the sample cargo in the historical transaction orders of the sample cargo, based on the intersection dimension of the route and the registered vehicle length, is determined to obtain the third sample bottom line price; At least two of the first sample baseline price, the second sample baseline price, and the third sample baseline price are determined as the sample baseline prices for the plurality of target dimensions; Determine the target difference between the sample floor price and the sample source price; the target difference includes the absolute difference and / or the relative difference. The difference between the sample floor price and the sample target price is defined as the sample floor price feature.

3. The training method for the source ranking factor determination model according to claim 1, characterized in that, The acquisition of the price competitiveness characteristics and real-time price increase characteristics of the sample goods includes: Obtain the price percentile of the first sample source within a preset time period in each of the divided sample sub-markets to obtain the price competitiveness characteristics of the sample source. Obtain the sample price increase characteristics of the second sample source within a preset time period of the second sample, and obtain the real-time price increase characteristics of the sample source; wherein, the sample price increase characteristics include at least one of the sample price increase amount, the first sample time interval between the price increase time and the release time of the second sample source, and the second sample time interval between the last price increase and the current time.

4. The training method for the source ranking factor determination model according to claim 1, characterized in that, The initial supply ranking factor determination model includes an initial price feature extraction layer, an initial feature encoding layer, an initial scenario extraction layer, an initial task extraction layer, and an initial task pyramid layer; the training of the initial supply ranking factor determination model based on the sample floor price features, the sample supply price competitiveness features, and the sample supply real-time price increase features includes: The sample floor price feature, the sample source price competitiveness feature, and the sample source real-time price increase feature are used as sample price sharing features, and sample scenario-related features are obtained. The sample price sharing feature and the sample scenario related feature are input into the initial feature encoding layer. The initial feature encoding layer encodes the sample price sharing feature and the sample scenario related feature to obtain the sample price encoding feature and sample scenario encoding feature output by the initial feature encoding layer. The sample price encoding features are input into the initial price feature extraction layer, and the initial price feature extraction layer extracts price features from the sample price encoding features to obtain the sample price features output by the initial price feature extraction layer. The sample price encoding feature and the sample scene encoding feature are input into the initial scene extraction layer. The initial scene extraction layer extracts scene features from the sample price encoding feature and the sample scene encoding feature to obtain the sample scene features output by the initial scene extraction layer. The sample scene features are input into the initial task extraction layer, and the initial task extraction layer extracts task features from the sample scene features to obtain the sample task features output by the initial task extraction layer. The sample task features and sample price features are input into the initial task tower layer, and task identification is performed through the initial task tower layer to obtain the sample source ranking factor output by the initial task tower layer; the sample source ranking factor includes sample click rate and / or sample transaction rate. The model parameters of the initial source ranking factor determination model are adjusted based on the sample source ranking factor and sample label data until the initial source ranking factor determination model converges, thus obtaining the trained source ranking factor determination model.

5. The training method for the source ranking factor determination model according to claim 4, characterized in that, The step of inputting the sample price sharing feature and the sample scene-related feature into the initial feature encoding layer, and then performing feature encoding on the sample price sharing feature and the sample scene-related feature through the initial feature encoding layer to obtain the sample price encoding feature and sample scene encoding feature output by the initial feature encoding layer includes: Each type of sample price sharing feature is used as a sample bucketing feature. The non-default value price features in the sample bucketing feature are bucketed by equal frequency to obtain the first sample bucketing. The default value price feature in the sample to be binned is taken as a new bin and added to the boundary of the first sample bin, together with the first sample bin to form the second sample bin of the sample to be binned feature; Based on a preset gain coefficient, the bucket boundary value of the second sample bucket is increased to obtain the target sample bucket feature corresponding to the sample bucket feature to be bucketed. The target sample binning features and the sample scene-related features corresponding to the shared features of all types of sample prices are input into the initial feature encoding layer. The initial feature encoding layer encodes all the target sample binning features and the sample scene-related features to obtain the sample price encoding features and sample scene encoding features output by the initial feature encoding layer.

6. The training method for the source ranking factor determination model according to any one of claims 1 to 5, characterized in that, After obtaining the trained source ranking factor determination model, the training method of the source ranking factor determination model further includes: Collect real driver-cargo pairs samples, and increase the prices in the real driver-cargo pairs samples based on a preset price increase ratio to obtain virtual driver-cargo pairs samples; The real driver-cargo pair sample and the virtual driver-cargo pair sample are determined as test samples; For each test sample group in the test sample, the virtual driver-cargo pair samples in the test sample group are input into the cargo source ranking factor determination model to obtain the virtual cargo source ranking factor output by the cargo source ranking factor determination model; the virtual cargo source ranking factor includes virtual click-through rate and / or virtual transaction rate; The virtual ranking score of the virtual driver pair samples in the test sample group is determined based on the virtual cargo source ranking factor. Determine the Spearman correlation coefficient between the virtual ranking score of the test sample group and the source price of the real cargo pair sample in the test sample group; the Spearman correlation coefficient is used to characterize the price monotonicity of the source ranking factor determination model.

7. A method for determining the source ranking factor, characterized in that, include: Retrieve historical transaction orders for goods within the first preset time period; The baseline price feature is determined based on the historical transaction orders of the goods. The baseline price feature is used to characterize the baseline price of multiple target dimensions and the relationship between the price of the goods and the baseline price. Obtain the price competitiveness characteristics of goods within the second preset time period and the real-time price increase characteristics of goods within the third preset time period; The bottom line price feature, the source price competitiveness feature, and the source real-time price increase feature are used as price sharing features and input into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model. The product ranking factor determination model is trained based on the training method of the product ranking factor determination model as described in any one of claims 1 to 6; the product ranking factor includes the product's click-through rate and / or transaction rate.

8. The method for determining the source ranking factor according to claim 7, characterized in that, The source ranking factor determination model includes a price feature extraction layer, a feature encoding layer, a scenario extraction layer, a task extraction layer, and a task tower layer; the process of inputting the baseline price feature, the source price competitiveness feature, and the source real-time price increase feature as price sharing features into the source ranking factor determination model to obtain the source ranking factor output by the source ranking factor determination model includes: Obtain scene-related features; The baseline price feature, the source price competitiveness feature, and the source real-time price increase feature are used as price sharing features and input together with the scenario-related features into the feature encoding layer. The feature encoding layer encodes the price sharing feature and the scenario-related feature to obtain the price encoding feature and scenario encoding feature output by the feature encoding layer. The price coding feature is input into the price feature extraction layer, and the price feature extraction layer extracts price features from the price coding feature to obtain the price features output by the price feature extraction layer. The price encoding feature and the scene encoding feature are input into the scene extraction layer. The scene extraction layer extracts scene features from the price encoding feature and the scene encoding feature to obtain the scene features output by the scene extraction layer. The scene features are input into the task extraction layer, and the task extraction layer extracts task features from the scene features to obtain the task features output by the task extraction layer. The task features and price features are input into the task tower layer, and task identification is performed through the task tower layer. The source ranking factor output by the task tower layer is determined as the source ranking factor output by the source ranking factor determination model.

9. An electronic device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the training method for the source ranking factor determination model as described in any one of claims 1 to 6, or the steps of the source ranking factor determination method as described in claim 7 or 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the training method for the source ranking factor determination model as described in any one of claims 1 to 6, or the steps of the source ranking factor determination method as described in claim 7 or 8.

Images