Method for predicting purchase willingness of customer and related equipment thereof
By obtaining and analyzing the type of speech of the call role, using preset models to predict customers' purchasing intentions, solving the one-sided problem of keyword prediction results, and achieving more accurate predictions of customers' purchasing intentions.
Patent Information
- Application Number
- CN202411193734.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-12-03
AI Technical Summary
The existing technology only predicts customers' purchasing intentions through keywords, resulting in the prediction results being more one-sided and cannot accurately reflect the customers' real purchasing intentions.
By obtaining the target call content of the target call role, the preset speech type is determined using the preset speech classification model, and the customer's purchasing intention is predicted based on the preset intention prediction model, considering the influence of the speech types of different call roles on purchasing intention.
It improves the accuracy of customer purchase intention prediction, overcomes the one-sidedness of keyword prediction, and can more accurately reflect the customer's true purchasing intention.
Smart Images

Figure CN119067710A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method for predicting customer purchase intention and related equipment. Background Art
[0002] Currently, key words are usually extracted from historical call records between customer service and customers, and the customer's purchasing intention is predicted based on the key words.
[0003] However, although keywords can reflect customers' purchasing intention to a certain extent, predicting customers' purchasing intention only through keywords makes the prediction results rather one-sided, resulting in the inability to accurately predict customers' real purchasing intention. Summary of the invention
[0004] The main purpose of this application is to provide a method for predicting customer purchase intention and related equipment, aiming to solve the technical problem that predicting customer purchase intention only by key words makes the prediction result rather one-sided, resulting in the inability to accurately predict the customer's true purchase intention.
[0005] To achieve the above purpose, the present application proposes a method for predicting customer purchase intention, wherein the method comprises:
[0006] Obtain target call content of a target call role, where the target call role includes a marketing customer service representative and / or a marketing customer;
[0007] Determine the target speech type of the speech used in the target call content by a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them;
[0008] Based on the target speech type, the marketing customer's purchase intention is predicted by a preset intention prediction model to obtain the target purchase intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchase intention, and the speech types used by different call roles have different weights of influence on the customer's purchase intention.
[0009] In one embodiment, before the step of obtaining the target call content of the target call role, the method further includes:
[0010] Acquire a training sample and a sample label of the training sample, wherein the training sample includes a first call content of a first call role, and the sample label of the training sample includes a first speech type label of a speech used by the first call role;
[0011] Based on the first call content and the first speech type label, the preset first model to be trained is iteratively trained to obtain a speech classification model.
[0012] In one embodiment, the step of iteratively training the preset first model to be trained based on the first call content and the first speech type label to obtain the speech classification model includes:
[0013] Based on the first call content, pre-training a preset large model to obtain a first content analysis model, wherein the content analysis model has the ability to understand the call content in the call text data;
[0014] Adding a preset linear classification layer to the first content analysis model to obtain a second content analysis model, wherein the preset linear classification layer is used to perform a speech classification task;
[0015] Based on the first call content and the first speech type label, the second content analysis model is fine-tuned to obtain a speech classification model.
[0016] In one embodiment, the step of fine-tuning the second content analysis model based on the first call content and the first speech type label to obtain the speech classification model includes:
[0017] Inputting the first call content and the first speech type label into the second content analysis model;
[0018] Determining, from the first call content, by using the second content analysis model, a first probability that the speech used by the first call character is of a preset speech type;
[0019] Based on the first probability and the first speech type label, a first loss value is calculated by using a preset loss function;
[0020] Based on the first loss value, the model parameters of the second content analysis model are updated, and the step of inputting the first call content and the first speech type label into the second content analysis model is returned to continue iterative training until a speech classification model that meets the preset first accuracy condition is obtained.
[0021] In one embodiment, the first speech type includes a plurality of preset speech types, each of which is respectively set for the first conversation role and / or other conversation roles except the first conversation role, and the step of calculating the first loss value by using a preset loss function based on the first probability and the first speech type label includes:
[0022] Based on the first probability and the first speech type label, a second loss value corresponding to each of the preset speech types is calculated by a preset loss function;
[0023] Calculating a first loss value based on the second loss value and the first attention mask;
[0024] Among them, the first attention mask is used to decide to ignore the influence of the preset speech type corresponding to the other calling roles on the first loss value.
[0025] In one embodiment, after the step of iteratively training the preset first to-be-trained model based on the first call content and the first speech type label to obtain the speech classification model, the step further includes:
[0026] Acquire a test sample and a sample label of the test sample, wherein the test sample includes the second call content of the second call role, and the sample label of the test sample includes a second speech type label of the speech used by the second call role;
[0027] Based on the second call content, the speech classification model is tested to obtain a test result, wherein the test result includes a second probability that the speech used by the second call role is a preset speech type;
[0028] Based on the second probability and the second speech type label, calculating a third loss value by using a preset loss function;
[0029] Based on the third loss value, the model parameters of the second content analysis model are updated, and the step of testing the speech classification model based on the second call content to obtain the test result is returned, and iterative training is continued until a speech classification model that meets the preset second accuracy condition is obtained.
[0030] In one embodiment, the second speech type includes a plurality of preset speech types, and the step of calculating the third loss value by a preset loss function based on the second probability and the second speech type label includes:
[0031] Based on the second probability and the second speech type label, a fourth loss value corresponding to each of the preset speech types is calculated by a preset loss function;
[0032] Calculating a third loss value based on the fourth loss value and the second attention mask;
[0033] The second attention mask is used to decide to ignore the influence of other speech types on the third loss value, and the other speech types are speech types in the second speech type except for the easily confused speech types, and the easily confused speech types are determined based on the second probability and the second speech type label.
[0034] In one embodiment, before the step of obtaining historical call text data, the method further includes:
[0035] Obtaining a historical speech type of a speech used by a first call role in historical call content, wherein the first call role includes a historical customer service representative and / or a historical customer;
[0036] Based on the historical speech type, determine the purchase intention label;
[0037] Based on the historical speech types and purchase intention labels, the preset second model to be trained is iteratively trained to obtain the preset intention prediction model.
[0038] In one embodiment, the preset second model to be trained includes a binary classification model to be trained, and the step of iteratively training the preset second model to be trained based on the historical speech type and the purchase intention label to obtain the preset intention prediction model includes:
[0039] Inputting the historical speech type and purchase intention label into the binary classification model to be trained;
[0040] Predicting the purchase intention of the historical customers through the binary classification model to be trained to obtain the historical purchase probability;
[0041] Calculating the error between the historical purchase probability and the purchase intention label to obtain an error calculation result;
[0042] Based on the error calculation result, the model parameters of the binary classification model to be trained are updated, and the step of inputting the historical speech type and purchase intention label into the binary classification model to be trained is returned until a preset intention prediction model that meets the preset third precision condition is obtained.
[0043] In addition, to achieve the above-mentioned purpose, the present application also proposes a device for predicting customer purchase intention, the device for predicting customer purchase intention comprising:
[0044] An acquisition module, used to acquire target call content of a target call role, wherein the target call role includes a marketing customer service representative and / or a marketing customer;
[0045] A speech classification module, used to determine the target speech type of the speech used in the target call content through a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them;
[0046] The intention prediction module is used to predict the purchasing intention of the marketing customer based on the target speech type through a preset intention prediction model to obtain the target purchasing intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchasing intention, and the speech types used by different call roles have different influence weights on the customer's purchasing intention.
[0047] In addition, to achieve the above-mentioned purpose, the present application also proposes a device for predicting customer purchasing intention, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the method for predicting customer purchasing intention as described above.
[0048] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for predicting customer purchasing intention as described above are implemented.
[0049] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the method for predicting customer purchase intention as described above.
[0050] One or more technical solutions proposed in this application have at least the following technical effects:
[0051] Compared with the existing technology that extracts key words from the historical call records between customer service and customers and predicts the customer's purchase intention based on the key words, the technical means adopted in this application are:
[0052] First, the target call content of the target call role is obtained, specifically, the target call content of the marketing customer service and / or the marketing customer.
[0053] Secondly, since the preset speech classification model can characterize the correlation between the call content of different call roles and the speech types of the speech used by the corresponding call roles, that is, the preset speech classification model can analyze the call content of different call roles to obtain the speech types of the speech used by the corresponding call roles, therefore, the preset speech classification model can accurately determine the target speech type of the speech used in the target call content.
[0054] Finally, since the preset intention prediction model can characterize the correlation between the types of speech used by different call roles and the customer's purchasing intention, that is, the preset intention prediction model can accurately predict the customer's purchasing intention based on the target speech type used by each call role. Therefore, based on the target speech type used by the target call role in the target call content, the purchasing intention of the marketing customer is predicted by the preset intention prediction model, and the target purchasing intention of the marketing customer can be accurately predicted.
[0055] It should be noted that since the types of speeches used by different call roles have different weights of influence on the customer's purchasing intention, that is, assuming that the types of speeches used by customer service are the same as those used by customers, for the same type of speeches, the intensity of customer purchasing intention represented by the speeches spoken by customer service and by customers is different; therefore, the present application determines the target speech type of the speeches used by the target call role in the target call content through a preset speech classification model, and analyzes the target speech type through a preset intention prediction model, so as to accurately predict the purchasing intention of marketing customers. If the customer's purchasing intention is predicted only based on the key words in the historical call text data, the different intensities of customer purchasing intention represented by the key words in different call roles are ignored.
[0056] Based on the above, the above technical means of the present application overcomes the defect in the related art that the customer's purchasing intention is predicted only by key words, resulting in a relatively one-sided prediction result and an inability to accurately predict the customer's true purchasing intention, thereby improving the prediction accuracy of the customer's purchasing intention. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0058] In order to more clearly illustrate the embodiments of the present application 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0059] Figure 1 A flowchart diagram of the first embodiment of the method for predicting the purchase intention of customers in this application;
[0060] Figure 2 A schematic diagram of a scenario provided for the first embodiment of the method for predicting the purchase intention of customers of this application;
[0061] Figure 3A flow chart of the second embodiment of the method for predicting the purchase intention of customers in this application;
[0062] Figure 4 A schematic diagram of a scenario provided for Embodiment 2 of the method for predicting customer purchase intention of this application;
[0063] Figure 5 A flowchart diagram of the third embodiment of the method for predicting the purchase intention of customers in this application;
[0064] Figure 6 A schematic diagram of the module structure of a device for predicting customer purchase intention according to an embodiment of the present application;
[0065] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the method for predicting customer purchasing intention in the embodiment of the present application.
[0066] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0067] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.
[0068] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.
[0069] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a device for predicting customer purchase intention, etc. The following takes the device for predicting customer purchase intention as an example to illustrate this embodiment and the following embodiments.
[0070] Based on this, the present application embodiment provides a method for predicting customer purchase intention, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the method for predicting customer purchase intention of the present application.
[0071] In this embodiment, the method for predicting customer purchase intention includes steps S10 to S40:
[0072] Step S10, obtaining target call content of a target call role, wherein the target call role includes a marketing customer service representative and / or a marketing customer;
[0073] It should be noted that the target call content can be obtained from historical call text data. The historical call text data can be obtained by processing the call records in the telephone marketing process. Since the call records are voice data, the call records can be converted into historical call text data through voice recognition technology (Automatic Speech Recognition, ASR). At the same time, in order to facilitate the subsequent processing of the call content of different call roles separately, the call content belonging to different call roles in the call records can be distinguished based on the different timbres in the call records, so as to obtain the target call content of the target call role, for example, the call content corresponding to the marketing customer service and marketing customers in the historical call text data can be obtained.
[0074] Among them, the historical call text data may include the target call content corresponding to the marketing customer service and the marketing customers respectively, or may also include the target call content between other customers and other non-marketing personnel, etc.; therefore, the target call role can be any role that has a call with the marketing customer; the target call content can be the call content corresponding to the target call role in the historical call text data, and can be multiple sentences corresponding to the target call role.
[0075] Step S20, determining the target speech type of the speech used in the target call content by a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them;
[0076] Currently, key words are usually extracted from the historical call records between customer service and customers, and the customer's purchasing intention is predicted based on the key words; however, although key words can reflect the customer's purchasing intention to a certain extent, predicting the customer's purchasing intention only through key words makes the prediction result rather one-sided, resulting in the inability to accurately predict the customer's true purchasing intention.
[0077] This embodiment determines the target speech type of the speech used by the target call role in the target call content through a preset speech classification model, and can analyze the types of speech used by different call roles in the historical call text data respectively. Since each call role uses a variety of speech during the call process, different speech represents different meanings. For the same type of speech, when spoken by different call roles, the meaning represented will change.
[0078] Assuming that the type of speech used by customer service is the same as the type of speech used by customers, for the same type of speech, the intensity of the customer's willingness to buy that can be represented by the speech is different; therefore, this embodiment distinguishes the call content in the historical call text data based on the different call roles, and determines the target speech type of the speech used by the target call role through a preset speech classification model.
[0079] Among them, the preset speech classification model can be obtained by iterative training based on the call content of different call roles and the speech type labels of the speech used in the call content. It can be understood that the trained preset speech classification model can characterize the correlation between the call content of different call roles and the speech types of the speech used by them. Therefore, by determining the target speech type of the speech used by the target call role through the preset speech classification model, the accuracy of the target speech type can be guaranteed.
[0080] Specifically, the scripts used by different call roles may include opening remarks, product introduction, purchase intention inquiry, product transaction cost introduction, appointment for next call, closing remarks, thank you remarks, product and price consultation, company introduction, interest rate introduction, customer retention, etc.
[0081] It should be noted that among the target speech types of the above-mentioned speech used by different call roles, the target speech types used by marketing customer service can be divided into the following types: greetings, inquiries, sales promotion, customer retention, etc.; the target speech types used by marketing customers can be divided into the following types: inquiries, refusal of sales promotion, willingness to continue to understand, etc.
[0082] Step S30: Based on the target speech type, the purchasing intention of the marketing customer is predicted by a preset intention prediction model to obtain the target purchasing intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchasing intention, and the speech types used by different call roles have different influence weights on the customer's purchasing intention.
[0083] It can be understood that, for the inquiry type of language, the inquiry language spoken by the marketing customer service cannot directly reflect the customer's willingness to buy, but the inquiry language spoken by the marketing customer can reflect that the customer has a strong desire to understand; that is, the inquiry language types used by different call roles have different weights of influence on the customer's willingness to buy.
[0084] Therefore, iterative training can be performed based on the types of speech used by different call roles and the customer's purchase intention labels. The preset intention prediction model obtained through training can characterize the correlation between the types of speech used by different call roles and the customer's purchase intention. Therefore, based on the target speech type used by the target call role in the target call content, the purchase intention of the marketing customer is predicted through the preset intention prediction model, thereby ensuring the accuracy of the target purchase intention.
[0085] Specifically, before the step of obtaining the historical call text data, the historical speech type of the speech used by the first call role in the historical call content can be obtained, and the first call role includes a historical customer service and / or a historical customer; based on the historical speech type, the purchase intention label is determined;
[0086] Based on the historical speech types and purchase intention labels, the preset second model to be trained is iteratively trained to obtain the preset intention prediction model.
[0087] It should be noted that the historical speech type of the speech used by the first call role in the historical call content can be obtained by analyzing the historical call content of the historical call role through the above-mentioned preset speech classification model; specifically, it can be obtained by analyzing the historical call content of historical customer service and / or historical customers through the above-mentioned preset speech classification model.
[0088] Furthermore, in order to inform the model of the content that should be output based on the historical speech type, a purchase intention label corresponding to the historical speech type may be annotated. Specifically, the purchase intention label includes the purchase probability of the historical customer.
[0089] The specific implementation method of determining the purchase intention label based on the historical speech type may be: based on the historical speech type, a logistic regression algorithm or a decision tree algorithm is used to predict the purchase probability of the historical customer, the purchase probability of the historical customer is between 0 and 1, and the purchase probability of the historical customer is used as the purchase intention label, such as Figure 2 shown.
[0090] Furthermore, based on the historical speech types and purchase intention labels, the preset second model to be trained is iteratively trained to obtain the preset intention prediction model.
[0091] Specifically, the preset second model to be trained includes a binary classification model to be trained or a multi-classification model to be trained. The specific implementation method of iteratively training the preset second model to be trained based on the historical speech type and the purchase intention label to obtain the preset intention prediction model may be:
[0092] Input the historical speech type and purchase intention label into the binary classification model to be trained; predict the purchase intention of the historical customers through the binary classification model to be trained to obtain the historical purchase probability; calculate the error between the historical purchase probability and the purchase intention label to obtain the error calculation result; based on the error calculation result, update the model parameters of the binary classification model to be trained, and return to the step of inputting the historical speech type and purchase intention label into the binary classification model to be trained, until a preset intention prediction model that meets the preset third precision condition is obtained.
[0093] Specifically, before iteratively training the preset second model to be trained based on the historical speech types and purchase intention labels, the format of the historical speech types may also be converted.
[0094] Furthermore, the format-converted historical speech types and purchase intention labels are input into the binary classification model to be trained; the binary classification model to be trained analyzes the correspondence between the historical speech types and the purchase intentions of historical customers to obtain the predicted purchase intentions of historical customers, that is, the historical purchase probability of the historical customers is obtained.
[0095] Furthermore, the above-mentioned iterative training process also includes: calculating the error between the historical purchase probability and the purchase intention label to obtain an error calculation result; judging whether the error calculation result meets the error standard indicated by a preset error threshold range; wherein the preset error threshold includes a preset mean square error threshold, and the smaller the mean square error threshold is, the more accurate the model is considered to be, therefore, judging whether the mean square error calculation result is less than the preset mean square error threshold, and if so, it is considered that the error calculation result meets the error standard indicated by the preset error threshold range.
[0096] Furthermore, the above-mentioned iterative training process also includes: if the above-mentioned error calculation result does not meet the error standard indicated by the preset error threshold range, then the model parameters of the binary classification model to be trained are updated, and the step of inputting the historical speech type and purchase intention label into the binary classification model to be trained is returned, and the training is stopped until the error calculation result meets the error standard indicated by the preset error threshold range, and a willingness prediction model that meets the preset third precision condition is obtained; that is, through iterative training, the binary classification model to be trained is converged until the error calculation result meets the error standard indicated by the preset error threshold range, and the iterative training is completed to obtain the willingness prediction model that meets the preset third precision condition, thereby improving the accuracy of the willingness prediction model in predicting customer purchase intention.
[0097] Based on the above, since the types of speech used by different call roles have different weights of influence on the customer's purchasing intention, this embodiment obtains the target call content of the target call role, accurately determines the target speech type of the speech used in the target call content through a preset speech classification model, and predicts the purchasing intention of the marketing customer through a preset intention prediction model, and accurately predicts the target purchasing intention of the marketing customer, thereby overcoming the defect of the related technology that the customer's purchasing intention is predicted only by keywords, resulting in a relatively one-sided prediction result and an inability to accurately predict the customer's true purchasing intention, thereby improving the prediction accuracy of the customer's purchasing intention.
[0098] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 Before step S10, the method for predicting customer purchase intention further includes steps S01 to S02:
[0099] Step S01, obtaining a training sample and a sample label of the training sample, wherein the training sample includes a first call content of a first call role, and the sample label of the training sample includes a first speech type label of a speech used by the first call role;
[0100] It should be noted that since the types of words used by different roles in the call content have different influence weights on the customer's purchasing intention, this embodiment uses the first call content of the first call role as a training sample, and uses the first type of words used by the first call role in the first call content as a sample label.
[0101] It can be understood that there may be multiple speech scripts used in the first call content corresponding to each call role, and there may be multiple corresponding speech script types. Therefore, there may be multiple first speech script type labels for the speech scripts used by the first call role in the first call content.
[0102] The first call role includes historical customer service and / or historical customers.
[0103] Specifically, the first speech type label is used to characterize the type of speech used by the first call role in the first call content. For example, when the historical customer service uses speech such as retaining the customer or requesting the customer to keep talking in the first call content, the first speech type label corresponding to the speech may be retaining the customer. When the historical customer uses speech such as not needing the product or that the product is too expensive in the first call content, the first speech type label corresponding to the speech may be rejecting marketing.
[0104] It is understandable that the words used by customer service in the call content can correspond to multiple different words types, and the words used by customers in the call content can also correspond to multiple different words types. Figure 4 ; Assume that the types of words used by customer service in the call content correspond to 5 different word type labels m1-m5, and the types of words used by customers in the call content can also correspond to 5 different word type labels m6-m10.
[0105] Step S02: Based on the first call content and the first speech type label, iteratively train the preset first model to be trained to obtain a speech classification model.
[0106] Based on the above-mentioned first call content and the first speech type label, the preset first model to be trained is iteratively trained to obtain a speech classification model.
[0107] Among them, the preset first model to be trained can be a preset large model (Large Model) or a preset basic model. The preset large model refers to a machine learning model containing a large number of parameters and computing resources, and the preset basic model refers to a model with certain language analysis capabilities.
[0108] Specifically, the specific implementation of iteratively training the preset first to-be-trained model based on the first call content and the first speech type label to obtain the speech classification model may be:
[0109] Based on the first call content, a preset large model is pre-trained to obtain a first content analysis model, and the content analysis model has the ability to understand the call content in the call text data; a preset linear classification layer is added to the first content analysis model to obtain a second content analysis model, wherein the preset linear classification layer is used to perform a speech classification task; based on the first call content and the first speech type label, the second content analysis model is fine-tuned to obtain a speech classification model.
[0110] It should be noted that in order to fully fit the intention of the target speech type used by the target call role output by this embodiment, the preset large model can be pre-trained based on the first call content to obtain a content analysis model, and the content analysis model has the ability to understand the call content in the call text data; then, a preset linear classification layer for performing speech classification tasks is added to the first content analysis model to obtain a second content analysis model; based on the first call content and the first speech type label, the second content analysis model is fine-tuned to obtain a speech classification model.
[0111] Specifically, this approach utilizes the powerful learning and expression capabilities of the preset large model, which improves the accuracy of understanding the different speech techniques in the call content, thereby improving the accuracy of determining the speech types to which different speech techniques belong.
[0112] It should be noted that the number of first call contents used to train the preset large model is huge, for example, the order can be more than one million; the number of first call contents used to fine-tune the content analysis model is relatively small, for example, it can be 1,000 first call text data or 10,000 first call contents, etc.
[0113] Among them, pre-training enables the preset large model to learn general, cross-domain, and conversational language representations without labels.
[0114] It can be understood that after pre-training the preset large model based on the first call content, a first content analysis model capable of understanding the call content in the call text data can be obtained.
[0115] Furthermore, by adding a preset linear classification layer for performing the task of speech classification to the first content analysis model, a second content analysis model is obtained; wherein the preset linear classification layer is capable of performing the task of classifying the speech.
[0116] Thus, based on the first call content and the first speech type label, the second content analysis model is fine-tuned, and on the basis of pre-training, for specific tasks (for example, speech classification tasks), the second content analysis model can be further trained on training samples with sample labels to enable the second content analysis model to adapt to specific task requirements; in this process, some or all parameters of the second content analysis model with a preset linear classification layer added will be adjusted to achieve the purpose of fine-tuning, so that the preset speech classification model obtained after fine-tuning can accurately output the target speech type of the speech used by the target call role according to the first call content.
[0117] Specifically, the preset linear classification layer may be a fully connected layer.
[0118] In a feasible implementation, the first call content and the first speech type label can be input into a second content analysis model, content analysis processing can be performed based on the second content analysis model, and the historical speech type of the speech used by the first call character can be output. Based on the error between the historical speech type and the first speech type label, the model parameters of the content analysis model can be adjusted, and the step of inputting the first call content and the first speech type label into the second content analysis model can be returned until a preset speech classification model that meets the preset requirements is obtained.
[0119] In addition, the specific implementation of fine-tuning the second content analysis model based on the first call content and the first speech type label to obtain the speech classification model may also be:
[0120] The first call content and the first speech type label are input into the second content analysis model; the first probability that the speech used by the first call character is of a preset speech type is determined from the first call content by the second content analysis model; based on the first probability and the first speech type label, a first loss value is calculated by a preset loss function; based on the first loss value, the model parameters of the second content analysis model are updated, and the step of inputting the first call content and the first speech type label into the second content analysis model is returned to continue iterative training until a speech classification model that meets the preset first accuracy condition is obtained.
[0121] Specifically, the first call content and the first speech type label can be input into the second content analysis model, and the second content analysis model can be used to determine from the first call content that the speech used by the first call role is a preset speech type ( m 1— m 10)'s first probability; for example, the probability that the first conversation role uses the speech m1 is 0.8, ..., the probability that the first conversation role uses the speech m1 0 is 0.2, etc.
[0122] It should be noted that the first speech type includes multiple preset speech types, each of which is set for the first conversation role and / or other conversation roles except the first conversation role; it can be understood that the first conversation role corresponds to multiple preset speech types m1-m5, and other conversation roles except the first conversation role correspond to multiple preset speech types m6-m7. m 10.
[0123] Furthermore, based on the first probability and the first speech type label, a first loss value can be calculated by using a preset loss function.
[0124] Specifically, the expression of the preset loss function can be:
[0125] l(x, y) = L = {l1, ..., l N} T , l n =-u n [y n log x n +(1-y n )·log(1-x n )],
[0126] Where N is the number of preset speech types.
[0127] It can be understood that the preset loss function calculates the second loss value l corresponding to each of the preset speech types. n , and then based on the second loss value l corresponding to each preset speech type n The first loss value is calculated; however, the first loss value calculated in this way will be affected by the preset speech type corresponding to the other call roles.
[0128] For example, the probability that the first conversation role uses the speech m1 is 0.8, ..., the probability that the first conversation role uses the speech m10 is 0.2, etc.; because the speech type m10 is not set for the first conversation role, but for other conversation roles except the first conversation role.
[0129] Therefore, in order to ignore the influence of the preset speech types corresponding to the other conversation roles on the first loss value, this embodiment multiplies the transpose L of the matrix formed by the second loss values corresponding to each speech type by the first attention mask mask to calculate the first loss value: loss = L*mask.
[0130] Specifically, the first attention mask is used to decide to ignore the influence of the preset speech types corresponding to the other call roles on the first loss value. The first attention mask can be expressed as: mask = [1, 1, 1, 1, 0, 0, 0, 0, 0], that is, by multiplying the second loss value of the speech category corresponding to m6-m10 by mask and changing it to 0, thereby achieving the purpose of ignoring the influence of the preset speech types corresponding to the other call roles on the first loss value; improving the accuracy of the dialogue classification model in predicting the speech type used by the call role.
[0131] Furthermore, based on the first loss value, the model parameters of the second content analysis model are updated, and the step of inputting the first call content and the first speech type label into the second content analysis model is returned to continue iterative training to obtain a speech classification model that meets the preset first accuracy condition.
[0132] In this embodiment, by setting the loss function to ignore the influence of the preset speech types corresponding to other conversation roles on the first loss value during the model training process, the accuracy of the dialogue classification model in predicting the speech type used by the conversation role is improved.
[0133] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 5, after step S02, further comprising steps A1-A4:
[0134] Step A1, obtaining a test sample and a sample label of the test sample, wherein the test sample includes the second call content of the second call role, and the sample label of the test sample includes a second speech type label of the speech used by the second call role;
[0135] In order to optimize the above-mentioned speech classification model (which may be a speech classification model that meets the preset first accuracy condition), the second call content of the second call role may be used as a test sample, and the test sample may be labeled with a second speech type label, where the second speech type label is used to characterize the speech type used by historical customer service and / or historical customers in the second call content.
[0136] Step A2, based on the second call content, testing the speech classification model to obtain a test result, wherein the test result includes a second probability that the speech used by the second call role is a preset speech type;
[0137] Specifically, the second call content can be input into the speech classification model, and the second call content can be classified by the speech classification model to obtain the test result. It can be understood that the test result includes the second probability that the speech used by the second call role is the preset speech type.
[0138] For example, the test results include that the probability that the second conversation character uses the speech m1 is 0.15, ..., and the probability that the second conversation character uses m10 is 0.9.
[0139] Step A3, calculating a third loss value by using a preset loss function based on the second probability and the second speech type label;
[0140] Furthermore, based on the second probability and the second speech type label, a third loss value can be calculated by using a preset loss function.
[0141] Specifically, the expression of the preset loss function is the same as above, and the fourth loss value corresponding to each of the preset speech types can be calculated separately, and then the third loss value can be calculated based on the fourth loss value corresponding to each preset speech type; however, when there are multiple speech types that are easy to predict errors in the model prediction process, the preset loss function cannot accurately calculate the third loss value.
[0142] Therefore, in this embodiment, based on the second probability and the second speech type label, the third loss value is calculated by using a preset loss function and the following implementation method can be adopted:
[0143] Based on the second probability and the second speech type label, a fourth loss value corresponding to each of the preset speech types is calculated by a preset loss function; based on the fourth loss value and the second attention mask, a third loss value is calculated.
[0144] It should be noted that the second speech type includes multiple preset speech types. For example, the second speech type includes preset speech types m1-m10.
[0145] Specifically, the easily confused speech type can be determined by the second probability that the speech used by the second call character is the preset speech type obtained in the above test process and its consistency with the second speech type label; for example, if more than 80% or more than 60% of the test samples show that the second probability that the speech used by the second call character is m1 is inconsistent with the second speech type label (the second speech type label is m7), then it means that the preset speech types m1 and m7 are easily confused in the prediction process; that is, the easily confused speech types are determined to be the preset speech types m1 and m7.
[0146] Therefore, in order to enable the speech classification model to accurately distinguish easily confused speech types, this embodiment first calculates the fourth loss value corresponding to each preset speech type based on the second probability and the second speech type label through a preset loss function, and calculates the third loss value based on the fourth loss value and the second attention mask.
[0147] The second attention mask is used to decide to ignore the influence of other speech types (speech types in the second speech type except the easily confused speech type) on the third loss value.
[0148] Specifically, the second attention mask can be expressed as: mask = [1,0,0,0,0,0,1,0,0,0], that is, by multiplying the fourth loss value of the speech category corresponding to m2-m6 and m8-m10 by mask and changing it to 0, the purpose of ignoring the influence of other speech types on the third loss value is achieved, and the accuracy of the speech classification model in predicting easily confused speech types is improved.
[0149] Step A4, based on the third loss value, updates the model parameters of the second content analysis model, and returns to the step of testing the speech classification model based on the second call content to obtain the test result, and continues iterative training until a speech classification model that meets the preset second accuracy condition is obtained.
[0150] Furthermore, based on the third loss value, the model parameters of the second content analysis model are updated, and the step of testing the speech classification model based on the second call content to obtain the test result is returned, and iterative training is continued to obtain a speech classification model that meets the preset second accuracy condition.
[0151] It should be noted that the above-mentioned preset first precision condition, preset second precision condition, and preset third precision condition may be that the loss value is less than a preset value or reaches a preset number of iterations, etc.
[0152] In this embodiment, by setting the loss function to ignore the influence of other speech types on the third loss value during the model training process, the accuracy of the speech classification model in predicting easily confused speech types is improved.
[0153] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the method for predicting customer purchase intentions of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.
[0154] This application also provides a device for predicting customer purchase intentions. Figure 6 , the customer purchase intention prediction device comprises:
[0155] An acquisition module 10 is used to acquire target call content of a target call role, where the target call role includes a marketing customer service representative and / or a marketing customer;
[0156] A speech classification module 20, used to determine the target speech type of the speech used in the target call content through a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them;
[0157] The intention prediction module 30 is used to predict the purchasing intention of the marketing customer based on the target speech type through a preset intention prediction model to obtain the target purchasing intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchasing intention, and the speech types used by different call roles have different influence weights on the customer's purchasing intention.
[0158] The device for predicting customer purchase intention provided by the present application adopts the method for predicting customer purchase intention in the above-mentioned embodiment, which can solve the technical problem that the prediction result of the customer's purchase intention is relatively one-sided and the customer's true purchase intention cannot be accurately predicted by only predicting the customer's purchase intention through key words. Compared with the prior art, the beneficial effect of the device for predicting customer purchase intention provided by the present application is the same as the beneficial effect of the method for predicting customer purchase intention provided by the above-mentioned embodiment, and the other technical features in the device for predicting customer purchase intention are the same as the features disclosed in the above-mentioned embodiment method, which will not be described in detail here.
[0159] The present application provides a device for predicting customer purchase intention, and the device for predicting customer purchase intention includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for predicting customer purchase intention in the above-mentioned embodiment one.
[0160] Reference below Figure 7 , which shows a schematic diagram of the structure of a device suitable for predicting customer purchase intentions in the embodiment of the present application. The device for predicting customer purchase intentions in the embodiment of the present application may include, but is not limited to, mobile terminals such as mobile phones, tablet computers, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The customer purchase intention prediction device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0161] like Figure 7As shown, the prediction device of customer purchase intention may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the prediction device of customer purchase intention are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the prediction device of the customer's purchase intention to communicate wirelessly or wired with other devices to exchange data. Although the figure shows the prediction device of the customer's purchase intention with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.
[0162] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.
[0163] The customer purchase intention prediction device provided by the present application adopts the customer purchase intention prediction method in the above embodiment, which can solve the technical problem that the customer purchase intention is predicted only by key words, which makes the prediction result relatively one-sided and cannot accurately predict the customer's true purchase intention. Compared with the prior art, the beneficial effect of the customer purchase intention prediction device provided by the present application is the same as the beneficial effect of the customer purchase intention prediction method provided by the above embodiment, and the other technical features in the customer purchase intention prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0164] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0165] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0166] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the method for predicting customer purchase intention in the above-mentioned embodiment.
[0167] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.
[0168] The computer-readable storage medium may be included in the device for predicting customer purchase intention; or may exist independently without being incorporated into the device for predicting customer purchase intention.
[0169] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the device for predicting customer purchase intention, the device for predicting customer purchase intention executes the method for predicting customer purchase intention.
[0170] Computer program code for performing the operations of the present application may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0171] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0172] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.
[0173] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for predicting the customer's purchase intention, and can solve the technical problem that the customer's purchase intention is predicted only by key words, which makes the prediction result relatively one-sided and makes it impossible to accurately predict the customer's true purchase intention. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the method for predicting the customer's purchase intention provided by the above-mentioned embodiment, and will not be repeated here.
[0174] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for predicting customer purchase intention when executed by a processor.
[0175] The computer program product provided by the present application can solve the technical problem that the prediction result of the customer's purchase intention is relatively one-sided and the customer's real purchase intention cannot be accurately predicted by only predicting the customer's purchase intention through key words. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the method for predicting the customer's purchase intention provided by the above embodiment, and will not be elaborated here.
[0176] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.
Claims
1. A method for predicting customer purchase intention, characterized in that: The method includes: Obtain target call content of a target call role, where the target call role includes a marketing customer service representative and / or a marketing customer; Determine the target speech type of the speech used in the target call content by a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them; Based on the target speech type, the marketing customer's purchase intention is predicted by a preset intention prediction model to obtain the target purchase intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchase intention, and the speech types used by different call roles have different weights of influence on the customer's purchase intention.
2. The method according to claim 1, characterized in that Before the step of obtaining the target call content of the target call role, the method further includes: Acquire a training sample and a sample label of the training sample, wherein the training sample includes a first call content of a first call role, and the sample label of the training sample includes a first speech type label of a speech used by the first call role; Based on the first call content and the first speech type label, the preset first model to be trained is iteratively trained to obtain a speech classification model.
3. The method according to claim 2, characterized in that The step of iteratively training the preset first model to be trained based on the first call content and the first speech type label to obtain a speech classification model includes: Based on the first call content, pre-training a preset large model to obtain a first content analysis model, wherein the content analysis model has the ability to understand the call content in the call text data; Adding a preset linear classification layer to the first content analysis model to obtain a second content analysis model, wherein the preset linear classification layer is used to perform a speech classification task; Based on the first call content and the first speech type label, the second content analysis model is fine-tuned to obtain a speech classification model.
4. The method according to claim 3, characterized in that The step of fine-tuning the second content analysis model based on the first call content and the first speech type label to obtain a speech classification model includes: Inputting the first call content and the first speech type label into the second content analysis model; Determining, from the first call content, by using the second content analysis model, a first probability that the speech used by the first call character is of a preset speech type; Based on the first probability and the first speech type label, a first loss value is calculated by using a preset loss function; Based on the first loss value, the model parameters of the second content analysis model are updated, and the step of inputting the first call content and the first speech type label into the second content analysis model is returned to continue iterative training until a speech classification model that meets the preset first accuracy condition is obtained.
5. The method according to claim 4, characterized in that The first speech type includes a plurality of preset speech types, each of which is set for the first conversation role and / or other conversation roles except the first conversation role. The step of calculating the first loss value by using a preset loss function based on the first probability and the first speech type label includes: Based on the first probability and the first speech type label, a second loss value corresponding to each of the preset speech types is calculated by a preset loss function; Calculating a first loss value based on the second loss value and the first attention mask; Among them, the first attention mask is used to decide to ignore the influence of the preset speech type corresponding to the other calling roles on the first loss value.
6. The method according to claim 2, characterized in that After the step of iteratively training the preset first model to be trained based on the first call content and the first speech type label to obtain a speech classification model, the method further includes: Acquire a test sample and a sample label of the test sample, wherein the test sample includes the second call content of the second call role, and the sample label of the test sample includes a second speech type label of the speech used by the second call role; Based on the second call content, the speech classification model is tested to obtain a test result, wherein the test result includes a second probability that the speech used by the second call role is a preset speech type; Based on the second probability and the second speech type label, calculating a third loss value by using a preset loss function; Based on the third loss value, the model parameters of the second content analysis model are updated, and the step of testing the speech classification model based on the second call content to obtain the test result is returned, and iterative training is continued until a speech classification model that meets the preset second accuracy condition is obtained.
7. The method according to claim 6, characterized in that The second speech type includes a plurality of preset speech types, and the step of calculating the third loss value by a preset loss function based on the second probability and the second speech type label includes: Based on the second probability and the second speech type label, a fourth loss value corresponding to each of the preset speech types is calculated by a preset loss function; Calculating a third loss value based on the fourth loss value and the second attention mask; The second attention mask is used to decide to ignore the influence of other speech types on the third loss value, and the other speech types are speech types in the second speech type except for the easily confused speech types, and the easily confused speech types are determined based on the second probability and the second speech type label.
8. The method according to claim 1, characterized in that Before the step of obtaining historical call text data, the method further includes: Obtaining a historical speech type of a speech used by a first call role in historical call content, wherein the first call role includes a historical customer service representative and / or a historical customer; Based on the historical speech type, determine the purchase intention label; Based on the historical speech types and purchase intention labels, the preset second model to be trained is iteratively trained to obtain the preset intention prediction model.
9. The method according to claim 8, characterized in that The preset second model to be trained includes a binary classification model to be trained, and the step of iteratively training the preset second model to be trained based on the historical speech type and the purchase intention label to obtain the preset intention prediction model includes: Inputting the historical speech type and purchase intention label into the binary classification model to be trained; Predicting the purchase intention of the historical customers through the binary classification model to be trained to obtain the historical purchase probability; Calculating the error between the historical purchase probability and the purchase intention label to obtain an error calculation result; Based on the error calculation result, the model parameters of the binary classification model to be trained are updated, and the step of inputting the historical speech type and purchase intention label into the binary classification model to be trained is returned until a preset intention prediction model that meets the preset third precision condition is obtained.
10. A device for predicting customer purchase intention, characterized in that: The device comprises: An acquisition module, used to acquire target call content of a target call role, wherein the target call role includes a marketing customer service representative and / or a marketing customer; A speech classification module, used to determine the target speech type of the speech used in the target call content through a preset speech classification model, wherein the preset speech classification model is used to characterize the association between the call content of different call roles and the speech type of the speech used by them; The intention prediction module is used to predict the purchasing intention of the marketing customer based on the target speech type through a preset intention prediction model to obtain the target purchasing intention, wherein the preset intention prediction model is used to characterize the correlation between the speech types used by different call roles and the customer's purchasing intention, and the speech types used by different call roles have different influence weights on the customer's purchasing intention.
11. A device for predicting customer purchase intention, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for predicting customer purchase intention according to any one of claims 1 to 9.
12. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for predicting customer purchase intention according to any one of claims 1 to 9 are implemented.
13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for predicting customer purchase intention according to any one of claims 1 to 9 are implemented.
