Training method of express claim settlement model, express claim settlement prediction method and device
By determining sampling weight information in the express delivery claims model and performing weighted sampling, the problem of low classification accuracy in existing express delivery claims models is solved, and higher prediction accuracy is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SF TECH CO LTD
- Filing Date
- 2021-06-29
- Publication Date
- 2026-06-16
AI Technical Summary
The classification accuracy of existing express delivery claims models is low, especially in cases of imbalanced samples, where random sampling methods miss a lot of hidden information, resulting in low accuracy of the classification models.
By acquiring the first express delivery claim sample set and the pre-trained express delivery claim classification model, the prediction results of candidate negative samples are determined, and the sampling weight information is determined based on the prediction results to perform weighted sampling, forming the third express delivery claim sample set. Finally, the pre-trained model is trained to obtain the target express delivery claim model.
The prediction accuracy of the express delivery claims model has been improved. By weighted sampling of sampling weight information, more feature information is obtained, and the trained target model has higher prediction ability.
Smart Images
Figure CN115545249B_ABST
Abstract
Description
Technical Field
[0001] This application mainly relates to the field of big data technology, specifically to a training method for a parcel claim model, a prediction method for parcel claims, and a device. Background Technology
[0002] With the development of mobile internet and the logistics industry, the volume of express parcels in my country is rapidly increasing every year. Parcels may be damaged or lost during transportation, leading to customer claims against logistics companies. Some companies use machine learning methods to predict whether a claim will be made, aiming to take preventative measures and mitigate losses beforehand. However, the ratio of claimable to non-claimable parcels is extremely imbalanced, making it difficult for models to fit the data. For classification models, after training, a decision boundary is fitted, and then the samples are classified based on this boundary. The further a sample is from the decision boundary, the higher the model's confidence in its classification. In imbalanced sampling problems, it is often necessary to sample some samples from the category with the larger sample size to avoid severe data skew. How to extract the samples that are most helpful for model classification from a large amount of data is a problem that sampling mechanisms need to solve. Most existing technologies use random sampling, which is simple to operate but also misses a large amount of hidden information, resulting in low accuracy for classification models in classifying the sampled samples.
[0003] In other words, the accuracy of claims classification in existing express delivery claims models is relatively low. Summary of the Invention
[0004] This application provides a training method for a parcel claim model, a prediction method for parcel claims, and an apparatus, aiming to solve the problem of low accuracy in the classification of claims by existing parcel claim models.
[0005] Firstly, this application provides a training method for a parcel claim model, the training method comprising:
[0006] Obtain a first express delivery claim sample set and a pre-trained express delivery claim classification model, wherein the pre-trained express delivery claim classification model is a model trained on a preset classification model using a second express delivery claim sample set, and the second express delivery claim sample set is a subset of the first express delivery claim sample set;
[0007] Multiple candidate negative samples from the first express delivery claim sample set are input into the pre-trained express delivery claim classification model to obtain the prediction results of each candidate negative sample.
[0008] The sampling weight information is determined based on the prediction results of each candidate negative sample;
[0009] Based on the sampling weight information, the first express delivery claim sample set is weighted and sampled to obtain the third express delivery claim sample set.
[0010] The pre-trained express claim classification model is trained based on the third express claim sample set to obtain the target express claim model.
[0011] Optionally, training the pre-trained express delivery claim classification model based on the third express delivery claim sample set to obtain the target express delivery claim model includes:
[0012] The third express delivery claim sample set is determined as the second express delivery claim sample set and subjected to multiple iterative weighted sampling to obtain the third express delivery claim sample set after multiple iterations.
[0013] The pre-trained express claim classification model is trained based on the third express claim sample set after multiple iterations to obtain the target express claim model.
[0014] Optionally, the prediction result for each candidate negative sample includes the predicted probability that each candidate negative sample is predicted as a positive sample category;
[0015] The determination of sampling weight information based on the prediction results of each candidate negative sample includes:
[0016] The confidence modulus of each candidate negative sample is determined based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0017] The sampling weight information is determined based on the confidence modulus of each candidate negative sample.
[0018] Optionally, the sampling weight information includes the sampling weight coefficients for each preset modulus interval;
[0019] The determination of the sampling weight information based on the confidence modulus of each candidate negative sample includes:
[0020] Obtain the number of sample distributions of the multiple candidate negative samples in each preset modulus interval;
[0021] The sampling weight coefficient of each preset module length interval is determined based on the number of samples distributed in each preset module length interval, wherein the sampling weight coefficient of the preset module length interval decreases as the number of samples distributed in the preset module length interval increases.
[0022] Optionally, the plurality of preset modulus intervals are a plurality of modulus intervals of equal length that are continuously distributed in the interval [0,1].
[0023] The determination of the confidence modulus of each candidate negative sample based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample includes:
[0024] The initial confidence modulus of each candidate negative sample is determined based on the difference between the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0025] The initial confidence modulus of each candidate negative sample is normalized based on the initial confidence modulus of each candidate negative sample, the minimum confidence modulus among the initial confidence modulus of each candidate negative sample, and the maximum confidence modulus among the initial confidence modulus of each candidate negative sample, to obtain the confidence modulus of each candidate negative sample.
[0026] Optionally, the step of weighting the first express delivery claim sample set based on the sampling weight information to obtain the third express delivery claim sample set includes:
[0027] Based on the sampling weight coefficients of each preset modulus interval, the candidate negative samples of the multiple candidate negative samples in each preset modulus interval are sampled to obtain multiple negative sample sampling sets;
[0028] The third express delivery claim sample set is determined by combining multiple negative sample sets with the positive samples in the first express delivery claim sample set.
[0029] Optionally, obtaining the first express delivery claim sample set and the pre-trained express delivery claim classification model includes:
[0030] Retrieve multiple shipment information and corresponding claim information for multiple historical shipments within a preset historical time period;
[0031] Based on the claim information corresponding to each express shipment, the multiple express shipment information are labeled as positive sample type and negative sample type to obtain the first express shipment claim sample set.
[0032] Secondly, this application provides a method for predicting express delivery claims, the method comprising:
[0033] Obtain the shipment information of the shipment to be predicted;
[0034] The shipment information of the shipment to be predicted is input into the target shipment claim model to obtain the predicted probability that the shipment to be predicted is a positive sample category, wherein the target shipment claim model is any one of the target shipment claim models described in the first aspect.
[0035] Thirdly, this application provides a training device for a parcel claim model, the training device for the parcel claim model comprising:
[0036] The acquisition unit is used to acquire a first express delivery claim sample set and a pre-trained express delivery claim classification model, wherein the pre-trained express delivery claim classification model is a model trained on a preset classification model using a second express delivery claim sample set, and the second express delivery claim sample set is a subset of the first express delivery claim sample set;
[0037] The prediction unit is used to input multiple candidate negative samples from the first express claim sample set into the pre-trained express claim classification model to obtain the prediction results of each candidate negative sample.
[0038] The weight calculation unit is used to determine the sampling weight information based on the prediction results of each candidate negative sample;
[0039] A weighted sampling unit is used to perform weighted sampling on the first express claim sample set based on the sampling weight information to obtain a third express claim sample set.
[0040] The model training unit is used to train the pre-trained express claim classification model based on the third express claim sample set to obtain the target express claim model.
[0041] Optionally, the model training unit is used for:
[0042] The third express delivery claim sample set is determined as the second express delivery claim sample set and subjected to multiple iterative weighted sampling to obtain the third express delivery claim sample set after multiple iterations.
[0043] The pre-trained express claim classification model is trained based on the third express claim sample set after multiple iterations to obtain the target express claim model.
[0044] Optionally, the prediction result for each candidate negative sample includes the predicted probability that each candidate negative sample is predicted as a positive sample category;
[0045] The weight calculation unit is used for:
[0046] The confidence modulus of each candidate negative sample is determined based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0047] The sampling weight information is determined based on the confidence modulus of each candidate negative sample.
[0048] Optionally, the sampling weight information includes the sampling weight coefficients for each preset modulus interval;
[0049] The weight calculation unit is used for:
[0050] Obtain the number of sample distributions of the multiple candidate negative samples in each preset modulus interval;
[0051] The sampling weight coefficient of each preset module length interval is determined based on the number of samples distributed in each preset module length interval, wherein the sampling weight coefficient of the preset module length interval decreases as the number of samples distributed in the preset module length interval increases.
[0052] Optionally, the plurality of preset modulus intervals are a plurality of modulus intervals of equal length that are continuously distributed in the interval [0,1].
[0053] The weight calculation unit is used for:
[0054] The initial confidence modulus of each candidate negative sample is determined based on the difference between the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0055] The initial confidence modulus of each candidate negative sample is normalized based on the initial confidence modulus of each candidate negative sample, the minimum confidence modulus among the initial confidence modulus of each candidate negative sample, and the maximum confidence modulus among the initial confidence modulus of each candidate negative sample, to obtain the confidence modulus of each candidate negative sample.
[0056] Optionally, the weighted sampling unit is used for:
[0057] Based on the sampling weight coefficients of each preset modulus interval, the candidate negative samples of the multiple candidate negative samples in each preset modulus interval are sampled to obtain multiple negative sample sampling sets;
[0058] The third express delivery claim sample set is determined by combining multiple negative sample sets with the positive samples in the first express delivery claim sample set.
[0059] Optionally, the acquisition unit is configured to:
[0060] Retrieve multiple shipment information and corresponding claim information for multiple historical shipments within a preset historical time period;
[0061] Based on the claim information corresponding to each express shipment, the multiple express shipment information are labeled as positive sample type and negative sample type to obtain the first express shipment claim sample set.
[0062] Fourthly, this application provides a predictive device for express delivery claims, the predictive device for express delivery claims comprising:
[0063] The acquisition unit is used to acquire the express information of the express shipment to be predicted;
[0064] The prediction unit is used to input the express information of the express to be predicted into the target express claim model to obtain the predicted probability that the express to be predicted is a positive sample category, wherein the target express claim model is any one of the target express claim models described in the first aspect.
[0065] Fifthly, this application provides a computer device, the computer device comprising:
[0066] One or more processors;
[0067] Memory; and
[0068] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the training method for the express delivery claims model as described in any one of the first aspects or the prediction method for express delivery claims as described in any one of the second aspects.
[0069] Sixthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform steps in the training method of the express delivery claims model as described in the first aspect or the prediction method of express delivery claims as described in the second aspect.
[0070] This application provides a training method for a parcel claim model, a prediction method for parcel claims, and an apparatus. The training method for the parcel claim model first predicts multiple candidate negative samples in a first parcel claim sample set to obtain the prediction results of each candidate negative sample. Based on the prediction results of the candidate negative samples, sampling weight information is determined. Then, the first parcel claim sample set is weighted and sampled according to the sampling weight information to obtain a third parcel claim sample set. The collected third parcel claim sample set has more feature information than the randomly sampled claim sample set. The target parcel claim model trained based on the third parcel claim sample set has a more accurate prediction ability and can improve the prediction accuracy of parcel claims. Attached Figure Description
[0071] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0072] Figure 1 This is a schematic diagram of a scenario for the express delivery claims prediction system provided in this application embodiment;
[0073] Figure 2 This is a schematic flowchart of an embodiment of the training method for the express delivery claims model provided in this application.
[0074] Figure 3 This is a schematic flowchart of an embodiment of S203 in this application;
[0075] Figure 4 This is a schematic flowchart of an embodiment of the express delivery claim prediction method provided in this application.
[0076] Figure 5 This is a schematic diagram of an embodiment of the training device for the express delivery claims model provided in this application.
[0077] Figure 6 This is a schematic diagram of an embodiment of the express delivery claim prediction device provided in this application.
[0078] Figure 7 This is a schematic diagram of an embodiment of the computer device provided in this application. Detailed Implementation
[0079] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0080] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.
[0081] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0082] This application provides a training method for a courier claims model, a prediction method for courier claims, an apparatus, a computer device, and a storage medium, which will be described in detail below.
[0083] Please see Figure 1 , Figure 1 This is a schematic diagram of a scenario for the express delivery claims prediction system provided in this application embodiment. The express delivery claims prediction system may include a computer device 100, in which a training device for an express delivery claims model and / or a prediction device for express delivery claims are integrated.
[0084] In this embodiment, the computer device 100 can be a standalone server, a server network, or a server cluster. For example, the computer device 100 described in this embodiment includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing.
[0085] In this embodiment, the computer device 100 described above can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device 100 can be a desktop computer, a portable computer, a network server, a handheld computer (Personal Digital Assistant, PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, an embedded device, etc. This embodiment does not limit the type of computer device 100.
[0086] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the solution in this application and does not constitute a limitation on the application scenario of the solution in this application. Other application environments may include more than one application scenario. Figure 1 The number of computer devices shown is more or less, for example Figure 1 Only one computer device is shown in the document. It is understood that the express delivery claims prediction system may also include one or more other computer devices capable of processing data, which are not specifically limited here.
[0087] In addition, such as Figure 1 As shown, the express delivery claim prediction system may also include a memory 200 for storing data.
[0088] It should be noted that, Figure 1 The schematic diagram of the express delivery claims prediction system shown is merely an example. The express delivery claims prediction system and scenarios described in this application are for the purpose of more clearly illustrating the technical solutions of this application, and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of the express delivery claims prediction system and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.
[0089] First, this application provides a method for training a parcel claim model, comprising: acquiring a first parcel claim sample set and a pre-trained parcel claim classification model, wherein the pre-trained parcel claim classification model is a model trained on a preset classification model using a second parcel claim sample set, and the second parcel claim sample set is a subset of the first parcel claim sample set; inputting multiple candidate negative samples from the first parcel claim sample set into the pre-trained parcel claim classification model to obtain prediction results for each candidate negative sample; determining sampling weight information based on the prediction results of each candidate negative sample; performing weighted sampling on the first parcel claim sample set based on the sampling weight information to obtain a third parcel claim sample set; and training the pre-trained parcel claim classification model based on the third parcel claim sample set to obtain a target parcel claim model.
[0090] like Figure 2 As shown, Figure 2 This is a schematic flowchart of an embodiment of the training method for the express delivery claims model provided in this application. The training method includes the following steps S201 to S204:
[0091] S201. Obtain the first express delivery claim sample set and the pre-trained express delivery claim classification model.
[0092] Among them, the pre-trained express delivery claim classification model is a model obtained by training the preset classification model using the second express delivery claim sample set, which is a subset of the first express delivery claim sample set.
[0093] In this embodiment of the application, the express delivery claim sample set includes positive samples and negative samples. Positive samples refer to express delivery samples for which claims have been processed, while negative samples refer to express delivery samples for which claims have not been processed.
[0094] In a specific embodiment, obtaining the first express delivery claims sample set and the pre-trained express delivery claims classification model may include:
[0095] (1) Obtain multiple parcel information and corresponding claim information for multiple historical parcels within a preset historical time period.
[0096] The preset historical time period can be one month, two months, etc., which can be set according to the specific situation.
[0097] The express shipment information includes at least one of the following: the origin of the express shipment, the type of item being shipped, the shipping cost, whether the express shipment is insured, and the weight of the express shipment.
[0098] The claims information includes a label indicating whether the shipment has been claimed.
[0099] (2) Based on the claim information corresponding to each express item, multiple express item information are labeled as positive sample type and negative sample type to obtain the first express item claim sample set.
[0100] Specifically, parcel information marked with "1" indicating whether a claim has been filed is labeled as a positive sample type, meaning that the sample has undergone a claim, and "1" can be used as the true label for this type of sample; parcel information marked with "0" indicating whether a claim has been filed is labeled as a negative sample type, meaning that the sample has not undergone a claim, and "0" can be used as the true label for this type of sample. The labeled positive and negative samples are determined as the first parcel claim sample set.
[0101] The pre-trained express delivery claims classification model is obtained by training a pre-defined classification model using a second express delivery claims sample set. The second express delivery claims sample set is a subset of the first express delivery claims sample set. The pre-trained express delivery claims classification model is obtained by training the pre-defined classification model using the second express delivery claims sample set through supervised learning. The pre-defined classification model can be a LightGBM model, a GBDT model, or an XGBoost model. Pre-training the pre-trained express delivery claims classification model using a subset of the first express delivery claims sample set can accelerate the training speed of the pre-trained express delivery claims classification model.
[0102] Preferably, the preset classification model is the LightGBM model. The LightGBM model is characterized by its fast training speed compared to other models. Furthermore, the LightGBM model natively supports categorical features, eliminating the need for 0-1 encoding of categorical features. For example, whether a package is eligible for compensation or whether it is insured does not require further 0-1 encoding.
[0103] Specifically, the ratio of positive to negative samples in the first shipment claims sample set is less than that in the second shipment claims sample set. The ratio of positive to negative samples in the second shipment claims sample set is a preset ratio, which can be set according to specific circumstances. For example, the ratio of positive to negative samples in the first shipment claims sample set is 3:10000; the ratio in the second shipment claims sample set is 1:10.
[0104] In one specific embodiment, a first express delivery claim sample set is obtained, and negative samples in the first express delivery claim sample set are randomly removed until the ratio of positive samples to negative samples in the first express delivery claim sample set is a preset ratio of 1:10, thus obtaining a second express delivery claim sample set.
[0105] S202. Input multiple candidate negative samples from the first express claim sample set into the pre-trained express claim classification model to obtain the prediction results of each candidate negative sample.
[0106] In this embodiment, the prediction result of each candidate negative sample includes the prediction probability P that each candidate negative sample is predicted as a positive sample. predict .
[0107] S203. Determine the sampling weight information based on the prediction results of each candidate negative sample.
[0108] In this embodiment of the application, sampling weight information is determined based on the prediction results of each candidate negative sample, including S301-S302:
[0109] S301. Determine the confidence modulus of each candidate negative sample based on the predicted probability that each candidate negative sample is predicted to be a positive sample category and the true label of each candidate negative sample.
[0110] For example, the true label P of the candidate negative sample true All are 0. Of course, the true label P of the candidate negative sample is... true It can be defined according to the actual situation, such as -1, etc., but this application does not limit it.
[0111] In one specific embodiment, the multiple preset modulus intervals are multiple modulus intervals of equal length that are continuously distributed on the interval [0,1]. For example, N preset modulus intervals are N modulus intervals of equal length that are continuously distributed on the interval [0,1]. If N=2, then the N preset modulus intervals are [0,0.5) and [0.5,1].
[0112] Specifically, the confidence modulus of each candidate negative sample is determined based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample, including:
[0113] (1) Determine the initial confidence modulus of each candidate negative sample based on the difference between the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0114] The prediction probability P of each candidate negative sample being predicted as a positive sample predict The true label P of each candidate negative sample true The relationship between the initial confidence modulus G of each candidate negative sample and the initial confidence modulus G is shown in Equation (1):
[0115] G = |P predict -P true |,0≤G≤1 (1)
[0116] (2) Based on the initial confidence modulus of each candidate negative sample, the minimum confidence modulus among the initial confidence modulus of each candidate negative sample, and the maximum confidence modulus among the initial confidence modulus of each candidate negative sample, the initial confidence modulus of each candidate negative sample is normalized to obtain the confidence modulus of each candidate negative sample. The maximum confidence modulus G among all negative samples is calculated. max and minimum confidence modulus G min The initial confidence modulus G of each candidate negative sample, and the maximum confidence modulus G among the initial confidence modulus of each candidate negative sample. max The minimum confidence modulus G among the initial confidence moduli of each candidate negative sample. min The confidence modulus G of each candidate negative sample s Satisfying the relationship shown in formula (2),
[0117]
[0118] S302. Determine the sampling weight information based on the confidence modulus of each candidate negative sample.
[0119] Specifically, the sampling weight information includes the sampling weight coefficients for each preset modulus interval.
[0120] In a specific embodiment, the sampling weight information is determined based on the confidence modulus of each candidate negative sample, including:
[0121] (1) Obtain the number of sample distributions of multiple candidate negative samples in each preset modulus interval.
[0122] Specifically, calculate the number of negative samples in each preset modulus interval, and denote the number of negative samples in the nth preset modulus interval as num. n .
[0123] Of course, in other embodiments, the initial confidence modulus G of each candidate negative sample can also be determined as the confidence modulus G of each candidate negative sample. sThe minimum and maximum initial confidence modulus G of each candidate negative sample are used to form a modulus interval, and the modulus interval is divided equally to obtain each preset modulus interval.
[0124] (2) The sampling weight coefficient of each preset module length interval is determined based on the number of samples distributed in each preset module length interval. The sampling weight coefficient of the preset module length interval decreases as the number of samples distributed in the preset module length interval increases.
[0125] A larger confidence modulus indicates a greater discrepancy between the model's prediction and the sample's true label, meaning the sample is more difficult to classify. Our research has found that samples correctly classified by the model with high confidence levels provide far less information than samples with low confidence levels, or even misclassified samples. This is because the model needs to perform deep splits to correctly classify the latter when fitting samples. Therefore, samples with low confidence levels or those that are misclassified often play a more significant role in improving the model's training performance. Most negative samples in express mail samples provide limited information and can be easily and accurately classified as negative. The samples that are truly beneficial for model fitting are the small number of difficult-to-fit samples, namely those negative samples with large confidence moduli. Increasing the proportion of negative samples with large confidence moduli can improve the model's training performance.
[0126] Specifically, the number of samples distributed across each preset modulus interval, num. n And the sampling weight coefficient W for each preset modulus interval n The relationship is shown in formula (3).
[0127]
[0128] Alternatively, the proportion of the number of samples distributed within the preset modulus interval to the total number of candidate negative samples can be determined as the sampling weight coefficient W of the preset modulus interval. n You can set it according to the specific situation.
[0129] S204. Based on the sampling weight information, the first express claim sample set is weighted and sampled to obtain the third express claim sample set.
[0130] In this embodiment of the application, negative samples in the first express delivery claim sample set are sampled based on sampling weight information to obtain the third express delivery claim sample set, including:
[0131] (1) Sample multiple candidate negative samples in each preset module length interval based on the sampling weight coefficient of each preset module length interval to obtain multiple negative sample sampling sets.
[0132] Specifically, the proportion of negative samples in the negative sample set to the total number of negative samples in multiple negative sample sets is equal to the sampling weight coefficient of the preset modulus interval corresponding to the negative sample set. For example, if the sampling weight coefficient of the preset modulus interval A is 0.1, then the proportion of negative samples in the negative sample set obtained from the preset modulus interval A to the total number of samples is 0.1.
[0133] (2) The positive samples from multiple negative sample sets and the first express claim sample set are determined as the third express claim sample set.
[0134] The ratio of positive to negative samples in the third express delivery claims sample set is a preset ratio. For example, the preset ratio is 1:10. Preferably, the number of samples in the third express delivery claims sample set is equal to that in the second express delivery claims sample set.
[0135] S205. The pre-trained express claim classification model is trained based on the third express claim sample set to obtain the target express claim model.
[0136] In one specific embodiment, a pre-trained express delivery claim classification model is trained based on a third express delivery claim sample set to obtain a target express delivery claim model, including:
[0137] (1) The third express claim sample set is determined as the second express claim sample set and subjected to multiple iterations of weighted sampling to obtain the third express claim sample set after multiple iterations.
[0138] The third express delivery claim sample set is determined as the second express delivery claim sample set. S201-S204 is repeated to obtain the iterated third express delivery claim sample set. The newly generated third express delivery claim sample set is then determined as the second express delivery claim sample set again, and S201-S204 is repeated to obtain the iterated third express delivery claim sample set. By repeatedly executing S201-S204, multiple iterations of the third express delivery claim sample set can be obtained. Preferably, the number of iterative weighted sampling iterations is 3-5 times.
[0139] (2) The pre-trained express claim classification model is trained based on the third express claim sample set after multiple iterations to obtain the target express claim model.
[0140] In one specific embodiment, a pre-trained express delivery claim classification model is trained using a supervised learning method based on a third express delivery claim sample set after multiple iterations, to obtain the target express delivery claim model.
[0141] In another specific embodiment, a third express delivery claim sample set is obtained after each iteration of weighted sampling, resulting in multiple third express delivery claim sample sets. A pre-trained express delivery claim classification model is then trained on each of these multiple third express delivery claim sample sets to obtain multiple candidate express delivery claim classification models. These candidate models are then selected based on a preset model evaluation metric to obtain the target express delivery claim model. The preset model evaluation metric can be accuracy, which refers to the ratio of correctly predicted samples to the total number of predicted samples. The target express delivery claim model is the model with the highest accuracy among the multiple candidate express delivery claim classification models.
[0142] Furthermore, this application also provides a method for predicting express delivery claims, such as... Figure 4 As shown, the methods for predicting express delivery claims include:
[0143] S401. Obtain the express information of the express shipment to be predicted.
[0144] The express delivery information includes at least one of the following: the origin of the express delivery, the type of item being shipped, the shipping cost, whether the express delivery is insured, and the weight of the express delivery.
[0145] S402. Input the shipment information of the shipment to be predicted into the target shipment claim model to obtain the predicted probability that the shipment to be predicted is a positive sample category.
[0146] The target shipment claim model is any of the target shipment claim models described above.
[0147] To better implement the training method of the express delivery claims model in the embodiments of this application, based on the training method of the express delivery claims model, the embodiments of this application also provide a training device for the express delivery claims model, such as... Figure 5 As shown, the training device 500 for the express delivery claims model includes:
[0148] The acquisition unit 501 is used to acquire a first express delivery claim sample set and a pre-trained express delivery claim classification model, wherein the pre-trained express delivery claim classification model is a model trained on a preset classification model using a second express delivery claim sample set, and the second express delivery claim sample set is a subset of the first express delivery claim sample set.
[0149] The prediction unit 502 is used to input multiple candidate negative samples from the first express claim sample set into the pre-trained express claim classification model to obtain the prediction results of each candidate negative sample.
[0150] The weight calculation unit 503 is used to determine the sampling weight information based on the prediction results of each candidate negative sample;
[0151] The weighted sampling unit 504 is used to perform weighted sampling on the first express claim sample set based on the sampling weight information to obtain the third express claim sample set.
[0152] Model training unit 505 is used to train the pre-trained express claim classification model based on the third express claim sample set to obtain the target express claim model.
[0153] Optionally, the model training unit 505 is used for:
[0154] The third express delivery claim sample set is determined as the second express delivery claim sample set and subjected to multiple iterations of weighted sampling to obtain the third express delivery claim sample set after multiple iterations.
[0155] The pre-trained express claim classification model is trained based on the third express claim sample set after multiple iterations to obtain the target express claim model.
[0156] Optionally, the prediction result for each candidate negative sample includes the predicted probability that each candidate negative sample is predicted as a positive sample category;
[0157] Weight calculation unit 503 is used for:
[0158] The confidence modulus of each candidate negative sample is determined based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0159] The sampling weight information is determined based on the confidence modulus of each candidate negative sample.
[0160] Optionally, the sampling weight information includes the sampling weight coefficients for each preset modulus interval;
[0161] Weight calculation unit 503 is used for:
[0162] Obtain the number of sample distributions of multiple candidate negative samples in each preset modulus interval;
[0163] The sampling weight coefficient of each preset module length interval is determined based on the number of samples distributed in each preset module length interval. The sampling weight coefficient of the preset module length interval decreases as the number of samples distributed in the preset module length interval increases.
[0164] Optionally, the multiple preset modulus intervals are multiple modulus intervals of equal length that are continuously distributed in the interval [0,1].
[0165] Weight calculation unit 503 is used for:
[0166] The initial confidence modulus of each candidate negative sample is determined based on the difference between the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample.
[0167] The initial confidence modulus of each candidate negative sample is normalized based on the initial confidence modulus of each candidate negative sample, the minimum confidence modulus among the initial confidence modulus of each candidate negative sample, and the maximum confidence modulus among the initial confidence modulus of each candidate negative sample, to obtain the confidence modulus of each candidate negative sample.
[0168] Optionally, the weighted sampling unit 504 is used for:
[0169] Based on the sampling weight coefficients of each preset modulus interval, multiple candidate negative samples are sampled in each preset modulus interval to obtain multiple negative sample sampling sets;
[0170] The positive samples from multiple negative sample sets and the first express delivery claim sample set are determined as the third express delivery claim sample set.
[0171] Optionally, the acquisition unit 501 is used for:
[0172] Retrieve multiple shipment information and corresponding claim information for multiple historical shipments within a preset historical time period;
[0173] Based on the claims information corresponding to each express shipment, multiple express shipment information are labeled as positive sample type and negative sample type to obtain the first express shipment claims sample set.
[0174] To better implement the express delivery claim prediction method in the embodiments of this application, based on the express delivery claim prediction method, the embodiments of this application also provide an express delivery claim prediction device, such as... Figure 6 As shown, the express delivery claim prediction device 600 includes:
[0175] The acquisition unit 601 is used to acquire the express information of the express to be predicted;
[0176] The prediction unit 602 is used to input the express information of the express to be predicted into the target express claim model to obtain the predicted probability that the express to be predicted is a positive sample category, wherein the target express claim model is any of the above-mentioned target express claim models.
[0177] This application also provides a computer device that integrates a training device for any of the express delivery claims models provided in this application and / or a prediction device for express delivery claims. The computer device includes:
[0178] One or more processors;
[0179] Memory; and
[0180] One or more applications, wherein the applications are stored in memory and configured to be executed by a processor, comprising the steps of training a method for a delivery claims model or predicting a delivery claims method as described above.
[0181] like Figure 7 As shown, it illustrates a structural schematic diagram of the computer device involved in the embodiments of this application, specifically:
[0182] The computer device may include components such as a processor 701 with one or more processing cores, a memory 702 with one or more computer-readable storage media, a power supply 703, and an input unit 704. Those skilled in the art will understand that the computer device structure shown in the figures does not constitute a limitation on the computer device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:
[0183] Processor 701 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in memory 702, and by calling data stored in memory 702, thereby providing overall monitoring of the computer device. Optionally, processor 701 may include one or more processing cores; processor 701 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Preferably, processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and application programs, and the modem processor mainly handles wireless communication. It is understood that the aforementioned modem processor may not be integrated into processor 701.
[0184] The memory 702 can be used to store software programs and modules. The processor 701 executes various functional applications and data processing by running the software programs and modules stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 702 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 702 may also include a memory controller to provide the processor 701 with access to the memory 702.
[0185] The computer device also includes a power supply 703 that supplies power to the various components. Preferably, the power supply 703 can be logically connected to the processor 701 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 703 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.
[0186] The computer device may also include an input unit 704, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
[0187] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 701 in the computer device loads the executable files corresponding to the processes of one or more application programs into the memory 702 according to the following instructions, and the processor 701 runs the application programs stored in the memory 702 to realize various functions, as follows:
[0188] Obtain a first express delivery claim sample set and a pre-trained express delivery claim classification model. The pre-trained express delivery claim classification model is a model trained on a pre-defined classification model using a second express delivery claim sample set, which is a subset of the first express delivery claim sample set.
[0189] Multiple candidate negative samples from the first express claim sample set are input into the pre-trained express claim classification model to obtain the prediction results for each candidate negative sample.
[0190] The sampling weight information is determined based on the prediction results of each candidate negative sample;
[0191] The first express delivery claim sample set is weighted and sampled based on the sampling weight information to obtain the third express delivery claim sample set.
[0192] The pre-trained express claim classification model is trained based on the third express claim sample set to obtain the target express claim model;
[0193] or,
[0194] Obtain the shipment information of the shipment to be predicted;
[0195] Input the shipment information of the shipment to be predicted into the target shipment claim model to obtain the predicted probability that the shipment is a positive sample category.
[0196] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0197] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a disk, or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any of the training methods or prediction methods for express delivery claims provided in embodiments of this application. For example, the computer program loaded by the processor can execute the following steps:
[0198] Obtain a first express delivery claim sample set and a pre-trained express delivery claim classification model. The pre-trained express delivery claim classification model is a model trained on a pre-defined classification model using a second express delivery claim sample set, which is a subset of the first express delivery claim sample set.
[0199] Multiple candidate negative samples from the first express claim sample set are input into the pre-trained express claim classification model to obtain the prediction results for each candidate negative sample.
[0200] The sampling weight information is determined based on the prediction results of each candidate negative sample;
[0201] The first express delivery claim sample set is weighted and sampled based on the sampling weight information to obtain the third express delivery claim sample set.
[0202] The pre-trained express claim classification model is trained based on the third express claim sample set to obtain the target express claim model;
[0203] or,
[0204] Obtain the shipment information of the shipment to be predicted;
[0205] Input the shipment information of the shipment to be predicted into the target shipment claim model to obtain the predicted probability that the shipment is a positive sample category.
[0206] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.
[0207] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.
[0208] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0209] The above provides a detailed description of a training method, apparatus, computer device, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A training method for a parcel claim model, characterized in that, The training method includes: Obtain a first express delivery claim sample set and a pre-trained express delivery claim classification model, wherein the pre-trained express delivery claim classification model is a model trained on a preset classification model using a second express delivery claim sample set, the second express delivery claim sample set is a subset of the first express delivery claim sample set, and the ratio of positive samples to negative samples in the first express delivery claim sample set is less than the ratio of positive samples to negative samples in the second express delivery claim sample set; Multiple candidate negative samples from the first express delivery claim sample set are input into the pre-trained express delivery claim classification model to obtain the prediction results of each candidate negative sample. The sampling weight information is determined based on the prediction results of each candidate negative sample; Based on the sampling weight information, the first express delivery claim sample set is weighted and sampled to obtain multiple negative sample sample sets. The multiple negative sample sample sets and the positive samples in the first express delivery claim sample set are determined as the third express delivery claim sample set. The pre-trained express claim classification model is trained based on the third express claim sample set to obtain the target express claim model.
2. The training method according to claim 1, characterized in that, The step of training the pre-trained express delivery claim classification model based on the third express delivery claim sample set to obtain the target express delivery claim model includes: The third express delivery claim sample set is determined as the second express delivery claim sample set and subjected to multiple iterations of weighted sampling to obtain the third express delivery claim sample set after multiple iterations; The pre-trained express claim classification model is trained based on the third express claim sample set after multiple iterations to obtain the target express claim model.
3. The training method according to claim 1, characterized in that, The prediction results for each candidate negative sample include the predicted probability that each candidate negative sample is predicted as a positive sample category; The determination of sampling weight information based on the prediction results of each candidate negative sample includes: The confidence modulus of each candidate negative sample is determined based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample. The sampling weight information is determined based on the confidence modulus of each candidate negative sample.
4. The training method according to claim 3, characterized in that, The sampling weight information includes the sampling weight coefficients for each preset modulus interval; The determination of the sampling weight information based on the confidence modulus of each candidate negative sample includes: Obtain the number of sample distributions of the multiple candidate negative samples in each preset modulus interval; The sampling weight coefficient of each preset module length interval is determined based on the number of samples distributed in each preset module length interval, wherein the sampling weight coefficient of the preset module length interval decreases as the number of samples distributed in the preset module length interval increases.
5. The training method according to claim 4, characterized in that, The plurality of preset module length intervals are a plurality of equal-length module length intervals continuously distributed in the interval [0,1]. The determination of the confidence modulus of each candidate negative sample based on the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample includes: The initial confidence modulus of each candidate negative sample is determined based on the difference between the predicted probability of each candidate negative sample being predicted as a positive sample class and the true label of each candidate negative sample. The initial confidence modulus of each candidate negative sample is normalized based on the initial confidence modulus of each candidate negative sample, the minimum confidence modulus among the initial confidence modulus of each candidate negative sample, and the maximum confidence modulus among the initial confidence modulus of each candidate negative sample, to obtain the confidence modulus of each candidate negative sample.
6. The training method according to claim 4, characterized in that, The first express delivery claim sample set is weighted and sampled based on the sampling weight information to obtain multiple negative sample sample sets, including: Based on the sampling weight coefficients of each preset modulus interval, the candidate negative samples of the multiple candidate negative samples in each preset modulus interval are sampled to obtain multiple negative sample sampling sets.
7. The training method according to claim 1, characterized in that, The process of obtaining the first express delivery claim sample set and the pre-trained express delivery claim classification model includes: Retrieve multiple shipment information and corresponding claim information for multiple historical shipments within a preset historical time period; Based on the claim information corresponding to each express shipment, the multiple express shipment information are labeled as positive sample type and negative sample type to obtain the first express shipment claim sample set.
8. A method for predicting express delivery claims, characterized in that, The prediction methods for express delivery claims include: Obtain the shipment information of the shipment to be predicted; The shipment information of the shipment to be predicted is input into the target shipment claim model to obtain the predicted probability that the shipment to be predicted is a positive sample category, wherein the target shipment claim model is the target shipment claim model as described in any one of claims 1 to 7.
9. A training device for a parcel claim model, characterized in that, The training device for the express delivery claims model includes: The acquisition unit is used to acquire a first express delivery claim sample set and a pre-trained express delivery claim classification model, wherein the pre-trained express delivery claim classification model is a model trained on a preset classification model using a second express delivery claim sample set, the second express delivery claim sample set is a subset of the first express delivery claim sample set, and the ratio of positive samples to negative samples in the first express delivery claim sample set is less than the ratio of positive samples to negative samples in the second express delivery claim sample set; The prediction unit is used to input multiple candidate negative samples from the first express claim sample set into the pre-trained express claim classification model to obtain the prediction results of each candidate negative sample. The weight calculation unit is used to determine the sampling weight information based on the prediction results of each candidate negative sample; The weighted sampling unit is used to perform weighted sampling on the first express claim sample set based on the sampling weight information to obtain multiple negative sample sample sets, and to determine the multiple negative sample sample sets and the positive samples in the first express claim sample set as the third express claim sample set. The model training unit is used to train the pre-trained express claim classification model based on the third express claim sample set to obtain the target express claim model.
10. A predictive device for express delivery claims, characterized in that, The express delivery claim prediction device includes: The acquisition unit is used to acquire the express information of the express shipment to be predicted; A prediction unit is used to input the express information of the express to be predicted into the target express claim model to obtain the predicted probability that the express to be predicted is a positive sample category, wherein the target express claim model is the target express claim model as described in any one of claims 1 to 7.
11. A computer device, characterized in that, The computer device includes: One or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the training method of the express delivery claims model of any one of claims 1 to 7 or the prediction method of express delivery claims of claim 8.
12. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the steps in the training method of the express delivery claim model according to any one of claims 1 to 7 or the prediction method of express delivery claim according to claim 8.