Customer purchase willingness prediction method and device, medium and equipment

Abstract

The invention relates to the technical field of behavior prediction, and particularly provides a customer purchase willingness prediction method and device, a medium and equipment, and the method can comprise the steps: obtaining feature information corresponding to the data information of a product interested by a telemarketing customer; the feature information comprises basic features corresponding to the basic information of the customer, customer behavior features corresponding to the customer behavior information and text embedding vectors corresponding to the historical call information of the customer; the text embedding vector is determined based on label content of a process node of a dialogue text in the historical call information of the customer; splicing the basic features, the customer behavior features and the text embedding vectors to obtain a feature set; and predicting the feature set by using a pre-trained willingness prediction model to obtain the purchase willingness of the telemarketing customer to the interested product. According to the embodiment of the invention, the purchase willingness of the telemarketing customer can be accurately predicted.

Description

Technical Field

[0001] This application relates to the field of behavioral prediction technology, and more specifically, to a method, apparatus, medium, and device for predicting customer purchase intentions. Background Technology

[0002] With the widespread adoption of communication technologies and big data, telemarketing has become a crucial channel for enterprise marketing and customer relationship management. To improve sales efficiency, reduce operating costs, and achieve precise marketing in a highly competitive market, effectively predicting customer purchase intentions has become essential. Currently, predicting customer purchase intentions typically involves collecting static customer attribute information (such as age, gender, region, and occupation) and limited historical interaction data (such as past purchase records and product browsing duration). Domain experts then set a series of judgment rules; these rules are used to predict customer purchase intentions. However, the multi-dimensional, static, historical structured data relied upon in current technologies has a relatively singular data dimension, making accurate prediction of customer purchase intentions impossible.

[0003] Therefore, how to provide a technical solution for accurately predicting customer purchase intentions has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of some embodiments of this application is to provide a method, apparatus, medium and device for predicting customer purchase intentions. The technical solutions of the embodiments of this application can improve the accuracy of predicting customer purchase intentions and improve the accuracy and efficiency of telemarketing.

[0005] In a first aspect, some embodiments of this application provide a method for predicting customer purchase intention, comprising: acquiring feature information corresponding to data information of products of interest to telemarketing customers; wherein, the data information includes basic customer information, customer behavior information, and customer historical call information; the customer behavior information includes purchase records, application usage time, or consumption data; the feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer historical call information; the text embedding vectors are determined based on the label content of the flow nodes of the dialogue text in the customer historical call information; concatenating the basic features, the customer behavior features, and the text embedding vectors to obtain a feature set; and using a pre-trained intention prediction model to predict the feature set to obtain the purchase intention degree of the telemarketing customer for the products of interest.

[0006] Some embodiments of this application obtain feature information from data on products of interest to telemarketing customers, then concatenate these feature information to obtain a feature set. Finally, the feature set is input into a pre-trained intention prediction model to predict the purchase intention of telemarketing customers for products of interest. Embodiments of this application can accurately predict the purchase intention of telemarketing customers, thereby improving the accuracy of telemarketing and customer experience.

[0007] In some embodiments, obtaining the feature information corresponding to the data information of products of interest to telemarketing customers includes: dividing the customer behavior information according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

[0008] Some embodiments of this application process customer behavior information to obtain first and second behavioral features, thereby achieving feature analysis of customer behavior information and improving data support for subsequent predictions.

[0009] In some embodiments, obtaining the feature information corresponding to the data information of products of interest to telemarketing customers includes: extracting multiple key tag contents corresponding to the process nodes in the customer's historical call information; concatenating the multiple key tag contents according to time nodes to obtain a concatenated sentence; inputting the concatenated sentence into a language embedding model and outputting the text embedding vector.

[0010] Some embodiments of this application extract multiple key tags from customer historical call information, concatenate them to obtain concatenated sentences, and then combine them with a language embedding model to obtain text embedding vectors, thereby achieving feature processing of customer historical call information and improving data support for subsequent predictions.

[0011] In some embodiments, extracting multiple key tag contents corresponding to the process nodes in the customer's historical call information includes: identifying multiple key process nodes and multiple customer question nodes in the customer's historical call information; and using the text summaries corresponding to the multiple key process nodes and the multiple customer question nodes as the multiple key tag contents.

[0012] Some embodiments of this application obtain corresponding text summaries as key dialogue text by identifying multiple key process nodes and question nodes in the customer's historical call information, thereby achieving accurate extraction of effective information from the customer's historical call information.

[0013] In some embodiments, the step of using a pre-trained intention prediction model to predict the feature set and obtain the purchase intention of the telemarketing customer for the product of interest includes: obtaining the intention prediction results of the basic features and the customer behavior features; and correcting the intention prediction results using the text embedding vector to obtain the purchase intention.

[0014] Some embodiments of this application first obtain initial intention prediction results through a decision tree model, and then correct them using text embedding vectors to obtain the purchase intention level, thereby improving the accuracy of the analysis of the purchase intention of telemarketing customers.

[0015] Secondly, some embodiments of this application provide an apparatus for predicting customer purchase intention, comprising: a feature processing module, used to acquire feature information corresponding to data information of products of interest to telemarketing customers; wherein, the data information includes basic customer information, customer behavior information, and customer historical call information; the customer behavior information includes purchase records, application usage time, or consumption data; the feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer historical call information; the text embedding vectors are determined based on the label content of the flow nodes of the dialogue text in the customer historical call information; a feature concatenation module, used to concatenate the basic features, the customer behavior features, and the text embedding vectors to obtain a feature set; and a prediction module, used to predict the feature set using a pre-trained intention prediction model to obtain the purchase intention degree of the telemarketing customer for the products of interest.

[0016] In some embodiments, the feature processing module is used to: divide the customer behavior information according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

[0017] In some embodiments, the feature processing module is used to: extract multiple key tag contents corresponding to the process nodes in the customer's historical call information; concatenate the multiple key tag contents according to time nodes to obtain a concatenated sentence; input the concatenated sentence into a language embedding model and output the text embedding vector.

[0018] In some embodiments, the feature processing module is used to: identify multiple key process nodes and multiple customer question nodes in the customer's historical call information; and use the text summaries corresponding to the multiple key process nodes and the multiple customer question nodes as the content of the multiple key tags.

[0019] In some embodiments, the prediction module is used to: obtain the intention prediction results of the basic features and the customer behavior features; and correct the intention prediction results using the text embedding vector to obtain the purchase intention level.

[0020] Thirdly, some embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the method described in any embodiment of the first aspect.

[0021] Fourthly, some embodiments of this application provide an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, can implement the method as described in any embodiment of the first aspect.

[0022] Fifthly, some embodiments of this application provide a computer program product, the computer program product including a computer program, wherein the computer program, when executed by a processor, can implement the method described in any embodiment of the first aspect. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of some embodiments of this application, the accompanying drawings used in some embodiments of this application will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A system diagram for predicting customer purchase intentions is provided for some embodiments of this application; Figure 2 One of the flowcharts for a method of predicting customer purchase intentions provided for some embodiments of this application; Figure 3 Model structure diagrams of DistilBERT provided for some embodiments of this application; Figure 4 A second flowchart illustrating a method for predicting customer purchase intentions, provided for some embodiments of this application; Figure 5 Block diagrams of apparatus for predicting customer purchase intentions provided for some embodiments of this application; Figure 6 A schematic diagram of an electronic device provided for some embodiments of this application. Detailed Implementation

[0025] The technical solutions of some embodiments of this application will now be described with reference to the accompanying drawings.

[0026] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] In related technologies, by collecting static customer attribute information (such as age, gender, region, occupation, etc.) and limited historical interaction data (such as past purchase records, product browsing time, etc.), domain experts set a series of judgment rules. For example, different weights are assigned to characteristics such as "recent consultation behavior," "belonging to high-net-worth individuals," and "having purchased related products," constructing a scoring card or decision tree model. Before making outbound telemarketing calls, the system filters and prioritizes the customer list according to these preset rules, predicts their approximate purchase intention level, and thus guides sales personnel to prioritize calls.

[0028] Furthermore, building upon the expert rule model, a classification prediction model can be constructed using the same data and classic machine learning algorithms such as logistic regression, support vector machines, or random forests. First, a large number of feature variables are extracted from the customer database and interaction history, including demographic characteristics, historical spending power, response history to marketing campaigns, and simple behavioral sequences (such as the outcome of the most recent call). Then, the model is trained using historical telemarketing data labeled "successfully converted" or "not converted." The trained model can calculate a "purchase probability" score for new customer data, and the telemarketing system sorts the customer list based on this score, prioritizing contact with high-probability customers.

[0029] While the aforementioned existing technologies can help improve telemarketing efficiency to some extent, they all have significant limitations. For example, the main problem with current solutions is the limited data dimensions and incomplete customer profiles; that is, whether it's a rule-based model or a traditional machine learning model, the underlying features they rely on are mostly static, historical, and structured data. Furthermore, this data often fails to directly reflect the causal relationship between customer purchase intentions. They cannot effectively capture and quantify the massive amounts of unstructured information generated during real-time call interactions, such as the topic of the conversation, the customer's tone and attitude, their true semantic intent, and subtle changes in the rhythm of the conversation. This results in incomplete customer profiles, biased prediction results, and poor prediction accuracy.

[0030] In view of this, some embodiments of this application provide a method for predicting customer purchase intention. This method involves collecting data on products of interest to telemarketing customers, performing feature processing to obtain feature information, and then combining this feature information with a pre-trained intention prediction model to predict the purchase intention level. The embodiments of this application can achieve accurate prediction of telemarketing customers' purchase intention, improving the accuracy and efficiency of telemarketing and providing customers with a better marketing experience.

[0031] The following is in conjunction with the appendix Figure 1 The overall structure of a system for predicting customer purchase intentions provided by some embodiments of this application is illustrated by way of example.

[0032] like Figure 1 As shown, some embodiments of this application provide a system diagram for predicting customer purchase intention. This system for predicting customer purchase intention may include a terminal 100 and a server 200. The terminal 100 can send data information about products of interest to telemarketing customers to the server 200. After receiving the data information, the server 200 can perform feature processing on it, and then analyze the feature set obtained after feature processing using a pre-trained intention prediction model to obtain the purchase intention level of the telemarketing customer for the product of interest. A higher purchase intention level indicates a stronger purchase intention from the customer.

[0033] In some embodiments of this application, the terminal 100 can be a mobile terminal or a non-portable computer terminal, and the embodiments of this application are not specifically limited here.

[0034] In some other embodiments of this application, if the terminal 100 has the function of feature processing of data information, the feature processing step can be performed by the terminal 100. It is also understood that the intention prediction model is pre-trained and then deployed to the server 200; its training process can be performed by the server 200 or trained by other external devices and then deployed to the server 200. The embodiments of this application are not specifically limited herein.

[0035] The following is in conjunction with the appendix Figure 2 The present application provides an exemplary embodiment of the process for predicting the purchase intention of telemarketing customers, performed by server 200.

[0036] Please see the appendix Figure 2 , Figure 2 A flowchart illustrating a method for predicting customer purchase intentions is provided for some embodiments of this application. This method for predicting customer purchase intentions may include: S210, acquire feature information corresponding to data information of products of interest to telemarketing customers; wherein, the data information includes basic customer information, customer behavior information, and customer historical call information; the customer behavior information includes purchase records, application usage time, or consumption data; the feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer historical call information; the text embedding vectors are determined based on the tag content of the flow nodes of the dialogue text in the customer historical call information.

[0037] For example, in a specific embodiment of this application, three types of raw customer data (as a specific example of data information) are first obtained for a certain telemarketing customer regarding a certain type of product of interest. The customer's basic information includes name, mobile phone number, ID card number, etc. Customer behavior information includes at least purchase records, app usage time (as a specific example of application usage time), and loan information (as a specific example of consumption data). Customer historical call information includes at least call duration, human-machine / human-human call duration, and historical dialogue text. The customer's historical dialogue text refers to past telemarketing call records for a specific customer; it includes the communication content between the customer and the salesperson, and the communication text between the customer and the voice robot.

[0038] The corresponding feature information is obtained by performing feature engineering on the above three types of raw customer data. It should be noted that, in addition to the above three types of raw customer data, it can also be expanded according to the actual telemarketing scenario, and the embodiments of this application are not limited to this.

[0039] In some embodiments of this application, S210 may include: dividing the customer behavior information according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

[0040] For example, in a specific embodiment of this application, customer behavior information is processed through feature engineering into multiple short-term behavioral features (as a specific example of the first behavioral feature) and long-term behavioral features (as a specific example of the second behavioral feature). Long-term behavioral features include, but are not limited to, relevant data of the customer within a set time period; such as app usage time and loan information over the past 1 month, 3 months, and 6 months (i.e., the set time period). For example, total loan amount and online shopping information, such as the total consumption amount of a certain category (i.e., products of interest) in the past month (as a specific example of consumption data). Specifically, app usage time is converted from the customer's login and logout times into online time through the app's recorded customer behavior logs; app usage time can then be further expanded to the total online time over a certain period (e.g., the past week), average online time, standard deviation of online time, etc. Loan information may include the customer's total loan amount over a certain period (e.g., the past three months) processed from the customer's online lending limit, lending frequency, repayment amount, number of overdue payments, and overdue amount.

[0041] In some embodiments of this application, S210 may include: S211, extract multiple key tag contents corresponding to the process nodes in the customer's historical call information.

[0042] For example, in a specific embodiment of this application, a large language model or a pre-trained extraction model is used to extract key text information of key process nodes in historical dialogue text (as a specific example of key tag content, hereinafter referred to as tag content).

[0043] In some embodiments of this application, S211 may include: identifying multiple key process nodes and multiple customer question nodes in the customer's historical call information; and using the text summaries corresponding to the multiple key process nodes and the multiple customer question nodes as the content of the multiple key tags.

[0044] For example, in specific embodiments of this application, the information extraction task is mainly divided into two categories: one is customer questions – customer inquiries or responses to salesperson's scripts (as a specific example of a customer question node). The other is process nodes – key process nodes in the salesperson's dialogue text where the salesperson guides the customer. The candidate set of text summaries corresponding to these two types of node information is a predefined, enumerable set of tags, where each tag is a phrase of no more than 10 characters, which can be flexibly adjusted according to actual circumstances. In addition to the two process nodes listed above, it can also be flexibly expanded according to actual circumstances, and the embodiments of this application are not limited to this.

[0045] The following example uses a historical conversation text to illustrate the process of obtaining multiple key tag contents. This historical conversation text could be the last conversation regarding the product of interest.

[0046] 1) Course Marketing Consultant: Hello, parent. I am the teacher who contacted you before. The score-boosting course you signed up for for your child will start soon. Could you tell me your child's grade again so I can show you how the class is conducted? 2) Customer: Call immediately.

[0047] 3) Customer: Oh, my child is currently living at school, so he doesn't have time to attend classes.

[0048] 4) Course Marketing Consultant: It's okay, parents. Since many children in junior high and high school board at school, our courses have taken all that into consideration. You can choose the subjects your child is weak in and focus on. These are specially designed to extract key problem-solving techniques. Your child can watch the recordings after school, and if there's anything they don't understand, they can ask a tutor for help. What grade is your child in? 5) Customer: I'm in the third year of junior high school now.

[0049] 6) Course Marketing Consultant: You're in the third year of junior high school, right? This is the stage where children need to prepare for the high school entrance exam. Every point lost now could mean missing out on their ideal high school. If they can't answer the key points in reading comprehension, the combined problems in functions and geometry in math will directly differentiate their grades. So, which subject is your child weakest in right now, Chinese, math, or English? 7) Client: Well, his is mainly about physics.

[0050] 8) Course Marketing Consultant: Yes, for science subjects, physics and chemistry are quite abstract, so you should focus on this. Our course is a 10-lesson intensive improvement course that will target your child's weakest subject for focused improvement. For subjects like physics, chemistry, and biology, we'll teach your child basic problem-solving techniques and key strategies to quickly improve their grades. It's currently part of Gaotu's anniversary promotion, so there's only one registration fee. The course, all materials and handouts, and one-on-one Q&A throughout the course are included—no additional charges. Is this phone number your social media number? I'll add you on social media and send you this registration poster so your child can claim their spot, okay? 9) Customer: Okay, thank you.

[0051] The above dialogue was extracted using a large language model, breaking it down into predefined customer questions and key process nodes. This process essentially involves encoding and embedding the original text using the underlying language model of the large model, followed by decoding. By extracting the above historical dialogue text, the following key tags were obtained for each process node: 1) Key moments: opening + asking about grade level 2)-3) Customer question: My child is busy and doesn't have time to attend classes.

[0052] 4) Key Nodes: Q&A + Inquiry about Grade Level 5) Customer inquiries: Reply to grade level 6) Key juncture: Inquire about weak subjects 7) Customer inquiries: Responding to weak departments 8) Key Node: Inviting Students to Take the Course 9) Customer Inquiry: Accepting the Invitation S212: Concatenate multiple key tag contents according to time nodes to obtain the concatenated statement.

[0053] For example, the customer's questions and key node tags can be reassembled according to the time sequence (i.e., time nodes) to form a natural language sentence (i.e., a concatenated sentence) composed of the tag sequence. The concatenated sentence is: [Key node: opening + asking about grade level] [Customer question: My child is busy and doesn't have time for class] [Key node: answering questions + asking about grade level] [5 Customer question: replying about grade level] [Key node: asking about weak subjects] [Customer question: replying about weak subjects] [Key node: inviting to take a class] [Customer question: accepting the invitation].

[0054] S213, input the concatenated sentence into the language embedding model and output the text embedding vector.

[0055] For example, in a specific embodiment of this application, DistilBERT (as a specific example of a language embedding model) is used to perform a second embedding on the concatenated sentences generated by the above-mentioned large model, outputting a text embedding vector of length n. This vector can be regarded as an N-dimensional feature in the application scenario. The model structure of DistilBERT is as follows: Figure 3 As shown, it includes Input, Token Embeddings, Segment Embeddings, and Position Embeddings. The Input example uses a simple sentence input to illustrate the subsequent processing steps in the model. This model structure yields an N-dimensional text embedding vector corresponding to the concatenated sentence described in this application.

[0056] S220, the basic features, the customer behavior features, and the text embedding vector are concatenated to obtain a feature set.

[0057] For example, in a specific embodiment of this application, after processing the above features, the customer's three-dimensional basic features can be obtained, such as age, gender, and province / city / district of residence [bi1, bi2, bi3]. Eighteen-dimensional customer behavior features: last ring duration, last call duration, whether the last call used a voice assistant, whether the last call was completely silent, whether the last call ended instantly, total number of historical outbound calls, total number of historical connected calls, historical connection rate, total historical ring duration, average historical ring duration, total historical call duration, average historical call duration, number of historical voice assistant calls, historical voice assistant rate, number of historical instant hang-ups, historical instant hang-up rate, number of historical instances of complete silence, historical complete silence rate [b1, b2, ..., b18]. An N-dimensional text embedding vector [emb1, emb2, ..., embN]. Concatenating the above three types of features yields the feature set features = [bi1, bi2, bi3, b1, ..., b18, emb1, ..., embN].

[0058] S230, the feature set is predicted using a pre-trained intention prediction model to obtain the purchase intention of the telemarketing customer for the product of interest.

[0059] For example, in a specific embodiment of this application, the feature set obtained above is input into a pre-trained LightGBM model (as a specific example of an intention prediction model), which outputs the purchase intention level of telemarketing customers. The essence of the LightGBM model is an optimized implementation of gradient boosting decision trees. Its core idea is to iteratively train a series of weak decision trees (i.e., base learners), each tree working to correct the prediction error of the previous tree, and finally, the prediction results of all trees are weighted and summed to obtain a powerful ensemble model (i.e., an intention prediction model). In each training stage, features with high information content are incorporated into the algorithm to predict customer purchase intentions.

[0060] In some embodiments of this application, S230 may include: obtaining the intention prediction results of the basic features and the customer behavior features; and correcting the intention prediction results using the text embedding vector to obtain the purchase intention level.

[0061] For example, in a specific embodiment of this application, the LightGBM model first analyzes the first two features in the feature set, namely the basic features and the customer behavior features, to obtain the intention prediction result. Then, by introducing features from text embedding vectors, it corrects the prediction error caused by the intention prediction result of the combination of the customer's basic features and customer behavior features, thereby obtaining the final purchase intention level. Essentially, the LightGBM model uses information from actual telephone conversations to more accurately predict customer purchase intentions based on existing information.

[0062] The following is in conjunction with the appendix Figure 4 The present application provides an exemplary description of the specific process for predicting the purchase intention of telemarketing customers through some embodiments.

[0063] Please see the appendix Figure 4 , Figure 4 A flowchart illustrating a method for predicting customer purchase intentions, provided for some embodiments of this application.

[0064] The above process is illustrated below by example.

[0065] S410 obtains basic customer information, customer behavior information, and customer call history information for products that telemarketing customers are interested in.

[0066] S420 performs feature engineering on basic customer information to obtain basic features.

[0067] S430 divides customer behavior information according to time length to obtain the first and second behavioral characteristics in customer behavior features.

[0068] S440 extracts multiple key tags from the customer's historical call information; concatenates the multiple key tags according to time nodes to obtain the concatenated sentence; inputs the concatenated sentence into the language embedding model and outputs the text embedding vector.

[0069] S450 concatenates the basic features, customer behavior features, and text embedding vectors to obtain the feature set.

[0070] S460 inputs the feature set into a pre-trained intention prediction model to obtain the purchase intention of telemarketing customers for products they are interested in.

[0071] It is understood that the specific implementation process of S410~S460 can refer to the method implementation examples provided above. To avoid repetition, detailed descriptions are omitted here.

[0072] As can be seen from the embodiments of this application described above, this application obtains a feature set by extracting and concatenating various features to predict customer intentions. In particular, extracting information from customer dialogue text before embedding it effectively reduces noise data in the audio-to-text conversion, noise in customer statements within the text, and improves the interpretability of the model. Furthermore, the second embedding of the extracted information, rather than using the embedding of extracted entities, further incorporates the contextual information of the telephone conversation, significantly contributing to the improved accuracy of predicting purchase intentions.

[0073] Please refer to Figure 5 , Figure 5The diagram illustrates a block diagram of a customer purchase intention prediction apparatus provided in some embodiments of this application. It should be understood that this customer purchase intention prediction apparatus corresponds to the method embodiments described above and is capable of performing the various steps involved in the method embodiments. The specific functions of this customer purchase intention prediction apparatus can be found in the description above; detailed descriptions are omitted here to avoid repetition.

[0074] Figure 5 The device for predicting customer purchase intention includes at least one software functional module that can be stored in a memory or embedded in the device in the form of software or firmware. The device includes: a feature processing module 510, used to acquire feature information corresponding to data information about products of interest to telemarketing customers; wherein the data information includes basic customer information, customer behavior information, and customer historical call information; the customer behavior information includes purchase records, application usage time, or consumption data; the feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer historical call information; the text embedding vectors are determined based on the label content of the flow nodes in the dialogue text of the customer historical call information; a feature concatenation module 520, used to concatenate the basic features, the customer behavior features, and the text embedding vectors to obtain a feature set; and a prediction module 530, used to predict the feature set using a pre-trained intention prediction model to obtain the purchase intention degree of the telemarketing customer for the products of interest.

[0075] In some embodiments of this application, the feature processing module 510 is used to: divide the customer behavior information according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

[0076] In some embodiments of this application, the feature processing module 510 is used to: extract multiple key tag contents corresponding to the process nodes in the customer's historical call information; concatenate the multiple key tag contents according to time nodes to obtain a concatenated sentence; input the concatenated sentence into a language embedding model and output the text embedding vector.

[0077] In some embodiments of this application, the feature processing module 510 is used to: identify multiple key process nodes and multiple customer question nodes in the customer's historical call information; and use the text summaries corresponding to the multiple key process nodes and the multiple customer question nodes as the content of the multiple key tags.

[0078] In some embodiments of this application, the prediction module 530 is used to obtain the intention prediction results of the basic features and the customer behavior features; and to correct the intention prediction results using the text embedding vector to obtain the purchase intention degree.

[0079] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the aforementioned method, and will not be elaborated further here.

[0080] Some embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can perform the operation of any of the methods corresponding to the methods provided in the above embodiments.

[0081] Some embodiments of this application also provide a computer program product, which includes a computer program, wherein when the computer program is executed by a processor, it can implement the operation of any of the methods corresponding to the above embodiments provided in the above embodiments.

[0082] like Figure 6 As shown, some embodiments of this application provide an electronic device 600, which includes a memory 610, a processor 620, and a computer program stored in the memory 610 and executable on the processor 620. When the processor 620 reads the program from the memory 610 via a bus 630 and executes the program, it can implement the methods of any of the above embodiments.

[0083] Processor 620 can process digital signals and can include various computing architectures. For example, it can be a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements multiple instruction set combinations. In some examples, processor 620 can be a microprocessor.

[0084] The memory 610 can be used to store instructions executed by the processor 620 or data related to the execution of instructions. These instructions and / or data may include code for implementing some or all of the functions of one or more modules described in the embodiments of this application. The processor 620 of this disclosure embodiment can be used to execute the instructions in the memory 610 to implement the methods shown above. The memory 610 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memories well known to those skilled in the art.

[0085] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0086] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for predicting customer purchase intention, characterized in that, include: The system acquires feature information corresponding to data on products of interest to telemarketing customers. This data includes basic customer information, customer behavior information, and customer call history information. Customer behavior information includes purchase records, application usage time, or consumption data. Feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer call history information. The text embedding vectors are determined based on the tag content of the flow nodes in the dialogue text within the customer call history information. The basic features, customer behavior features, and text embedding vectors are concatenated to obtain a feature set; The feature set is predicted using a pre-trained intention prediction model to obtain the purchase intention of the telemarketing customer for the product they are interested in.

2. The method as described in claim 1, characterized in that, The feature information corresponding to the data information of products of interest to telemarketing customers includes: The customer behavior information is divided according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

3. The method as described in claim 1 or 2, characterized in that, The feature information corresponding to the data information of products of interest to telemarketing customers includes: Extract multiple key tag contents corresponding to the process nodes from the customer's historical call information; The content of the multiple key tags is concatenated according to time nodes to obtain the concatenated sentence; The concatenated sentence is input into the language embedding model, and the text embedding vector is output.

4. The method as described in claim 3, characterized in that, The extraction of multiple key tag contents corresponding to the process nodes in the customer's historical call information includes: Identify multiple key process nodes and multiple customer question nodes in the customer's historical call information; The text summaries corresponding to the multiple key process nodes and the multiple customer question nodes are used as the content of the multiple key tags.

5. The method as described in claim 1 or 2, characterized in that, The step of using a pre-trained intention prediction model to predict the feature set and obtain the purchase intention of the telemarketing customer for the product of interest includes: Obtain the intention prediction results of the basic features and the customer behavior features; The purchase intention level is obtained by correcting the intention prediction result using the text embedding vector.

6. A device for predicting customer purchase intention, characterized in that, include: The feature processing module is used to acquire feature information corresponding to data information of products of interest to telemarketing customers; wherein, the data information includes basic customer information, customer behavior information, and customer historical call information; the customer behavior information includes purchase records, application usage time, or consumption data; the feature information includes basic features corresponding to the basic customer information, customer behavior features corresponding to the customer behavior information, and text embedding vectors corresponding to the customer historical call information; the text embedding vectors are determined based on the tag content of the flow nodes of the dialogue text in the customer historical call information; The feature concatenation module is used to concatenate the basic features, the customer behavior features, and the text embedding vector to obtain a feature set; The prediction module is used to predict the feature set using a pre-trained intention prediction model to obtain the purchase intention of the telemarketing customer for the product of interest.

7. The apparatus as claimed in claim 6, characterized in that, The feature processing module is used for: The customer behavior information is divided according to the time length to obtain a first behavior feature and a second behavior feature; wherein the first behavior feature and the second behavior feature constitute the customer behavior feature.

8. The apparatus as claimed in claim 6 or 7, characterized in that, The feature processing module is used for: Extract multiple key tag contents corresponding to the process nodes from the customer's historical call information; The content of the multiple key tags is concatenated according to time nodes to obtain the concatenated sentence; The concatenated sentence is input into the language embedding model, and the text embedding vector is output.

9. The apparatus as claimed in claim 8, characterized in that, The feature processing module is used for: Identify multiple key process nodes and multiple customer question nodes in the customer's historical call information; The text summaries corresponding to the multiple key process nodes and the multiple customer question nodes are used as the content of the multiple key tags.

10. The apparatus as claimed in claim 6 or 7, characterized in that, The prediction module is used for: Obtain the intention prediction results of the basic features and the customer behavior features; The purchase intention level is obtained by correcting the intention prediction result using the text embedding vector.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to perform the method as described in any one of claims 1-5.

12. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the computer program is executed by the processor to perform the method as described in any one of claims 1-5.

Images