Potential customer recommendation method and device and electronic equipment
By building a potential customer recommendation model based on encoder-decoder and convolutional gated recurrent network, using diversified transaction and evaluation index data, the problems of low accuracy and high cost of potential customer recommendation in the existing technology are solved, and the problem of low accuracy and high cost of potential customer recommendation in the existing technology is realized. More accurate and efficient customer recommendations.
Patent Information
- Application Number
- CN202410952439.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-12-03
AI Technical Summary
The existing potential customer recommendation method only considers the company's trading indicators based on a certain product, and ignores the transaction time correlation, resulting in low recommendation accuracy and high cost.
By building a potential customer recommendation model, using transaction indicator data and evaluation indicator data, combining the encoder-decoder structure and convolutional gating recurrent network, diversified data processing and training are carried out to screen out potential customers and reduce conversion costs.
Improves the accuracy and accuracy of potential customer recommendations, reduces marketing and conversion costs, and is suitable for a variety of transaction scenarios.
Smart Images

Figure CN119067729A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of customer recommendation, and in particular to a method, device and electronic equipment for recommending potential customers. Background Art
[0002] In any field where transactions occur, especially in industries with data accumulation such as e-commerce, retail, finance, and life services, identifying potential customers can further reduce marketing costs, shorten marketing cycles, and increase conversion rates. Potential customer recommendation refers to recommending suppliers with whom no transactions have been conducted to buyers and recommending buyers with whom no transactions have been conducted to suppliers for a certain type of product.
[0003] Existing potential customer recommendation methods are generally based on the company's transaction indicators for a certain type of product, using machine learning models to recommend potential customers to the company, and then convert potential customers into new customers under certain conditions, thereby increasing revenue and creating value for the company. However, the existing potential customer recommendation methods only consider the company's transaction indicators based on a certain product, the indicators are relatively single, and the time correlation of existing transactions is not considered. There are great limitations, and the accuracy of potential customer recommendation using existing potential customer recommendation methods is not high. Summary of the invention
[0004] In view of the above problems, the present invention provides a potential customer recommendation method, device and electronic device, which can accurately recommend potential customers and reduce marketing costs and potential customer conversion costs.
[0005] The present invention provides a potential customer recommendation method, the method comprising:
[0006] Determine the pre-recommended customer group based on the transaction needs of the target transaction subject; wherein the transaction needs include buying and selling needs and the target transaction product category;
[0007] Obtain input data corresponding to each pre-recommended customer in the pre-recommended customer group, and input the input data corresponding to each pre-recommended customer into a potential customer recommendation model corresponding to the buying and selling needs, so as to obtain a recommendation degree corresponding to each pre-recommended customer; wherein the input data includes transaction index data and evaluation index data, and the potential customer recommendation model is obtained by a preset model training strategy;
[0008] All potential customers are screened out from the pre-recommended customer group based on the recommendation degree, and all potential customers are recommended to the target transaction entity.
[0009] Furthermore, the preset model training strategy includes:
[0010] Determine multiple training data based on historical transaction data of multiple categories of products and evaluation score data of the recommended party among the two parties of the transaction; wherein the training data includes transaction index data and evaluation index data, and the recommended party is determined according to the preset recommendation purpose of the potential customer recommendation model;
[0011] Determine the label corresponding to each piece of training data respectively; wherein the label includes a recommendation degree label;
[0012] Each piece of training data and a label corresponding to the training data are respectively used as a training sample to construct a sample data set;
[0013] According to the sample data set, the preset network model is trained through the N-fold cross-validation method to obtain the potential customer recommendation model.
[0014] Furthermore, multiple training data are determined based on historical transaction data of multiple categories of products and evaluation score data of the recommended party among the two transaction parties, including:
[0015] Aggregate historical transaction data of multiple categories of products according to a preset strategy to obtain multiple initial transaction data sets; wherein each initial transaction data set includes all historical transaction data of the same transaction parties based on the same category of products;
[0016] Preprocessing the data in each initial transaction data set respectively to obtain transaction indicator data corresponding to the initial transaction data set;
[0017] The evaluation score data of the recommended party among the two transaction parties corresponding to each initial transaction data set is used as the evaluation index data corresponding to the initial transaction data set;
[0018] The transaction index data and the evaluation index data corresponding to each initial transaction data set are respectively used as one piece of training data to obtain multiple pieces of training data.
[0019] Furthermore, if the preset recommendation purpose of the potential customer recommendation model is to recommend a supplier to the purchaser, the recommended party is the supplier among the two parties of the transaction;
[0020] If the preset recommendation purpose of the potential customer recommendation model is to recommend a purchaser to a supplier, the recommended party is the purchaser among the two transaction parties.
[0021] Furthermore, the preset network model includes:
[0022] An encoder for extracting feature data from input data, a convolutional gated recurrent network for determining temporal connections based on the feature data, and a decoder for determining a recommendation degree prediction value based on a last state output by the convolutional gated recurrent network.
[0023] Furthermore, the transaction indicator data includes transaction quantity, transaction unit price, transaction amount and transaction date;
[0024] Evaluation index data include performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score.
[0025] Furthermore, the pre-recommended customer groups are determined according to the transaction needs of the target transaction subject, including:
[0026] Identify customer groups that meet transaction needs;
[0027] All customers in the customer group who have not concluded a transaction with the target transaction entity are regarded as the pre-recommended customer group.
[0028] The present invention also provides a potential customer recommendation device, the device comprising:
[0029] A pre-recommended customer determination module is used to determine the pre-recommended customer group according to the transaction needs of the target transaction subject; wherein the transaction needs include buying and selling needs and target transaction product categories;
[0030] A recommendation degree determination module is connected to the pre-recommended customer determination module, and is used to obtain input data corresponding to each pre-recommended customer in the pre-recommended customer group, and input the input data corresponding to each pre-recommended customer into a potential customer recommendation model corresponding to the buying and selling needs, so as to obtain the recommendation degree corresponding to each pre-recommended customer; wherein the input data includes transaction index data and evaluation index data, and the potential customer recommendation model is obtained by a preset model training strategy;
[0031] The recommendation module is connected to the recommendation degree determination module and is used to screen out all potential customers from the pre-recommended customer group according to the recommendation degree and recommend all potential customers to the target transaction subject.
[0032] The present invention also provides a computer-readable storage medium, wherein the computer program stored in the computer-readable storage medium implements the steps of the above method when executed by one or more processors.
[0033] The present invention also provides an electronic device, comprising a memory and one or more processors, wherein a computer program is stored in the memory, and when the computer program is executed by the one or more processors, the steps of the above method are executed.
[0034] The potential customer recommendation method, device and electronic device provided by the present invention have at least the following beneficial effects:
[0035] (1) When recommending potential customers, the transaction indicator data and evaluation indicator data corresponding to each pre-recommended customer are input into the potential customer recommendation model as input data, so that the indicators used when recommending potential customers are more diversified, and accurate recommendations for potential customers can be achieved. At the same time, the potential customer recommendation model selected when recommending is a model corresponding to buying and selling needs, which can further improve the accuracy of potential customer recommendations, greatly reduce conversion costs, and increase revenue.
[0036] (2) When training the model, multiple training data are determined based on the historical transaction data of multiple categories of products and the evaluation score data of the recommended party in the transaction. The training data includes transaction indicator data and evaluation indicator data. When training the model, the historical transaction data and evaluation score data are fully utilized. The training indicators are more diversified, which can obtain a more efficient potential customer recommendation model and improve the accuracy of model prediction.
[0037] (3) The potential customer recommendation model is based on the symmetric structure of "encoder-decoder", which improves the feature utilization rate. At the same time, it uses a convolutional gated recurrent network to effectively process time series data, further improving the generalization ability of the potential customer recommendation model and the accuracy of the prediction results. At the same time, the potential customer recommendation model can be applied to different business scenarios and is suitable for any field where transactions exist. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0039] It should also be noted that, for the convenience of description, only the parts related to the present disclosure are shown in the drawings. The drawings constituting a part of the specification of this application are used to provide a further understanding of the present application. The illustrative embodiments and their descriptions in this application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0040] Figure 1 A flowchart of a method for recommending potential customers provided in one embodiment of the present invention;
[0041] Figure 2 A flowchart of the steps of a preset model training strategy method provided in one embodiment of the present invention;
[0042] Figure 3 A schematic diagram of the structure of the Encode_Decode+C_GRU network model provided in one embodiment of the present invention;
[0043] Figure 4 A schematic diagram of the structure of a CGRU module provided in one embodiment of the present invention;
[0044] Figure 5 A schematic diagram of the structure of a potential customer recommendation device provided in yet another embodiment of the present invention;
[0045] Figure 6 A schematic diagram of the structure of an electronic device provided in one embodiment of the present invention;
[0046] Reference numerals:
[0047] Figure 5 In: 501-pre-recommended customer determination module, 502-recommendation degree determination module, 503-recommendation module;
[0048] Figure 6 In it: 600 - electronic device, 601 - processor, 602 - communication bus, 603 - user interface, 604 - communication interface, 605 - memory. DETAILED DESCRIPTION
[0049] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.
[0050] In one embodiment of the present invention, Figure 1 As shown, a potential customer recommendation method is provided, the method comprising:
[0051] Step S101: Determine a pre-recommended customer group according to the transaction needs of the target transaction subject.
[0052] Among them, transaction demand includes buying and selling demand and target transaction product category. Buying and selling demand refers to the demand to buy or sell products. When the buying and selling demand of the target transaction subject is the demand to buy products, the target transaction subject is the purchaser and the pre-recommended customer is the supplier. When the buying and selling demand of the target transaction subject is the demand to sell products, the target transaction subject is the supplier and the pre-recommended customer is the purchaser.
[0053] In one implementation, step S101 specifically includes the following steps:
[0054] Step S1011: Determine the customer group that meets the transaction requirements.
[0055] Taking the example that the target transaction subject has a buying demand for a certain category of products, that is, the target transaction subject is the purchaser, then the customers who have a selling demand for this category of products are identified as customers who meet the transaction demand, and all customers who meet the transaction demand are identified as the customer group that meets the transaction demand.
[0056] Optionally, before recommending potential customers, a data detail table can be pre-collected. The data detail table contains historical transaction data and evaluation score data of all purchasers and suppliers of each category of products. After the data detail table is statistically generated, the data detail table can be used to determine the customer group that meets the transaction needs.
[0057] For example, each data in the data details table includes the purchasing party's company name, the supplier's company name, transaction product category, transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, most recent transaction date, most recent transaction date, purchasing party's performance evaluation score, purchasing party's credit evaluation score, purchasing party's product evaluation score, purchasing party's innovation evaluation score, supplier's performance evaluation score, supplier's credit evaluation score, supplier's product evaluation score, and supplier's innovation evaluation score.
[0058] Specifically, when constructing the data details table, all historical transaction data of the same purchaser and the same supplier for the same category of products are counted and analyzed to obtain the total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, most recent transaction date, and most distant transaction date of the same purchaser and the same supplier for the same category of products. If there are multiple transactions with different transaction unit prices, the average value of the transaction unit prices can be taken as the final transaction unit price; at the same time, the purchaser's performance evaluation score, purchaser's credit evaluation score, purchaser's product evaluation score, purchaser's innovation evaluation score, supplier's performance evaluation score, supplier's credit evaluation score, supplier's product evaluation score, and supplier's innovation evaluation score are obtained to form a piece of data in the data details table.
[0059] It is understandable that when the purchaser does not sell products of this category (that is, the purchaser only buys but does not sell products of this category), the purchaser's product evaluation score is empty or a preset value.
[0060] Furthermore, if a transaction party (purchaser or supplier) of a certain category of products has not concluded a transaction with any other transaction party, then its transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, latest transaction date, and furthest transaction date are empty or preset values. If the transaction party is the purchaser, since there is no supplier with which it has conducted transactions, the supplier's performance evaluation score, supplier's credit evaluation score, supplier's product evaluation score, and supplier's innovation evaluation score are empty or preset values; if the transaction party is the supplier, the purchaser's performance evaluation score, purchaser's credit evaluation score, purchaser's product evaluation score, and purchaser's innovation evaluation score are empty or preset values.
[0061] Furthermore, historical transaction data and evaluation score data can be obtained from external systems. For example, historical transaction information can be obtained from an e-commerce transaction platform, and the score data corresponding to the transaction parties can be obtained from an enterprise information query system (such as Qichacha).
[0062] Step S1012: All customers in the customer group who have not concluded a transaction with the target transaction subject are taken as a pre-recommended customer group.
[0063] Specifically, the data details table can be used to filter out all customers who have not concluded a transaction with the target transaction entity from the customer group.
[0064] Step S102: Obtain input data corresponding to each pre-recommended customer in the pre-recommended customer group, and input the input data corresponding to each pre-recommended customer into a potential customer recommendation model corresponding to the buying and selling demand to obtain a recommendation degree corresponding to each pre-recommended customer.
[0065] Among them, the input data includes transaction indicator data and evaluation indicator data, and the potential customer recommendation model is obtained through a preset model training strategy.
[0066] If the buying and selling demand is a buying demand, that is, the target transaction subject is the purchaser, then the potential customer recommendation model whose purpose is to recommend the supplier to the purchaser is selected as the model corresponding to the buying and selling demand; if the buying and selling demand is a selling demand, that is, the target transaction subject is the supplier, then the potential customer recommendation model whose purpose is to recommend the purchaser to the supplier is selected as the model corresponding to the buying and selling demand.
[0067] Specifically, the input data corresponding to each pre-recommended customer can be input into the potential customer recommendation model in a preset order, and the recommendation degree prediction value corresponding to each pre-recommended customer can be determined in the order in which the potential customer recommendation model outputs the recommendation degree prediction value.
[0068] Step S103: Screen out all potential customers from the pre-recommended customer group according to the recommendation degree, and recommend all potential customers to the target transaction subject.
[0069] For example, for any pre-recommended customer in the pre-recommended customer group, when the predicted recommendation degree prediction value is not less than 0.5, the recommendation degree prediction value is taken as 1 after soft voting, and the customer is recommended as a potential customer. When the predicted recommendation degree prediction value is less than 0.5, the recommendation degree prediction value is taken as 0 after soft voting, that is, the pre-recommended customer is not recommended.
[0070] Furthermore, when there are multiple potential customers screened out, all potential customers are recommended. Specifically, the potential customers can be sorted from large to small according to the recommendation degree prediction value output by the model, and the sorting results are displayed to the target transaction entity.
[0071] The potential customer recommendation method provided in this embodiment, when making potential customer recommendations, inputs the transaction indicator data and evaluation indicator data corresponding to each pre-recommended customer as input data into the potential customer recommendation model, so that the indicators based on which potential customers are recommended are more diversified, and accurate recommendations for potential customers can be achieved. At the same time, the potential customer recommendation model selected when recommending is a model corresponding to buying and selling needs, which can further improve the accuracy of potential customer recommendations, greatly reduce conversion costs, and increase revenue.
[0072] In another embodiment of the present invention, based on the previous embodiment, Figure 2 As shown, the preset model training strategy includes the following steps:
[0073] Step S201: Determine a plurality of training data according to historical transaction data of products of multiple categories and evaluation score data of the recommended party among the two transaction parties.
[0074] Among them, the training data includes transaction indicator data and evaluation indicator data.
[0075] For example, the transaction index data includes the transaction quantity, transaction unit price, transaction amount and transaction date. The evaluation index data includes the performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score.
[0076] It should be understood that the input data input into the potential customer recommendation model in the prediction stage should be consistent with the training data input when training the potential customer recommendation model. Therefore, when the transaction indicator data in the training data include the transaction quantity, transaction unit price, transaction amount and transaction date, and the evaluation indicator data include the performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score, the transaction indicator data in the input data also include the transaction quantity, transaction unit price, transaction amount and transaction date, and the evaluation indicator data also include the performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score.
[0077] The recommended party among the two parties to the transaction is determined based on the preset recommendation purpose of the potential customer recommendation model.
[0078] For example, if the preset recommendation purpose of the potential customer recommendation model is to recommend the supplier to the purchaser, the recommended party is the supplier among the two transaction parties, that is, the supplier is recommended to the purchaser as a potential customer; if the preset recommendation purpose of the potential customer recommendation model is to recommend the purchaser to the supplier, the recommended party is the purchaser among the two transaction parties, that is, the purchaser is recommended to the supplier as a potential customer.
[0079] Similarly, the historical transaction data and evaluation score data mentioned in this step can be obtained from an external system. For example, historical transaction information can be obtained from an e-commerce transaction platform, and the score data corresponding to the transaction parties can be obtained from an enterprise information query system (such as Qichacha).
[0080] In one implementation, multiple training data are determined based on historical transaction data of multiple categories of products and evaluation score data of the recommended party of the transaction, including the following steps:
[0081] Step S2011: Aggregate historical transaction data of multiple categories of products according to a preset strategy to obtain multiple initial transaction data sets.
[0082] Among them, each initial transaction data set includes all historical transaction data of the same transaction parties based on the same category of products, that is, in step S2011, the preset strategy includes: aggregating all historical transaction data of the same transaction parties based on the same category of products to form an initial transaction data set.
[0083] For example, if a supplier A has conducted transactions with buyers B, C, and D based on a certain category of products, then all historical transaction data of supplier A and buyer B based on the category of products (multiple transactions may be conducted) are aggregated together; all historical transaction data of supplier A and buyer C based on the category of products are aggregated together; all historical transaction data of supplier A and buyer D based on the category of products are aggregated together.
[0084] It should be understood that if A has traded with B as a supplier and with B as a purchaser for a certain category of products, that is, both A and B have engaged in both buying and selling of the products in the category. At this time, it is necessary to aggregate all the historical transaction data of supplier A and purchaser B based on the products in the category to form an initial transaction data set; and aggregate all the historical transaction data of purchaser A and supplier B based on the products in the category to form an initial transaction data set.
[0085] Step S2012: pre-process the data in each initial transaction data set respectively to obtain transaction indicator data corresponding to the initial transaction data set.
[0086] For example, each piece of historical transaction data in the initial transaction data set includes the supplier name, the purchaser name, the transaction product category, the transaction quantity, the transaction unit price, the transaction amount, and the transaction date.
[0087] It is understandable that in different historical transaction data of the same initial transaction data set, the supplier name, the buyer name and the transaction product category are the same.
[0088] When preprocessing the data in a certain initial transaction data set, the total transaction amount, the maximum transaction amount, the minimum transaction amount, the average transaction amount, the total transaction quantity, the maximum transaction quantity, the minimum transaction quantity, the average transaction quantity, the most recent transaction date, and the most recent transaction date are obtained by performing data analysis on the transaction amount, transaction quantity, and transaction date.
[0089] The transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, latest transaction date, and longest transaction date are used as transaction indicator data.
[0090] It is understandable that if the transaction unit price is different each time a transaction is conducted, the transaction unit price can also be analyzed during preprocessing to calculate the average transaction unit price of all transactions as the final transaction unit price. The final transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, most recent transaction date, and most distant transaction date are used as transaction indicator data.
[0091] Step S2013: Using the evaluation score data of the recommended party among the two transaction parties corresponding to each initial transaction data set as the evaluation index data corresponding to the initial transaction data set.
[0092] The evaluation index data includes the performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score. Among them, each initial transaction data set corresponds to a certain category of products, and the product evaluation score is the evaluation score of the category of products corresponding to the initial transaction data set. Furthermore, if the recommended party is the purchaser and the recommended party does not sell this type of product, then the product evaluation score can be set to empty or a preset value (such as 0).
[0093] Step S2014: taking the transaction index data and the evaluation index data corresponding to each initial transaction data set as one piece of training data respectively, so as to obtain multiple pieces of training data.
[0094] It can be understood that if during training, the training data includes transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, most recent transaction date, farthest transaction date, performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score, then the input data also includes transaction unit price, total transaction amount, maximum transaction amount, minimum transaction amount, average transaction amount, total transaction quantity, maximum transaction quantity, minimum transaction quantity, average transaction quantity, most recent transaction date, farthest transaction date, performance evaluation score, credit evaluation score, product evaluation score (product evaluation score corresponding to the target transaction product category) and innovation evaluation score.
[0095] For example, taking the target transaction subject as the purchaser, for each pre-recommended customer, the input data is obtained by the following method:
[0096] The transaction data of all transactions conducted by the pre-recommended customer as a supplier based on the target transaction product category are counted to calculate the total transaction amount, the maximum transaction amount, the minimum transaction amount, the average transaction amount, the total transaction quantity, the maximum transaction quantity, the minimum transaction quantity, the average transaction quantity, the latest transaction date, and the furthest transaction date. If there are multiple transactions with different transaction unit prices, the average transaction unit price can be taken as the final transaction unit price, and the final transaction unit price, the total transaction amount, the maximum transaction amount, the minimum transaction amount, the average transaction amount, the total transaction quantity, the maximum transaction quantity, the minimum transaction quantity, the average transaction quantity, the latest transaction date, and the furthest transaction date are taken as the transaction indicator data corresponding to the pre-recommended customer. At the same time, the performance evaluation score, credit evaluation score, product evaluation score (the product evaluation score corresponding to the target transaction product category) and innovation evaluation score of the pre-recommended customer are obtained as the evaluation indicator data corresponding to the pre-recommended customer.
[0097] Step S202: Determine the label corresponding to each piece of training data.
[0098] The label includes a recommendation degree label. In one implementation, a recommendation degree of 0 indicates not recommending, and a recommendation degree of 1 indicates recommending. For example, the value of the recommendation degree can be determined by the latest transaction date. For example, if the latest transaction date is before a preset date, the recommendation degree is 0, and if the latest transaction date is between the preset date and after the preset date, the recommendation degree is 1.
[0099] The preset date can be set by those skilled in the art according to actual needs, and the present invention does not limit this. For example, the recommendation degree is determined based on whether a transaction has been conducted in the last two years. That is, assuming that the current date is July 9, 2023, the preset date can be set to July 9, 2021, or the preset date can be set to January 1, 2022. That is, if the last transaction date is before 2022, the recommendation degree is 0, and if the last transaction date is after 2022, the recommendation degree is 1.
[0100] Step S203: taking each piece of training data in the plurality of training data and the label corresponding to the training data as a training sample to construct a sample data set.
[0101] Step S204: training the preset network model through the N-fold cross-validation method according to the sample data set to obtain a potential customer recommendation model.
[0102] Optionally, the preset network model is the Encode_Decode+C_GRU network model, such as Figure 3 As shown, the model includes an encoder for extracting feature data from input data, a convolutional gated recurrent network (CGRU) for determining temporal connections based on the feature data, and a decoder for determining a recommendation degree prediction value based on the last state output by the convolutional gated recurrent network. The smallest unit of the model includes a convolutional layer, a batch normalization layer (BN), and a LeakyReLU layer, and the model is constructed as an L layer.
[0103] Specifically, the encoder is used to extract deep features. Each layer of the encoder includes two minimum units, and the dimension is reduced through the maximum pooling layer (Max-Pool) between each layer. The encoder inputs the extracted deep features into the convolutional gated recurrent network, which is used to obtain the temporal connection in multiple order transactions of the same purchaser and supplier, and inputs the last state of the convolutional gated recurrent network into the decoder. Each layer of the decoder contains two minimum units, and the dimension is restored through upsampling between each layer to obtain the final recommendation prediction value.
[0104] The CGRU module structure in the Encode_Decode+C_GRU network model is as follows Figure 4As shown in the figure, it has the ability to mine information from long-distance sequence data, infer the feature connections between sequence data, and solve the intensity similarity problem by capturing the feature details across sequences. CGRU uses a reset gate and an update gate. The reset gate determines which information to discard, and the update gate determines which information to remove or update.
[0105] CGRU takes as input {X1,X2,...,X t}、Hidden state{H1,H2,...,H t} and Z t 、r t They are all expanded into three-dimensional tensors, and the last two dimensions are spatial information (rows and columns). At this time, the update formula of CGRU is shown in the following equations (1) to (4):
[0106] z t =sigmoid((X t +H t-1 )*W z +b z ) (1)
[0107] r t =sigmoid((X t +H t-1 )*W r +b r ) (2)
[0108]
[0109]
[0110] Among them, * represents the convolution operation, Sigmoid represents the activation function, and X t and H t Respectively represent the input state and hidden state at time t, r t , Z t and They represent the reset gate, update gate and new information respectively, W is the two-dimensional convolution kernel, and b is the bias term.
[0111] In this embodiment, the potential customer recommendation model is obtained by training the preset network model through the N-fold cross validation method, preferably the 5-fold cross validation method, which can improve the persuasiveness of the experimental results and effectively avoid overfitting.
[0112] The training steps include: dividing all the training data in the sample data set into five parts, taking one of them as the test set and the other four as the training set. The training set is divided into an internal training set and an internal validation set at a ratio of 4:1 within the training set to achieve real-time verification of the segmentation performance and model preservation during the training process. Repeat the above training steps five times, and test all the training data of the sample data set through five cross-validations to complete the overall performance evaluation.
[0113] After the overall training and prediction are completed, a soft vote is performed on the recommendation degree (P) of the prediction result, as shown in the following formula (5):
[0114]
[0115] After soft voting, each recommendation prediction value is 0 or 1, where 0 means not recommended and 1 means recommended.
[0116] Optionally, the preset model training strategy also includes using the error between the actual value (label) of the recommendation degree and the predicted value of the recommendation degree after soft voting to evaluate the accuracy of the potential customer recommendation model.
[0117] When the predicted value of the recommendation is closer to the true value of the recommendation, the performance of the potential customer recommendation model is better, otherwise it is the opposite. In order to more intuitively judge the performance of the potential customer recommendation model, we can use accuracy, precision and recall for evaluation and analysis. Accuracy (Acc) is the proportion of the number of correct predictions of the model to the total number, as shown in formula (6); precision (PRE) is based on the prediction results, and the proportion of the samples predicted as positive examples is the correct proportion, as shown in formula (7); recall (RE) is based on the actual samples, and the proportion of the positive samples predicted correctly to the total actual positive samples is the proportion of the positive samples predicted correctly, as shown in formula (8).
[0118] Acc=(TP+TN) / (TP+FP+FN+TN) (6)
[0119] PRE=TP / (TP+FP) (7)
[0120] PRE=TP / (TP+FN) (8)
[0121] Among them, TP is a true positive example, FP is a false positive example, FN is a false negative example, and TN is a true negative example.
[0122] The potential customer recommendation method provided in this embodiment determines multiple training data based on historical transaction data of multiple categories of products and evaluation score data of the recommended party in the transaction when training the model. The training data includes transaction index data and evaluation index data. The transaction data and evaluation score data are fully utilized during model training. The training indicators are more diversified, and a more efficient potential customer recommendation model can be obtained to improve the accuracy of model prediction. In addition, the potential customer recommendation model is based on the symmetric structure of "encoder-decoder", which improves the feature utilization rate. At the same time, it uses a convolutional gated recurrent network to effectively process time series data, further improving the generalization ability of the potential customer recommendation model and the accuracy of the prediction results. At the same time, the potential customer recommendation model can be applied to different business scenarios and is applicable to any field where transactions exist.
[0123] In another embodiment of the present invention, a potential customer recommendation device is also provided. Figure 5 As shown, the device comprises:
[0124] The pre-recommended customer determination module 501 is used to determine the pre-recommended customer group according to the transaction needs of the target transaction subject; wherein the transaction needs include buying and selling needs and target transaction product categories.
[0125] The recommendation degree determination module 502 is connected to the pre-recommended customer determination module 501, and is used to obtain input data corresponding to each pre-recommended customer in the pre-recommended customer group, and input the input data corresponding to each pre-recommended customer into the potential customer recommendation model corresponding to the buying and selling needs, so as to obtain the recommendation degree corresponding to each pre-recommended customer, wherein the input data includes transaction indicator data and evaluation indicator data, and the potential customer recommendation model is obtained through a preset model training strategy.
[0126] The recommendation module 503 is connected to the recommendation degree determination module 502 and is used to screen out all potential customers from the pre-recommended customer group according to the recommendation degree and recommend all potential customers to the target transaction subject.
[0127] The potential customer recommendation device provided by the present invention can recommend potential customers more accurately, reduce the potential customer conversion cost, and greatly increase the profit.
[0128] Optionally, the pre-recommended customer determination module includes:
[0129] A customer determination unit, used to determine customer groups that meet transaction requirements;
[0130] The pre-recommended customer screening unit is connected to the customer determination unit and is used to select all customers in the customer group who have not concluded a transaction with the target transaction subject as the pre-recommended customer group.
[0131] Optionally, the preset model training strategy includes:
[0132] Determine multiple training data based on historical transaction data of multiple categories of products and evaluation score data of the recommended party among the two parties of the transaction; wherein the training data includes transaction index data and evaluation index data, and the recommended party is determined according to the preset recommendation purpose of the potential customer recommendation model;
[0133] Determine the label corresponding to each piece of training data, wherein the label includes a recommendation degree label;
[0134] Each piece of training data and a label corresponding to the training data are respectively used as a training sample to construct a sample data set;
[0135] According to the sample data set, the preset network model is trained through the N-fold cross-validation method to obtain the potential customer recommendation model.
[0136] Optionally, multiple training data are determined based on historical transaction data of multiple categories of products and evaluation score data of the recommended party among the two transaction parties, including:
[0137] Aggregate historical transaction data of multiple categories of products according to a preset strategy to obtain multiple initial transaction data sets; wherein each initial transaction data set includes all historical transaction data of the same transaction parties based on the same category of products;
[0138] Preprocessing the data in each initial transaction data set respectively to obtain transaction indicator data corresponding to the initial transaction data set;
[0139] The evaluation score data of the recommended party among the two transaction parties corresponding to each initial transaction data set is used as the evaluation index data corresponding to the initial transaction data set;
[0140] The transaction index data and the evaluation index data corresponding to each initial transaction data set are respectively used as one piece of training data to obtain multiple pieces of training data.
[0141] Optionally, if the preset recommendation purpose of the potential customer recommendation model is to recommend a supplier to the purchaser, the recommended party is the supplier among the two parties of the transaction;
[0142] If the preset recommendation purpose of the potential customer recommendation model is to recommend a purchaser to a supplier, the recommended party is the purchaser among the two transaction parties.
[0143] Optionally, the transaction indicator data includes transaction quantity, transaction unit price, transaction amount and transaction date;
[0144] Evaluation index data include performance evaluation score, credit evaluation score, product evaluation score and innovation evaluation score.
[0145] Optionally, the preset network model includes: an encoder for extracting feature data from input data, a convolutional gated recurrent network for determining temporal connections based on the feature data, and a decoder for determining a recommendation degree prediction value based on a last state output by the convolutional gated recurrent network.
[0146] In another embodiment of the present invention, a computer program product is provided, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, all or part of the steps of the method in the above method embodiment are implemented. This embodiment will not be repeated here.
[0147] Further, the computer program product may include one or more computer executable components configured to execute the embodiments when the program is run; the computer program product may also include a computer program tangibly contained on a readable medium, the computer program including program code for executing any method in the embodiments of the present invention. In such an embodiment, the computer program may be downloaded and installed from a network through a communication part, and / or installed from a removable medium.
[0148] In another embodiment of the present invention, a computer-readable storage medium is further provided. The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements all or part of the steps of the method in the above method embodiment. This embodiment will not be repeated here.
[0149] In yet another embodiment of the present invention, an electronic device 600 is further provided. Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the electronic device 600 includes: at least one processor 601, at least one communication bus 602, a user interface 603, at least one external communication interface 604, and a memory 605. Among them, the communication bus 602 is configured to realize the connection and communication between these components. Among them, the user interface 603 may include a display screen, and the external communication interface 604 may include a standard wired interface and a wireless interface. A computer program is stored in the memory 605, and the memory 605 and one or more processors 601 are connected to each other in communication. When the computer program is executed by one or more processors, the processor 601 is configured to execute the computer program stored in the memory to implement all or part of the steps of the method in the above method embodiment, and this embodiment will not be repeated here.
[0150] The potential customer recommendation method, device and electronic device provided by the present invention, when recommending potential customers, inputs the transaction index data and evaluation index data corresponding to each pre-recommended customer as input data into the potential customer recommendation model, so that the indicators based on which the potential customers are recommended are more diversified, and accurate recommendations for potential customers can be achieved. At the same time, the potential customer recommendation model selected when recommending is a model corresponding to the buying and selling needs, which can further improve the accuracy of potential customer recommendations, greatly reduce conversion costs, and increase revenue. At the same time, when training the model, multiple training data are determined according to the historical transaction data of multiple categories of products and the evaluation score data of the recommended party in the transaction. The training data includes transaction index data and evaluation index data. When training the model, the transaction data and evaluation score data are fully utilized, and the training indicators are more diversified, so that a more efficient potential customer recommendation model can be obtained, and the accuracy of model prediction can be improved. In addition, the potential customer recommendation model is based on the symmetric structure of "encoder-decoder", which improves the feature utilization rate, and uses a convolutional gated recurrent network to effectively process time series data, further improving the generalization ability of the potential customer recommendation model and the accuracy of the prediction results, and at the same time, the potential customer recommendation model can be applied to different business scenarios and is applicable to any field where transactions exist.
[0151] The various embodiments in the present disclosure are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0152] The protection scope of the present disclosure is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the scope and spirit of the present disclosure. If these changes and modifications fall within the scope of the claims of the present disclosure and their equivalents, the intention of the present disclosure also includes these changes and modifications.
Claims
1. A potential customer recommendation method, characterized in that: The method comprises: Determine the pre-recommended customer group according to the transaction needs of the target transaction subject; wherein the transaction needs include buying and selling needs and target transaction product categories; Obtaining input data corresponding to each pre-recommended customer in the pre-recommended customer group, and inputting the input data corresponding to each pre-recommended customer into a potential customer recommendation model corresponding to the buying and selling demand, so as to obtain a recommendation degree corresponding to each pre-recommended customer; wherein the input data includes transaction index data and evaluation index data, and the potential customer recommendation model is obtained by a preset model training strategy; All potential customers are screened out from the pre-recommended customer group according to the recommendation degree, and all the potential customers are recommended to the target transaction subject.
2. The potential customer recommendation method according to claim 1, characterized in that: The preset model training strategy includes: Determine multiple training data based on historical transaction data of multiple categories of products and evaluation score data of the recommended party among the two parties of the transaction; wherein the training data includes transaction index data and evaluation index data, and the recommended party is determined according to the preset recommendation purpose of the potential customer recommendation model; Determine the label corresponding to each piece of training data respectively; wherein the label includes a recommendation degree label; Taking each piece of training data and a label corresponding to the training data as a training sample, respectively, to construct a sample data set; The potential customer recommendation model is obtained by training a preset network model according to the sample data set through an N-fold cross-validation method.
3. The potential customer recommendation method according to claim 2, characterized in that: The multiple training data are determined based on the historical transaction data of multiple categories of products and the evaluation score data of the recommended party of the transaction, including: Aggregate historical transaction data of multiple categories of products according to a preset strategy to obtain multiple initial transaction data sets; wherein each initial transaction data set includes all historical transaction data of the same transaction parties based on the same category of products; Preprocessing the data in each initial transaction data set respectively to obtain transaction indicator data corresponding to the initial transaction data set; The evaluation score data of the recommended party among the two transaction parties corresponding to each initial transaction data set is used as the evaluation index data corresponding to the initial transaction data set; The transaction index data and the evaluation index data corresponding to each initial transaction data set are respectively used as one piece of training data to obtain multiple pieces of training data.
4. The potential customer recommendation method according to claim 2, characterized in that: If the preset recommendation purpose of the potential customer recommendation model is to recommend a supplier to the purchaser, the recommended party is the supplier among the two parties to the transaction; If the preset recommendation purpose of the potential customer recommendation model is to recommend a purchaser to the supplier, then the recommended party is the purchaser among the two transaction parties.
5. The potential customer recommendation method according to claim 2, characterized in that: The preset network model includes: An encoder for extracting feature data from input data, a convolutional gated recurrent network for determining a temporal connection based on the feature data, and a decoder for determining a recommendation degree prediction value based on a last state output by the convolutional gated recurrent network.
6. The potential customer recommendation method according to claim 1 or 2, characterized in that: The transaction index data includes transaction quantity, transaction unit price, transaction amount and transaction date; The evaluation index data includes a performance evaluation score, a credit evaluation score, a product evaluation score and an innovation evaluation score.
7. The potential customer recommendation method according to claim 1, characterized in that: Determining the pre-recommended customer group according to the transaction needs of the target transaction subject includes: Identify customer groups that meet the transaction requirements; All customers in the customer group who have not concluded a transaction with the target transaction subject are taken as the pre-recommended customer group.
8. A potential customer recommendation device, characterized in that: The device comprises: A pre-recommended customer determination module is used to determine the pre-recommended customer group according to the transaction needs of the target transaction subject; wherein the transaction needs include buying and selling needs and target transaction product categories; A recommendation degree determination module, connected to the pre-recommended customer determination module, is used to obtain input data corresponding to each pre-recommended customer in the pre-recommended customer group, and input the input data corresponding to each pre-recommended customer into a potential customer recommendation model corresponding to the buying and selling demand, so as to obtain a recommendation degree corresponding to each pre-recommended customer; wherein the input data includes transaction index data and evaluation index data, and the potential customer recommendation model is obtained by a preset model training strategy; A recommendation module is connected to the recommendation degree determination module and is used to screen out all potential customers from the pre-recommended customer group according to the recommendation degree and recommend all potential customers to the target transaction subject.
9. A computer-readable storage medium, characterized in that: The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements the steps of the method according to any one of claims 1 to 7.
10. An electronic device, characterized in that: The method comprises a memory and one or more processors, wherein a computer program is stored in the memory, and when the computer program is executed by the one or more processors, the steps of the method according to any one of claims 1 to 7 are performed.
Citation Information
Cited By
Potential customer prediction method and device, electronic equipment and program product
CN119741054A
Potential customer prediction method, device, electronic equipment and program product
CN119741054B
