Method and device for training conversion rate prediction model and method and device for predicting conversion rate

By training the conversion rate prediction model, using deep cross network and Bayesian smoothing algorithm, combined with visit record information, the problem of low clue conversion rate of logistics enterprises is solved, and the precise identification and visit priority of large order customers is achieved, and the return on investment is improved rate and work efficiency.

CN119066421APending Publication Date: 2024-12-03BEIJING JINGDONG YUANSHENG TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411150788.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Logistics companies have low lead conversion rates, resulting in low return on investment, and frequent contracts for small orders lead to losses. An effective method is needed to explore large order customers and improve conversion rates.

Method used

The training conversion rate prediction model is adopted, and the model parameters are adjusted using deep cross network and Bayesian smoothing algorithm to predict the conversion rate, and the model is adjusted according to the prediction results, and combined with the visit record information to optimize the model to achieve the Accurate identification and visits of high potential customers are preferred.

Benefits of technology

It improves the efficiency of accurate identification and visiting large order customers, reduces labor costs, improves work efficiency and conversion rates, and ensures an increase in return on investment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066421A_ABST
    Figure CN119066421A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a method and device for training a conversion rate prediction model and predicting a conversion rate. The specific embodiment of the method comprises the following steps: extracting features from customer data to obtain feature vectors; the feature vector is labeled with positive and negative samples according to a time period, the positive sample shows that the conversion rate is 1, and the negative sample shows that the conversion rate is 0; the feature vectors are input into a conversion rate prediction model, a predicted conversion rate is obtained, and the conversion rate prediction model comprises a deep cross network; and adjusting network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label. According to the embodiment, the accuracy of conversion rate prediction can be improved, so that a judgment standard is provided for selling and mining high-potential large-order customers, and accurate follow-up is carried out.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method and device for training a conversion rate prediction model and predicting a conversion rate. Background Art

[0002] For logistics companies, the investment cost from a lead to a contract is often quite high, including not only the sales staff's visits, the headquarters staff's management, the couriers' visits and follow-up, but also the operations colleagues' attention, etc. Therefore, if after a series of follow-up efforts, the merchant's contract volume is not high, resulting in low income, then the return on investment of the lead contract may be relatively low.

[0003] In addition, if the customers who sign contracts are small orders, it will lead to insufficient income and eventually losses, which is undesirable for a company. Therefore, it is particularly important to explore and accurately locate customers with large orders. Summary of the invention

[0004] The embodiments of the present disclosure provide a method and device for training a conversion rate prediction model.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for training a conversion rate prediction model, comprising: extracting features from customer data to obtain a feature vector; labeling the feature vector with positive and negative samples according to a time period, wherein a positive sample indicates a conversion rate of 1 and a negative sample indicates a conversion rate of 0; inputting the feature vector into a conversion rate prediction model to obtain a predicted conversion rate, wherein the conversion rate prediction model includes a deep cross network; and adjusting the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label.

[0006] In some embodiments, before adjusting the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label, the method further includes: adjusting the predicted conversion rate by using a Bayesian smoothing algorithm.

[0007] In some embodiments, extracting features from customer data to obtain feature vectors includes: extracting features from at least one of the following data of the customer to obtain feature vectors: portrait information, industrial and commercial registration information, order volume and package volume in different time dimensions.

[0008] In some embodiments, extracting features from customer data to obtain feature vectors includes: segmenting attribute information in the customer data to obtain attribute keywords; and converting attribute keywords that appear more than a predetermined threshold in the sample set into feature vectors.

[0009] In some embodiments, extracting features from customer data to obtain feature vectors includes: counting whether predetermined types of information appear in the customer data to obtain statistical results; and performing multi-hot encoding on the statistical results to generate feature vectors.

[0010] In some embodiments, extracting features from customer data to obtain feature vectors includes: using a pipeline mechanism to splice the extracted feature vectors into a total feature vector.

[0011] In some embodiments, the method further includes: receiving visit record information to characterize the reasons why customers failed to convert; extracting a failure cause feature vector from the reasons; concatenating the failure cause feature vector and the feature vector of customers who failed to convert as negative samples, and retraining the conversion rate prediction model.

[0012] In a second aspect, an embodiment of the present disclosure provides a method for predicting conversion rate, comprising: extracting a feature vector of each candidate customer from relevant information of the candidate customer; inputting the feature vector of each candidate customer into a conversion rate prediction model trained according to the method described in any one of the first aspects, and outputting the conversion rate of each candidate customer; and outputting a list of candidate customers in descending order of conversion rate.

[0013] In a third aspect, an embodiment of the present disclosure provides a device for training a conversion rate prediction model, comprising: an extraction unit, configured to extract features from customer data to obtain a feature vector; a labeling unit, configured to label the feature vector with positive and negative samples according to a time period, wherein a positive sample indicates a conversion rate of 1 and a negative sample indicates a conversion rate of 0; a prediction unit, configured to input the feature vector into a conversion rate prediction model to obtain a predicted conversion rate, wherein the conversion rate prediction model includes a deep cross network; an adjustment unit, configured to adjust the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label.

[0014] In some embodiments, the apparatus further comprises a smoothing unit configured to: before adjusting the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label, adjust the predicted conversion rate by using a Bayesian smoothing algorithm.

[0015] In some embodiments, the extraction unit is further configured to extract features from at least one of the following data of the customer to obtain a feature vector: portrait information, business registration information, order volume and package volume in different time dimensions.

[0016] In some embodiments, the extraction unit is further configured to: segment the attribute information in the customer data to obtain attribute keywords; and convert the attribute keywords that appear more than a predetermined threshold in the sample set into feature vectors.

[0017] In some embodiments, the extraction unit is further configured to: count whether a predetermined type of information appears in the customer data to obtain a statistical result; and perform multi-hot encoding on the statistical result to generate a feature vector.

[0018] In some embodiments, the extraction unit is further configured to: utilize a pipeline mechanism to concatenate the extracted feature vectors into a total feature vector.

[0019] In some embodiments, the device also includes a feedback unit configured to: receive visit record information for characterizing the reasons why customers fail to convert successfully; extract a failure cause feature vector from the reasons; and concatenate the failure cause feature vector and the feature vector of customers who fail to convert successfully as negative samples to retrain the conversion rate prediction model.

[0020] In a fourth aspect, an embodiment of the present disclosure provides a device for predicting conversion rate, comprising: an extraction unit, configured to extract a feature vector of each candidate customer from relevant information of the candidate customer; a prediction unit, configured to input the feature vector of each candidate customer into a conversion rate prediction model trained according to the method described in any one of the first aspects, and output the conversion rate of each candidate customer; and an output unit, configured to output a list of candidate customers in descending order of conversion rate.

[0021] In a fifth aspect, an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device on which one or more computer programs are stored, and when the one or more computer programs are executed by the one or more processors, the one or more processors implement a method as described in any one of the first aspect or the second aspect.

[0022] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method as described in any one of the first aspect or the second aspect is implemented.

[0023] In a seventh aspect, an embodiment of the present disclosure provides a computer program product, comprising a computer program, wherein the computer program implements the method as described in any one of the first aspect or the second aspect when executed by a processor.

[0024] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0026] Figure 1 is an exemplary system architecture diagram in which an embodiment of the present disclosure may be applied;

[0027] Figure 2 is a flowchart of an embodiment of a method for training a conversion rate prediction model according to the present disclosure;

[0028] Figure 3a-3b is a schematic diagram of an application scenario of the method for training a conversion rate prediction model according to the present disclosure;

[0029] Figure 4 is a flow chart of an embodiment of a method for predicting conversion rate according to the present disclosure;

[0030] Figure 5 is a structural schematic diagram of an embodiment of a device for training a conversion rate prediction model according to the present disclosure;

[0031] Figure 6 is a schematic structural diagram of an embodiment of a device for predicting conversion rate according to the present disclosure;

[0032] Figure 7 It is a structural diagram of a computer system of an electronic device suitable for implementing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0033] The present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0034] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0035] Figure 1 An exemplary system architecture 100 is shown to which the method for training a conversion rate prediction model, the apparatus for training a conversion rate prediction model, the method for predicting a conversion rate, or the apparatus for predicting a conversion rate according to the embodiments of the present disclosure can be applied.

[0036] like Figure 1 As shown, the system architecture 100 may include terminals 101, 102, a network 103, a database server 104, and a server 105. The network 103 is used to provide a medium for communication links between the terminals 101, 102, the database server 104, and the server 105. The network 103 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0037] The user 110 can use the terminals 101 and 102 to interact with the server 105 through the network 103 to receive or send messages, etc. Various client applications can be installed on the terminals 101 and 102, such as model training applications, conversion rate prediction applications, shopping applications, payment applications, web browsers, and instant messaging tools.

[0038] The terminals 101 and 102 here can be hardware or software. When the terminals 101 and 102 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, Dynamic Image Experts Compression Standard Audio Layer 3), laptops and desktop computers, etc. When the terminals 101 and 102 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, to provide distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0039] The database server 104 may be a database server that provides various services. For example, a sample set may be stored in the database server. The sample set includes a large number of samples. The sample may include a feature vector and a label converted according to customer data, and the label is used to characterize the conversion rate of large-order customers. In this way, the user 110 can also select samples from the sample set stored in the database server 104 through the terminals 101 and 102.

[0040] The server 105 may also be a server that provides various services, such as a background server that supports various applications displayed on the terminals 101 and 102. The background server may train the initial model using samples in the sample set sent by the terminals 101 and 102, and may send the training results (such as the generated conversion rate prediction model) to the terminals 101 and 102. In this way, the user may use the generated conversion rate prediction model to perform conversion rate prediction.

[0041] The database server 104 and server 105 here can also be hardware or software. When they are hardware, they can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When they are software, they can be implemented as multiple software or software modules (for example, used to provide distributed services), or as a single software or software module. No specific limitation is made here. The database server 104 and server 105 can also be servers of a distributed system, or servers combined with blockchain. The database server 104 and server 105 can also be cloud servers, or intelligent cloud computing servers or intelligent cloud hosts with artificial intelligence technology.

[0042] It should be noted that the method for training a conversion rate prediction model or the method for predicting a conversion rate provided in the embodiments of the present disclosure is generally executed by the server 105. Accordingly, the device for training a conversion rate prediction model or the device for predicting a conversion rate is generally also disposed in the server 105.

[0043] It should be pointed out that, in the case where the server 105 can implement the related functions of the database server 104 , the database server 104 may not be provided in the system architecture 100 .

[0044] It should be understood that Figure 1 The number of terminals, networks, database servers and servers in the embodiment is only for illustration. Any number of terminals, networks, database servers and servers may be provided according to the implementation requirements.

[0045] Continue to see Figure 2 , which shows a process 200 of an embodiment of a method for training a conversion rate prediction model according to the present disclosure. The method for training a conversion rate prediction model may include the following steps:

[0046] Step 201: extract features from customer data to obtain feature vectors.

[0047] In this embodiment, the execution body of the method for training the conversion rate prediction model (for example Figure 1 The server shown in FIG. 1 can obtain customer data through a wired connection or a wireless connection. The customer data may include customer attribute information and order information. The customer data may be segmented into words, and then the keywords may be converted into feature vectors through a word vector conversion tool. If the feature is a numerical type, it is directly used as a feature vector.

[0048] In some optional implementations of this embodiment, the customer data may include at least one of the following: portrait information, business registration information, and order volume and package volume in different time dimensions.

[0049] The portrait information contains a summary of information of various dimensions of merchants on the logistics side, including the merchants' basic information, business conditions, business capabilities, behavioral analysis, etc. The dimensional data of industrial and commercial registration information can be queried using third-party tools, and may include legal representatives, business scope, registered capital, paid-in capital, number of insured persons, etc.

[0050] The order volume and package volume in different time dimensions can be, for example, the order volume information can be traced back to the recent weekly and monthly month-on-month information, which can be used as a supplement to the model feature information to help fully understand the recent order volume information dimension, thereby playing a positive role in mining large order customer information. The following is a derivation of the month-on-month information supplemented with the order volume and package volume features.

[0051] The order volume and package volume include recent total volume characteristics, long-term total volume characteristics, and year-on-year characteristics in different time dimensions. The year-on-year time dimension is mainly based on daily granularity, weekly granularity, and monthly granularity. The detailed calculation method is as follows:

[0052] 1) Based on the daily granularity, the calculation method for the current day's daily order volume compared to yesterday is as follows

[0053]

[0054] 2) Based on the weekly granularity, if the current day is the kth day of this week, the average daily order volume of this week up to the current day is calculated as follows:

[0055]

[0056] Among them, n 01 to n 0k They are the number of orders from the 1st day to the kth day of this week. The average number of orders per day up to the same day last week is calculated as follows:

[0057]

[0058] Among them, n 11 to n 1k They are the number of orders from the 1st day to the kth day of last week. The calculation method of the average daily order volume of this week up to today compared with the average daily order volume of the week up to the same day last week is:

[0059]

[0060] As shown in Table 1, based on the average daily order volume, a month-on-month comparison of the average daily orders from the 1st to the 4th week is established as of this week, based on the month-on-month comparison of the total order volume of this week compared with the total order volume of previous weeks, and the year-on-year comparison of the average daily order volume from this week to that day compared with the same week last year.

[0061] 3) Based on the monthly granularity, the calculation method is similar to the weekly granularity. It includes the month-on-month characteristics calculated as of a certain day of a certain month compared with the same day of the previous month, the month-on-month comparison of the total volume of a certain month relative to the total volume of the previous month, the year-on-year comparison of the daily average volume as of a certain day of this month relative to the same day of the same month last year, and the year-on-year comparison of the daily order volume of the current day relative to the same day of the previous month. Among them, the year-on-year calculation method for the daily order volume of the current day relative to the same day of the previous month is

[0062]

[0063] The detailed features are as follows, including: recent order volume distribution value, long-term order volume distribution value, recent weekly year-on-year distribution value, recent monthly year-on-year distribution value, and other features.

[0064]

[0065] Table 1 Recent single volume distribution values

[0066]

[0067] Table 2 Long-term single quantity distribution value

[0068]

[0069] Table 3 Recent Weekly and Monthly Distribution Values

[0070]

[0071] Table 4 Recent Monthly and Monthly Distribution Values

[0072] Optionally, customer data can be preprocessed before feature extraction. For example, outliers can be eliminated using the 3sigma method (rule of thumb), and null values ​​can be filled with medians and averages, and duplicate data in merchant dimensions can be deduplicated. These data operations can improve the availability of data and features.

[0073] In some optional implementations of this embodiment, the feature extraction from the customer data to obtain the feature vector includes: segmenting the attribute information in the customer data to obtain the attribute keywords; converting the attribute keywords whose occurrence times in the sample set are greater than the predetermined threshold into the feature vector. The first-level information of the merchant in the mall, the main category information, the merchant industry information, etc. can be segmented into Chinese words, and divided into more fine-grained range attributes to form new features. For example, the main category information of a certain merchant is: household daily necessities / home decoration building materials / household cleaning / paper products / kitchenware. After Chinese word segmentation, it is [furniture daily necessities, home decoration building materials, household cleaning, paper products, kitchenware]. The features after Chinese word segmentation show the category attribute information of the merchant in a more detailed granularity. After word segmentation, it is helpful for the selection of keywords. In the subsequent model vectorization formation process, only the keywords whose occurrence times in the full text are greater than the predetermined threshold (for example, 50 times) will be retained, and the rest will be discarded, and more attention will be paid to the key information and fine-grained information of the category.

[0074] In some optional implementations of this embodiment, extracting features from customer data to obtain feature vectors includes: counting whether predetermined types of information appear in the customer data to obtain statistical results; and performing multi-hot encoding on the statistical results to generate feature vectors.

[0075] By counting the values ​​of other discrete features of the product, we count whether each feature data appears as the processed feature. Among them, the first-level information of the mall, the main category information, and the merchant industry information contain multiple feature words, which are multi-hot features. Multi-hot features are different from one-hot features and can describe the multi-dimensional values ​​of a sample attribute in more detail.

[0076] For example, a multi-hot feature corresponds to a sample with multiple positions having values ​​after encoding, such as [0,1,0,0,0,1]. A one-hot feature only has a value at one position, such as

[0077] [0,0,0,0,0,1]. Multi-hot can better describe the relationship between multiple values ​​in a sample.

[0078] In some optional implementations of this embodiment, extracting features from customer data to obtain feature vectors includes: using a pipeline mechanism to splice the extracted feature vectors into a total feature vector. The pipeline mechanism is used to serialize all feature engineering processing mechanisms and finally summarize them into a total feature attribute. The purpose of pipeline mechanism processing is to reduce coupling, increase flexibility, and improve performance.

[0079] Step 202: label the feature vector with positive and negative samples according to the time period.

[0080] In this embodiment, a positive sample represents a conversion rate of 1, and a negative sample represents a conversion rate of 0.

[0081] For large logistics companies, they often have various types of merchants. For example, merchants have a certain periodicity. For example, seasonal merchants place a large number of orders in the peak season and a small number of orders in the off-season. Therefore, the labels of customers in each stage may be different. For example, last month was a large order, which is a positive sample. Next month, it may not meet the standard of a large order, which is a negative sample. In this way, the data is more realistic, and the features are also taken according to the corresponding time period. The actual number of orders and other features. It is generally believed that customers with an estimated order volume of more than 300 orders are large-order customers, and sales will focus on following up. In addition, sales will judge large-order customers based on some industries or sales experience.

[0082] Positive samples: Mark historical offline logistics large-order contract customers as positive samples. Large-order customers from July 2021 to July 2022 can be selected as positive samples.

[0083] Negative samples: Historical offline logistics customers who did not sign large orders are sampled and marked as negative samples.

[0084] After defining the labeling method for positive and negative examples, the positive and negative samples are determined based on the data within a period.

[0085] Step 203: input the feature vector into the conversion rate prediction model to obtain the predicted conversion rate.

[0086] In this embodiment, the conversion rate prediction model includes a deep cross-network (DCN). After inputting the relevant feature data of the customer, the large-order customer prediction model will give the probability value of whether the customer can be converted into a large-order customer. Sales personnel can give priority to visiting merchants with higher predicted probabilities, thereby saving manpower and realizing AI's accurate selection of large-order customers with high potential. The DCN model can be abstracted into a 3-layer network, including an input layer, a hidden layer, and an output layer. The specific structure is as follows Figure 3a shown.

[0087] The features input to the input layer include categorical features (sparse features) and numerical features (dense features). After embedding, the categorical features are directly concatenated with the numerical features as the input of the model. In the hidden layer, DCN uses a cross network to increase the interaction between features and uses multiple cross layers to cross the input features. The cross network minimizes the parameters of the model while ensuring feature interaction. In addition, the DCN network composed of multiple cross networks automatically crosses features, avoiding more reliance on manual feature crossing. The cross layer architecture is as follows: Figure 3b shown.

[0088] Through the cross network and deep network, multi-dimensional detection and intrinsic mining of merchant data can be achieved; the output layer performs binary classification prediction.

[0089] The DCN neural network uses nonlinear activation and feature crossover. The nonlinear activation function mainly uses the RELU function, which can transform the original linear features into nonlinear features after the activation function, realize the nonlinear change of features, and observe the hidden dimensional relationship of features from multiple perspectives.

[0090] The input data for model training is divided into in-set and out-set data according to the time dimension. The in-set data is divided into training set and test set according to 7:3. The training set is used to train the model data, and the test set is used to evaluate the model effect. The training set and test set of the in-set data belong to the same data distribution, and there is a certain correlation between the features, which is used to judge the model effect. The out-set data and the in-set data are completely isolated, and the actual effect can be predicted and evaluated.

[0091] Step 204: adjust the network parameters of the conversion rate prediction model according to the predicted conversion rate and the difference between the labels.

[0092] In this embodiment, the predicted conversion rate is the probability of a customer being converted into a large-order customer, and the label is 0 or 1. The network parameters of the conversion rate prediction model are adjusted according to the difference between the predicted conversion rate and the label, and supervised training is performed until the training completion condition of the model is met.

[0093] In some optional implementations of this embodiment, the method further includes: performing parameter tuning and feature tuning on the conversion rate prediction model.

[0094] Parameter tuning may include the maximum number of iterations, the selection of linear activation function, the selection of convergence method, the setting of convergence threshold, etc.

[0095] The feature optimization is aimed at the mining rate of the most influential large-order customers (i.e. the conversion rate of customers into large-order customers). The calculation method of the mining rate of large-order customers is:

[0096] Large-order customer mining rate = number of large-order customers signed / number of customers visited

[0097] If all the customers visited by a salesperson with a small number of customers are converted, the confidence of the model judgment will be reduced. Therefore, Bayesian smoothing is introduced to adjust the mining rate of large single customers. The Bayesian smoothing calculation method is to count the overall mean and variance of all samples, and then use the overall mean and variance as the prior probability, and then adjust the conversion rate of each piece of information, thereby increasing the credibility of the results issued.

[0098] In some optional implementations of this embodiment, the method further includes: evaluating the conversion rate prediction model, which mainly includes evaluation indicators of the model and evaluation indicators of the business.

[0099] 1) Model evaluation indicators:

[0100] The main indicators include the model's AUC index and the precision, recall and F1 value of positive and negative samples. The evaluation method is explained using positive samples as an example:

[0101] Precision = data predicted as positive samples and actually positive samples / all data predicted as positive samples.

[0102] Recall rate = data predicted to be positive samples and actually positive samples / all positive sample data.

[0103] F1 value = 2*precision*recall / (recall+precision).

[0104] The calculation method of negative sample and positive sample evaluation indicators is the same, except that the statistical caliber is changed from positive samples to negative samples.

[0105] 2) Business evaluation indicators:

[0106] It mainly mines conversion rate prediction models for large-order customers and collects conversion rates of large-order customers visited by sales by region.

[0107] The calculation method of large-order customer mining rate is:

[0108] Large-order customer mining rate = number of large-order customers signed / number of customers visited

[0109] The final conversion rate of large-order customers obtained in this way is the actual conversion rate of each high-potential list.

[0110] In some optional implementations of this embodiment, the method also includes: receiving visit record information to characterize the reasons why customers failed to convert; extracting a failure cause feature vector from the reasons; concatenating the failure cause feature vector and the feature vector of the unsuccessful converted customer as negative samples, and retraining the conversion rate prediction model.

[0111] Feedback on regional sales visit records includes users who have converted and the reasons why merchants did not convert as predicted by the model. Combined with the reasons for the visits, this provides other dimensional support for the next optimization of the model, thereby continuously optimizing the model accuracy.

[0112] The method provided by the above embodiment of the present disclosure selects the corresponding merchant information for the characteristic attributes of each dimension of the merchant through the feature engineering method based on big data, and establishes a targeted deep learning model. Based on the model, it evaluates whether the merchant has the potential to become a large-order customer, and then explores the offline large-order customer merchants of logistics, so that sales personnel can make accurate visits, thereby reducing labor costs and improving work efficiency.

[0113] Further references Figure 4 , which shows a process 400 of an embodiment of a method for predicting conversion rate. The process 400 of the method for predicting conversion rate includes the following steps:

[0114] Step 401: extract a feature vector of each candidate customer from the relevant information of the candidate customer.

[0115] In this embodiment, it is possible to segment words from information such as portrait information, business registration information, order volume and package volume in different time dimensions, and then extract the feature vector of the candidate customer by querying the embedding vector corresponding to the keyword.

[0116] Step 402: Input the feature vector of each candidate customer into the conversion rate prediction model, and output the conversion rate of each candidate customer.

[0117] In this embodiment, the conversion rate prediction model is a DCN trained by the method described in steps 201-204.

[0118] Step 403: Output a list of candidate customers in descending order of conversion rate.

[0119] In this embodiment, the conversion rate prediction model will sort the samples predicted to be positive samples, i.e., high-potential customers, by probability, and issue a list of merchants with relatively high probabilities according to the visiting capabilities of each region. Regional sales will visit high-potential customers and recommend conversions based on various factors, effectively saving manpower while improving the accuracy of merchant conversions.

[0120] This solution uses DCN to establish a binary classification model. After inputting the customer's relevant feature data, the large-order customer prediction model will give the probability value of whether the customer can be converted into a large-order customer. Sales personnel can give priority to visiting merchants with higher predicted probabilities, thereby saving manpower and realizing AI's accurate selection of large-order customers with high potential.

[0121] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for training a conversion rate prediction model. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0122] like Figure 5 As shown, the apparatus 500 for training a conversion rate prediction model in this embodiment includes: an extraction unit 501, a labeling unit 502, a prediction unit 503, and an adjustment unit 504. The extraction unit 501 is configured to extract features from customer data to obtain a feature vector; the labeling unit 502 is configured to label the feature vector with positive and negative samples according to a time period, wherein a positive sample indicates that the conversion rate is 1, and a negative sample indicates that the conversion rate is 0; the prediction unit 503 is configured to input the feature vector into a conversion rate prediction model to obtain a predicted conversion rate, wherein the conversion rate prediction model includes a deep cross network; the adjustment unit 504 is configured to adjust the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label.

[0123] In this embodiment, the specific processing of the extraction unit 501, the marking unit 502, the prediction unit 503 and the adjustment unit 504 of the device 500 for training the conversion rate prediction model can be referred to. Figure 2 Corresponding to step 201, step 202, step 203, and step 204 in the embodiment.

[0124] In some optional implementations of this embodiment, the device 500 also includes a smoothing unit (not shown in the drawings), which is configured to: adjust the predicted conversion rate through a Bayesian smoothing algorithm before adjusting the network parameters of the conversion rate prediction model according to the difference between the predicted conversion rate and the label.

[0125] In some optional implementations of this embodiment, the extraction unit 501 is further configured to extract features from at least one of the following data of the customer to obtain a feature vector: portrait information, industrial and commercial registration information, order volume and package volume in different time dimensions.

[0126] In some optional implementations of this embodiment, the extraction unit 501 is further configured to: segment the attribute information in the customer data to obtain attribute keywords; and convert the attribute keywords that appear more than a predetermined threshold in the sample set into feature vectors.

[0127] In some optional implementations of this embodiment, the extraction unit 501 is further configured to: count whether a predetermined type of information appears in the customer data to obtain a statistical result; and perform multi-hot encoding on the statistical result to generate a feature vector.

[0128] In some optional implementations of this embodiment, the extraction unit 501 is further configured to: use a pipeline mechanism to splice the extracted feature vectors into a total feature vector.

[0129] In some optional implementations of the present embodiment, the device 500 also includes a feedback unit (not shown in the drawings), which is configured to: receive visit record information for characterizing the reasons why customers are not successfully converted; extract a failure cause feature vector from the reasons; and concatenate the failure cause feature vector and the feature vector of the unsuccessfully converted customer as a negative sample to retrain the conversion rate prediction model.

[0130] Further references Figure 6 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for predicting conversion rate. Figure 4 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0131] like Figure 6 As shown, the device 600 for predicting conversion rate in this embodiment includes: an extraction unit 601, a prediction unit 602 and an output unit 603. The extraction unit 601 is configured to extract a feature vector of each candidate customer from the relevant information of the candidate customer; the prediction unit 602 is configured to input the feature vector of each candidate customer into the conversion rate prediction model trained by the device 500, and output the conversion rate of each candidate customer; the output unit 603 is configured to output a list of candidate customers in descending order of conversion rate.

[0132] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0133] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0134] An electronic device comprises: one or more processors; a storage device on which one or more computer programs are stored, and when the one or more computer programs are executed by the one or more processors, the one or more processors implement the method described in process 200 or 400.

[0135] A computer-readable medium stores a computer program, wherein the computer program implements the method described in process 200 or 400 when executed by a processor.

[0136] Figure 7A schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0137] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0138] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0139] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as a road area planning method. For example, in some embodiments, the road area planning method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the road area planning method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the road area planning method in any other appropriate manner (e.g., by means of firmware).

[0140] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0141] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0142] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0143] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0144] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0145] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server may be a server of a distributed system, or a server combined with a blockchain. The server may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. The server may be a server of a distributed system, or a server combined with a blockchain. The server may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0146] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0147] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for training a conversion rate prediction model, comprising: Extract features from customer data to obtain feature vectors; Labeling the feature vectors as positive and negative samples according to the time period, wherein a positive sample indicates a conversion rate of 1, and a negative sample indicates a conversion rate of 0; Inputting the feature vector into a conversion rate prediction model to obtain a predicted conversion rate, wherein the conversion rate prediction model includes a deep cross network; The network parameters of the conversion rate prediction model are adjusted according to the difference between the predicted conversion rate and the label.

2. The method according to claim 1, wherein: Before adjusting the network parameters of the conversion rate prediction model according to the predicted conversion rate and the difference between the labels, the method further includes: The predicted conversion rate is adjusted by a Bayesian smoothing algorithm.

3. The method according to claim 1, wherein: The step of extracting features from customer data to obtain a feature vector includes: Extract features from at least one of the following data of the customer to obtain a feature vector: Portrait information, business registration information, order volume and package volume in different time dimensions.

4. The method according to claim 1, wherein: The step of extracting features from customer data to obtain a feature vector includes: Segment the attribute information in the customer data to obtain attribute keywords; The attribute keywords whose occurrence times in the sample set are greater than a predetermined threshold are converted into feature vectors.

5. The method according to claim 1, wherein: The step of extracting features from customer data to obtain a feature vector includes: Count whether the predetermined type of information appears in the customer data and obtain the statistical results; The statistical results are multi-hot encoded to generate a feature vector.

6. The method according to claim 1, wherein: The feature extraction from the customer data, Get the feature vector, including: The extracted feature vectors are concatenated into a total feature vector using a pipeline mechanism.

7. The method according to claim 1, wherein: The method further comprises: Receive visit record information to indicate the reasons why customers were not successfully converted; extracting a failure cause feature vector from the causes; The failure reason feature vector and the feature vector of the customer who has not been successfully converted are concatenated as negative samples to retrain the conversion rate prediction model.

8. A method for predicting conversion rate, comprising: Extracting a feature vector of each candidate customer from the relevant information of the candidate customer; Inputting the feature vector of each candidate customer into the conversion rate prediction model trained according to any one of the methods of claims 1 to 7, and outputting the conversion rate of each candidate customer; Output a list of candidate customers in order of conversion rate from high to low.

9. A device for training a conversion rate prediction model, comprising: An extraction unit is configured to extract features from the customer data to obtain a feature vector; a labeling unit configured to label the feature vector with positive and negative samples according to a time period, wherein a positive sample indicates a conversion rate of 1, and a negative sample indicates a conversion rate of 0; A prediction unit, configured to input the feature vector into a conversion rate prediction model to obtain a predicted conversion rate, wherein the conversion rate prediction model includes a deep cross network; An adjustment unit is configured to adjust a network parameter of the conversion rate prediction model according to a difference between the predicted conversion rate and the label.

10. A device for predicting conversion rate, comprising: An extraction unit is configured to extract a feature vector of each candidate customer from the relevant information of the candidate customer; A prediction unit, configured to input the feature vector of each candidate customer into a conversion rate prediction model trained by the method according to any one of claims 1 to 7, and output the conversion rate of each candidate customer; The output unit is configured to output a list of candidate customers in descending order of conversion rate.

11. An electronic device, comprising: one or more processors; a storage device having one or more computer programs stored thereon, When the one or more computer programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

12. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

13. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 8.

Citation Information

Patent Citations

  • Search advertisement conversion rate prediction method and device for shopping section

    CN111352976A

  • Loan prediction method and device, electronic equipment and computer readable storage medium

    CN112785095A

  • Customer loss prediction method and device, storage medium and electronic equipment

    CN114022222A

  • Bank consumption loan customer demand mining method and device and storage medium

    CN115731030A