Marketing risk identification model training method, marketing risk identification method, marketing risk identification device and equipment

Abstract

The invention relates to the technical field of marketing risk control, and discloses a marketing risk identification model training method, device and equipment, and a marketing risk identification method, device and equipment, and the method comprises the steps: obtaining a training data set of a sample user, the training data set comprising dialogue text data and structured data; performing data preprocessing on the dialogue text data and the structured data to generate a dialogue text data feature vector and a structured data feature vector; performing first feature processing and second feature processing on the dialogue text data feature vector and the structured data feature vector respectively to generate a first training data set of a sample user and a second training data set of the sample user; according to the first training data set and the second training data set, training is carried out based on a classification algorithm, and a marketing risk identification model is generated. Through the method, the accuracy of marketing risk identification can be improved.

Description

Technical Field

[0001] The present invention relates to the field of marketing risk control technology, and in particular to a marketing risk identification model training, a marketing risk identification method, a device and equipment. Background Art

[0002] With the rapid development of communication technology, marketing methods have also become diversified, such as marketing through telephone, chat software, email, text messages and other methods. The above marketing methods can be targeted according to the customer's consumption behavior, interests, hobbies, purchasing power and other information, and communicate directly with customers, which can improve the success rate of marketing. However, the above marketing methods are easily disliked by customers. For most people, direct marketing is often regarded as interference and harassment. At the same time, direct marketing is to promote products and services through direct contact with customers, so its effect is difficult to predict and may sometimes achieve the opposite of the expected effect, such as leading to customer complaints.

[0003] At present, the existing technology mainly trains structured data such as marketing duration, marketing time period and marketing times to form a recognition model to identify whether the marketing has the risk of being complained. If there is a risk of complaint, then take post-remedial measures, such as adding such customers to the marketing blacklist. However, in reality, customers who are identified by the recognition model as having the risk of complaints still have the desire to buy. If the above-mentioned post-remedial measures are taken, it will result in the loss of potential customers. Therefore, there is an urgent need for a technical solution to solve the above-mentioned technical problems, thereby improving the accuracy of marketing risk identification. Summary of the invention

[0004] The present invention provides a marketing risk identification model training, a marketing risk identification method, a device and equipment to solve the technical problem of low accuracy of marketing risk identification in the prior art.

[0005] In a first aspect, a marketing risk identification model training method is provided, comprising:

[0006] Obtaining a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data;

[0007] After performing data preprocessing on the dialogue text data and the structured data respectively, a dialogue text data feature vector and a structured data feature vector are generated;

[0008] After performing a first feature processing on the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated;

[0009] After performing a second feature processing on the conversation text data feature vector and the structured data feature vector, a second training data set of sample users is generated;

[0010] According to the first training data set and the second training data set, training is performed separately based on a classification algorithm to generate a marketing risk identification model.

[0011] Secondly, a marketing risk identification method is provided, including:

[0012] Obtain the target user's data set;

[0013] The data set is input into a marketing risk identification model for risk identification, and an identification result is output, wherein the marketing risk identification model is trained according to the above-mentioned marketing risk identification model training method.

[0014] In a third aspect, a marketing risk identification model training device is provided, comprising:

[0015] Dataset acquisition module: used to acquire a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data;

[0016] A feature vector generating module: used for generating a feature vector of the dialogue text data and a feature vector of the structured data after performing data preprocessing on the dialogue text data and the structured data respectively;

[0017] A first feature processing module: used for performing a first feature processing on the feature vector of the conversation text data and the feature vector of the structured data to generate a first training data set of sample users;

[0018] A second feature processing module: used for performing a second feature processing on the conversation text data feature vector and the structured data feature vector to generate a second training data set of sample users;

[0019] Model training generation module: used to generate a marketing risk identification model by performing training based on the first training data set and the second training data set and based on a classification algorithm.

[0020] In a fourth aspect, a marketing risk identification device is provided, comprising:

[0021] Dataset acquisition module: used to obtain the target user's dataset;

[0022] Risk identification and identification result output module: used to input the data set into the marketing risk identification model for risk identification and output the identification result, wherein the marketing risk identification model is trained according to the above-mentioned marketing risk identification model training device.

[0023] In a fifth aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned marketing risk identification model training method or the marketing risk identification method are implemented.

[0024] The above-mentioned marketing risk identification model training, marketing risk identification method, device and equipment first obtain a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data; then, the conversation text data and the structured data are respectively preprocessed to generate a conversation text data feature vector and a structured data feature vector; secondly, the conversation text data feature vector and the structured data feature vector are subjected to a first feature processing to generate a first training data set of sample users; thirdly, the conversation text data feature vector and the structured data feature vector are subjected to a second feature processing to generate a second training data set of sample users; finally, according to the first training data set and the second training data set, training is performed based on a classification algorithm to generate a marketing risk identification model.

[0025] This application, through the model self-learning including dialogue text data and structured data of the diversified data set, compared with only learning structured data, can more accurately identify the customer's complaint tendency. At the same time, the training data set of the sample users in this application is pre-labeled with multiple category labels, wherein the category labels include one or more labels of complaint risk of telephone harassment, complaint risk of incomplete notification, complaint risk of poor service attitude of marketing personnel, and complaint risk of false information. That is, the marketing risk identification model in this application can self-learn and train the data set of multiple complaint reasons, and the marketing risk identification model generated at the end can not only identify whether there is a complaint risk, but also identify the specific complaint risk reasons. The recognition results output by the marketing risk identification model are more comprehensive and more accurate. Finally, this application generates two types of data sets by processing the features of dialogue text data and structured data in two ways, and trains and generates two marketing risk identification models in turn according to the two types of data sets. Finally, the two marketing risk identification models are used to identify the risk of marketing behavior, which can solve the technical problem of low prediction accuracy of a single model and further improve the stability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0027] Figure 1 It is a flow chart of a marketing risk identification model training method in one embodiment of the present invention;

[0028] Figure 2 is a schematic diagram of the structure of a training data set of sample users in one embodiment of the present invention;

[0029] Figure 3 It is a schematic diagram of a framework for training a marketing risk identification model in one embodiment of the present invention;

[0030] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation of step S30;

[0031] Figure 5 yes Figure 1 A schematic flow chart of a specific implementation of step S40;

[0032] Figure 6 yes Figure 1 A schematic flow chart of a specific implementation of step S50;

[0033] Figure 7 is a flow chart of a marketing risk identification method in one embodiment of the present invention;

[0034] Figure 8 yes Figure 7 A schematic flow chart of a specific implementation method of step S2-20;

[0035] Fig. 9 yes Figure 8 A schematic flow chart of a specific implementation method of step S2-23;

[0036] Fig.10 It is a structural schematic diagram of a marketing risk identification model training device in one embodiment of the present invention;

[0037] Fig.11 is a structural schematic diagram of a marketing risk identification device in one embodiment of the present invention;

[0038] Fig.12 It is a structural diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] See also Figure 1 As shown, Figure 1 The present invention is a flowchart of a marketing risk identification model training method according to an embodiment of the present invention, comprising the following steps:

[0041] S10: Obtain a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels and includes conversation text data and structured data.

[0042] The training data set in the embodiment of the present application is a multivariate data set. Figure 2 As shown, Figure 2 It is a schematic diagram of the structure of a training data set for sample users in one embodiment of the present invention. Among them, the training data set is composed of sample data of a large number of historical users, for example, sample data of historical users in the past six months or the past year can be collected. In order to improve the accuracy of the recognition results of the marketing risk identification model, the training data set not only includes structured data such as basic information of users and behavioral information of users, but also includes conversation text data of users, such as conversation records of phone calls, text messages and chat software. Among them, the basic information of users may include information such as gender, age and assets of users. The behavioral information of users may include information such as the number of times users click on the product introduction page and the time they stay on the product introduction page. The embodiment of the present application can improve the accuracy of the recognition results of the marketing risk identification model by obtaining a more diversified training data set and using the model to perform self-training and learning on the above training data set.

[0043] At the same time, in order to further ensure that the recognition results output by the marketing risk identification model are more comprehensive, the training data set is pre-labeled with multiple category labels. Among them, the category labels include one or more labels of complaint risks of telephone harassment, complaint risks of incomplete notification, complaint risks of poor service attitude of marketing personnel, and complaint risks of false information. That is, the marketing risk identification model in the embodiment of the present application can perform self-learning training on data sets of multiple complaint reasons, and the marketing risk identification model finally generated can not only identify whether there is a complaint risk, but also identify the specific reasons for the complaint risk.

[0044] S20: Preprocessing the conversation text data and the structured data respectively to generate a conversation text data feature vector and a structured data feature vector.

[0045] See also Figure 3 As shown, Figure 3 : is a schematic diagram of the framework of the marketing risk identification model training in one embodiment of the present invention. Figure 3 The data preprocessing module is mainly used to perform data preprocessing on the conversation text data and the structured data. For example, after the structured data is cleaned in advance, missing values ​​are filled, outliers are removed, etc., one-hot encoding is performed to generate a feature vector. For another example, for the conversation text data, a feature vector is generated based on a pre-trained bag-of-words vector model or a BERT (Bidirectional Encoder Representation from Transformers) model. It should be noted that the model pre-trained above extracts the vector representation corresponding to each word in the conversation text. In one embodiment, the feature vector of the conversation text data can be generated by taking the average value of the vector corresponding to each word in the conversation text.

[0046] In one case, the collected training data set has sample imbalance, such as more negative sample data from non-complaining users and less positive sample data from complaining users. Figure 3 The data preprocessing module in can also be used to perform data enhancement on part of the training data set, thereby expanding the number of positive samples. Data enhancement refers to a method of increasing the amount of data by adding minor changes to existing data or creating new synthetic data from existing data. In the embodiment of the present application, in the data preprocessing stage, the technical means of data enhancement are used to ensure that the ratio of positive and negative sample data in the training data set is within an appropriate range, which can further improve the accuracy of model recognition.

[0047] S30: After performing a first feature processing on the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated.

[0048] See also Figure 4 As shown, Figure 4 yes Figure 1 A specific implementation flow diagram of step S30 includes the following steps:

[0049] S31: Obtaining the dialogue text data feature vector and the structured data feature vector;

[0050] S32: After performing feature vector concatenation on the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated.

[0051] Feature vector concatenation mainly connects two or more feature vectors in a certain order to form a longer vector. Feature vector concatenation is a feature engineering technique that can obtain a more comprehensive and multivariate feature vector, thereby helping the model obtain more detailed and comprehensive feature information and improve model performance.

[0052] S40: After performing a second feature processing on the conversation text data feature vector and the structured data feature vector, a second training data set of sample users is generated.

[0053] See also Figure 5 As shown, Figure 5 yes Figure 1 A specific implementation flow diagram of step S40 includes the following steps:

[0054] S41: Acquire the dialogue text data feature vector and the structured data feature vector;

[0055] S42: Based on a preset weight coefficient, the feature vector of the conversation text data and the feature vector of the structured data are feature-fused to generate a second training data set of the sample user.

[0056] In one embodiment, the weight coefficient of the conversation text data is pre-set to 2 / 3, and the weight coefficient of the structured data is pre-set to 1 / 3, that is, the feature vector value of the conversation text data is multiplied by 2 / 3, and then the feature vector value of the structured data is multiplied by 1 / 3, and then the two are added to obtain the final feature vector value. Since the conversation text data is generally considered to be more important for risk identification in the application scenario of marketing risk identification, the weight coefficient of the conversation text data is set higher than the weight coefficient of the structured data, thereby helping the model learn more important data features. It should be noted that the weight coefficient pre-set in the embodiment of the present application is a hyperparameter of the model. The hyperparameter is a parameter set before the model starts the learning process. The hyperparameter will be optimized through subsequent training and learning of the training data set to improve the performance and effect of model learning.

[0057] S50: Performing training based on the first training data set and the second training data set and based on a classification algorithm to generate a marketing risk identification model.

[0058] See also Figure 6 As shown, Figure 6 yes Figure 1 A specific implementation flow diagram of step S50 includes the following steps:

[0059] S51: inputting the first training data set into the classification algorithm model for training until convergence, and generating a first marketing risk identification model;

[0060] S52: Input the second training data set into the classification algorithm model for training until convergence, and generate a second marketing risk identification model.

[0061] The first training data set and the second training data set in the embodiment of the present application adopt two different feature processing technical means, so the data features shown are different. After the model performs self-learning and training on the above two different training data sets, the performance of the model will also be different. The embodiment of the present application generates a first marketing risk identification model and a second marketing risk identification model through training, and then constructs a marketing risk identification model based on the above two different marketing risk identification models, which can solve the technical problem of low prediction accuracy of a single model and further improve the stability of the model.

[0062] At the same time, since the marketing risk identification model trained in the embodiment of the present application is a multi-classification prediction model, the algorithm of the model is a classification algorithm. In one embodiment, the classification algorithm model can adopt a capsule neural network model. The capsule neural network model is an improved model based on the convolutional neural network model, which mainly includes a three-layer network structure, namely a convolution layer, a main capsule layer, and a digital capsule layer. The capsule neural network model does not have the pooling operation of the traditional model, and through the above-mentioned network structure, it can learn more feature parameters than the convolutional neural network model, further improving the accuracy of marketing risk identification.

[0063] It should be noted that the marketing risk identification model in the embodiment of the present application is used to output the marketing risk identification result. Since the training data set of the marketing risk identification model is pre-labeled with multiple category labels. Among them, the category labels include complaint risk labels for multiple complaint reasons. Therefore, the identification result of the marketing risk identification model not only includes whether there is a complaint risk, but also includes the complaint risk reason in the case of a complaint.

[0064] At the same time, this application embodiment provides a marketing risk identification method, please refer to Figure 7 As shown, Figure 7 The present invention is a flowchart of a marketing risk identification method according to an embodiment of the present invention, comprising the following steps:

[0065] S2-10: Obtain the target user's data set;

[0066] S2-20: Input the data set into a marketing risk identification model for risk identification, and output an identification result, wherein the marketing risk identification model is trained according to the above-mentioned marketing risk identification model training method.

[0067] In the embodiment of the present application, the target user's data set also includes the target user's conversation text data and structured data. The data set is input into the marketing risk identification model for risk identification to obtain an identification result.

[0068] In one embodiment, the marketing risk identification model includes a first marketing risk identification model and a second marketing risk identification model. Figure 8 As shown, Figure 8 yes Figure 7 A specific implementation flow diagram of step S2-20 in the embodiment includes the following steps:

[0069] S2-21: outputting a first identification result based on the first marketing risk identification model;

[0070] S2-22: outputting a second identification result based on the second marketing risk identification model;

[0071] S2-23: Based on the first recognition result and the second recognition result, calculate and output the recognition result.

[0072] Similarly, since the training method of the above-mentioned marketing risk identification model forms two different types of marketing risk identification models, when the target user's data set is input into the marketing risk identification model, two identification results will be output, namely the first identification result and the second identification result. Finally, according to the first identification result and the second identification result, the identification result is calculated and output.

[0073] In one embodiment, the first identification result and the second identification result include complaint risk results of various reasons and corresponding probability coefficients, see Fig. 9 As shown, Fig. 9 yes Figure 8 A specific implementation flow diagram of step S2-23 in the embodiment includes the following steps:

[0074] S2-231: Calculate the probability coefficients corresponding to the complaint risk results for multiple reasons in the identification results according to the probability coefficients corresponding to the complaint risk results for multiple reasons in the first identification result and the second identification result;

[0075] For example, if the first recognition result is the complaint risk of telephone harassment, the corresponding probability coefficient is 0.55; the complaint risk of incomplete notification is 0.05; the complaint risk of poor service attitude of marketing personnel is 0.15; the complaint risk of false information is 0.1; and the corresponding probability coefficient of no complaint is 0.15. If the second recognition result is the complaint risk of telephone harassment, the corresponding probability coefficient is 0.45; the complaint risk of incomplete notification is 0.15; the complaint risk of poor service attitude of marketing personnel is 0.15; the complaint risk of false information is 0.1; and the corresponding probability coefficient of no complaint is 0.15. In one embodiment, the probability coefficients corresponding to the complaint risk results for multiple reasons in the identification result are the average of the sum of the probability coefficients corresponding to the complaint risk results for multiple reasons in the first and second identification results. Then the probability coefficients corresponding to the complaint risk results for multiple reasons in the identification result are: the complaint risk for telephone harassment has a corresponding probability coefficient of 0.5; the complaint risk for incomplete notification has a corresponding probability coefficient of 0.1; the complaint risk for poor service attitude of marketing personnel has a corresponding probability coefficient of 0.15; the complaint risk for false information has a corresponding probability coefficient of 0.1; and the corresponding probability coefficient for no complaint is 0.15.

[0076] S2-232: Calculate the complaint risk level according to the probability coefficients corresponding to the complaint risk results of various reasons in the identification results and the predefined complaint risk level determination rules, and determine whether there is a complaint risk;

[0077] For example, in one implementation, the predefined complaint risk level determination rule may be that if the total probability coefficient of the complaint risk label in the identification result is greater than 0.8, it is a high risk; if the total probability coefficient of the complaint risk label is between 0.3 and 0.8, it is a medium risk; if the total probability coefficient of the complaint risk label is less than 0.3, it is a low risk, indicating that there is no complaint risk. In the above example, since the total probability coefficient of the complaint risk label is 0.85, it is a high risk and there is a complaint risk.

[0078] S2-233: If there is a complaint risk, determine the complaint cause according to the probability coefficients corresponding to the complaint risk results of multiple reasons in the identification result and the predefined complaint cause determination rules, and output the complaint risk level and the complaint cause;

[0079] For example, the predefined complaint reason determination rule is to determine the top two reasons in the probability coefficients corresponding to the complaint risk results of the multiple reasons as the complaint reasons. In the above example, the complaint reasons are telephone harassment and poor service attitude of marketing personnel. At the same time, the complaint risk level is output as high risk.

[0080] S2-234: Otherwise, output the complaint risk level.

[0081] Compared with the prior art, the recognition results output by the marketing risk recognition model in the embodiment of the present application include not only the complaint risk level, but also the specific reasons for the complaint. Therefore, after receiving the model recognition results, the marketing personnel can adjust the marketing policy according to the reasons for the complaint and the actual situation of the users.

[0082] In the embodiment of the present application, the diversified data set including dialogue text data and structured data is self-learned by the model, which can more accurately identify the customer's complaint tendency compared to learning only structured data. At the same time, the training data set of the sample users in the embodiment of the present application is pre-labeled with multiple category labels, wherein the category labels include one or more labels of complaint risk of telephone harassment, complaint risk of incomplete notification, complaint risk of poor service attitude of marketing personnel, and complaint risk of false information. That is, the marketing risk identification model in the embodiment of the present application can perform self-learning training on the data set of multiple complaint reasons, and the marketing risk identification model generated at the end can not only identify whether there is a complaint risk, but also identify the specific complaint risk reasons. The recognition results output by the marketing risk identification model are more comprehensive and more accurate. Finally, the embodiment of the present application generates two types of data sets by performing two types of feature processing on dialogue text data and structured data, and trains and generates two marketing risk identification models in turn according to the two types of data sets. Finally, the two marketing risk identification models are used to identify the risk of marketing behavior, which can solve the technical problem of low prediction accuracy of a single model and further improve the stability of the model.

[0083] It should be understood that the order of execution of the steps in the above embodiments does not necessarily mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. In addition, the term "including" and its variants should be interpreted as an open term of "including but not limited to".

[0084] In one embodiment, a marketing risk identification model training device is provided, and the training device corresponds to the marketing risk identification model training method in the above embodiment. Fig.10 As shown, the training device includes: a data set acquisition module 201, a feature vector generation module 202, a first feature processing module 203 and a second feature processing module 204 and a model training generation module 205. The functional modules are described in detail as follows:

[0085] The data set acquisition module 201 is used to acquire a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels and includes conversation text data and structured data.

[0086] Feature vector generation module 202: used for performing data preprocessing on the conversation text data and the structured data respectively to generate a conversation text data feature vector and a structured data feature vector.

[0087] The first feature processing module 203 is used to generate a first training data set of sample users after performing a first feature processing on the conversation text data feature vector and the structured data feature vector.

[0088] The second feature processing module 204 is used to generate a second training data set of sample users after performing a second feature processing on the conversation text data feature vector and the structured data feature vector.

[0089] Model training generation module 205: used to generate a marketing risk identification model by performing training based on the first training data set and the second training data set and based on a classification algorithm.

[0090] A possible implementation method is provided in an embodiment of the present application. The first feature processing module 203 is further configured to:

[0091] Obtaining the dialogue text data feature vector and the structured data feature vector;

[0092] After feature vector concatenation of the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated.

[0093] A possible implementation method is provided in an embodiment of the present application, wherein the second feature processing module 204 is further configured to:

[0094] Obtaining the dialogue text data feature vector and the structured data feature vector;

[0095] Based on a preset weight coefficient, the feature vector of the conversation text data and the feature vector of the structured data are feature-fused to generate a second training data set of the sample user.

[0096] A possible implementation method is provided in an embodiment of the present application. The model training generation module 205 is further used to:

[0097] Inputting the first training data set into the classification algorithm model for training until convergence, thereby generating a first marketing risk identification model;

[0098] The second training data set is input into the classification algorithm model for training until convergence to generate a second marketing risk identification model.

[0099] In another embodiment, a marketing risk identification device is provided, and the identification device corresponds one-to-one with the marketing risk identification method in the above embodiment. Fig.11 As shown, the identification device includes: a data set acquisition module 301, a risk identification and identification result output module 302. The detailed description of each functional module is as follows:

[0100] The data set acquisition module 301 is used to acquire the data set of the target user.

[0101] The risk identification and identification result output module 302 is used to input the data set into the marketing risk identification model for risk identification and output the identification result, wherein the marketing risk identification model is trained according to the above-mentioned marketing risk identification model training device.

[0102] In an embodiment of the present application, a possible implementation is provided, wherein the marketing risk identification model includes a first marketing risk identification model and a second marketing risk identification model, and the risk identification and identification result output module 302 is further used to:

[0103] Based on the first identification model of marketing risks, output the first identification result;

[0104] Based on the second identification model of marketing risk, output the second identification result;

[0105] A recognition result is calculated and outputted based on the first recognition result and the second recognition result.

[0106] In an embodiment of the present application, a possible implementation method is provided, wherein the first identification result and the second identification result include complaint risk results of various reasons and corresponding probability coefficients, and the risk identification and identification result output module 302 is further used to:

[0107] Calculate the probability coefficients corresponding to the complaint risk results for multiple reasons in the identification results according to the probability coefficients corresponding to the complaint risk results for multiple reasons in the first identification result and the second identification result;

[0108] The complaint risk level is calculated based on the probability coefficients corresponding to the complaint risk results of various reasons in the identification results and the predefined complaint risk level determination rules, and it is determined whether there is a complaint risk;

[0109] If there is a complaint risk, determine the complaint cause according to the probability coefficients corresponding to the complaint risk results of multiple reasons in the identification result and the predefined complaint cause determination rules, and output the complaint risk level and the complaint cause;

[0110] Otherwise, the complaint risk level is output.

[0111] In the present application, first, a training data set of sample users is obtained, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data; then, the conversation text data and the structured data are respectively preprocessed to generate a conversation text data feature vector and a structured data feature vector; secondly, the conversation text data feature vector and the structured data feature vector are subjected to a first feature processing to generate a first training data set of sample users; thirdly, the conversation text data feature vector and the structured data feature vector are subjected to a second feature processing to generate a second training data set of sample users; finally, according to the first training data set and the second training data set, training is performed based on a classification algorithm to generate a marketing risk identification model.

[0112] This application, through the model self-learning including dialogue text data and structured data of the diversified data set, compared with only learning structured data, can more accurately identify the customer's complaint tendency. At the same time, the training data set of the sample users in this application is pre-labeled with multiple category labels, wherein the category labels include one or more labels of complaint risk of telephone harassment, complaint risk of incomplete notification, complaint risk of poor service attitude of marketing personnel, and complaint risk of false information. That is, the marketing risk identification model in this application can self-learn and train the data set of multiple complaint reasons, and the marketing risk identification model generated at the end can not only identify whether there is a complaint risk, but also identify the specific complaint risk reasons. The recognition results output by the marketing risk identification model are more comprehensive and more accurate. Finally, this application generates two types of data sets by processing the features of dialogue text data and structured data in two ways, and trains and generates two marketing risk identification models in turn according to the two types of data sets. Finally, the two marketing risk identification models are used to identify the risk of marketing behavior, which can solve the technical problem of low prediction accuracy of a single model and further improve the stability of the model.

[0113] For the specific limitations of the marketing risk identification model training device and the marketing risk identification device, please refer to the limitations of the marketing risk identification model training method and the marketing risk identification method mentioned above, which will not be repeated here. The various modules in the above-mentioned marketing risk identification model training device and the marketing risk identification device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0114] In one embodiment, a computer device is provided, whose internal structure diagram can be as follows: Fig.12As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements a marketing risk identification model training method or a marketing risk identification method service-side function or step.

[0115] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0116] Obtaining a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data;

[0117] After performing data preprocessing on the dialogue text data and the structured data respectively, a dialogue text data feature vector and a structured data feature vector are generated;

[0118] After performing a first feature processing on the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated;

[0119] After performing a second feature processing on the conversation text data feature vector and the structured data feature vector, a second training data set of sample users is generated;

[0120] According to the first training data set and the second training data set, training is performed separately based on a classification algorithm to generate a marketing risk identification model.

[0121] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0122] Obtain the target user's data set;

[0123] The data set is input into a marketing risk identification model for risk identification, and an identification result is output, wherein the marketing risk identification model is trained according to the above-mentioned marketing risk identification model training method.

[0124] It should be noted that the above functions or steps that can be implemented by the computer device can refer to the relevant description in the aforementioned method embodiment, and to avoid repetition, they will not be described one by one here.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0126] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0127] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A marketing risk identification model training method, characterized in that: include: Obtaining a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data; After performing data preprocessing on the dialogue text data and the structured data respectively, a dialogue text data feature vector and a structured data feature vector are generated; After performing a first feature processing on the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated; After performing a second feature processing on the conversation text data feature vector and the structured data feature vector, a second training data set of sample users is generated; According to the first training data set and the second training data set, training is performed separately based on a classification algorithm to generate a marketing risk identification model.

2. The method according to claim 1, characterized in that After performing the first feature processing on the conversation text data feature vector and the structured data feature vector, generating a first training data set of sample users includes: Obtaining the dialogue text data feature vector and the structured data feature vector; After feature vector concatenation of the conversation text data feature vector and the structured data feature vector, a first training data set of sample users is generated.

3. The method according to claim 1, characterized in that After performing the second feature processing on the conversation text data feature vector and the structured data feature vector, generating a second training data set of sample users, comprises: Obtaining the dialogue text data feature vector and the structured data feature vector; Based on a preset weight coefficient, the feature vector of the conversation text data and the feature vector of the structured data are feature-fused to generate a second training data set of the sample user.

4. The method according to claim 3, characterized in that The generating of the marketing risk identification model by training the first training data set and the second training data set based on a classification algorithm includes: Inputting the first training data set into the classification algorithm model for training until convergence, thereby generating a first marketing risk identification model; The second training data set is input into the classification algorithm model for training until convergence to generate a second marketing risk identification model.

5. A marketing risk identification method, characterized in that: include: Obtain the target user's data set; The data set is input into a marketing risk identification model for risk identification, and an identification result is output, wherein the marketing risk identification model is trained according to the marketing risk identification model training method according to any one of claims 1 to 4.

6. The method according to claim 5, characterized in that The marketing risk identification model includes a first marketing risk identification model and a second marketing risk identification model. The data set is input into the marketing risk identification model for risk identification, and the identification result is output, including: Based on the first identification model of marketing risks, output the first identification result; Based on the second identification model of marketing risk, output the second identification result; A recognition result is calculated and outputted based on the first recognition result and the second recognition result.

7. The method according to claim 6, characterized in that The first identification result and the second identification result include complaint risk results of various reasons and corresponding probability coefficients. The identification result is calculated and outputted based on the first identification result and the second identification result, including: Calculate the probability coefficients corresponding to the complaint risk results for multiple reasons in the identification results according to the probability coefficients corresponding to the complaint risk results for multiple reasons in the first identification result and the second identification result; The complaint risk level is calculated based on the probability coefficients corresponding to the complaint risk results of various reasons in the identification results and the predefined complaint risk level determination rules, and it is determined whether there is a complaint risk; If there is a complaint risk, determine the complaint cause according to the probability coefficients corresponding to the complaint risk results of multiple reasons in the identification result and the predefined complaint cause determination rules, and output the complaint risk level and the complaint cause; Otherwise, the complaint risk level is output.

8. A marketing risk identification model training device, characterized in that: include: Dataset acquisition module: used to acquire a training data set of sample users, wherein the training data set is pre-labeled with multiple category labels, including conversation text data and structured data; A feature vector generating module: used for generating a feature vector of the dialogue text data and a feature vector of the structured data after performing data preprocessing on the dialogue text data and the structured data respectively; A first feature processing module: used for performing a first feature processing on the feature vector of the conversation text data and the feature vector of the structured data to generate a first training data set of sample users; A second feature processing module: used for performing a second feature processing on the conversation text data feature vector and the structured data feature vector to generate a second training data set of sample users; Model training generation module: used to generate a marketing risk identification model by performing training based on the first training data set and the second training data set and based on a classification algorithm.

9. A marketing risk identification device, characterized in that: include: Dataset acquisition module: used to obtain the target user's dataset; Risk identification and identification result output module: used to input the data set into the marketing risk identification model for risk identification and output the identification result, wherein the marketing risk identification model is trained according to the marketing risk identification model training device according to claim 8.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the method according to any one of claims 1 to 4 or the method according to any one of claims 5 to 7.

Images