Predicting method and device for repeated incoming calls

Abstract

The application discloses a repeated incoming call prediction method and device, and belongs to the technical field of computers. The repeated incoming call prediction method comprises the following steps: obtaining historical call text of a target user; determining first data and second data of the target user based on the historical call text; wherein the first data comprises a first feature, and the first feature is used for indicating whether the initial intention of the target user for an incoming call is satisfied; the second data comprises a second feature and / or a third feature, the second feature is used for indicating whether the target user has a call abnormality, and the third feature is used for indicating whether the target user has an understanding abnormality; inputting the first data and the second data into a target machine learning model which has been trained to perform repeated incoming call prediction, and obtaining a prediction result of the target user, wherein the prediction result comprises whether the target user will repeat an incoming call.

Description

Technical Field

[0001] This application belongs to the field of computer technology, and specifically relates to a method and apparatus for predicting repeat calls. Background Technology

[0002] With the widespread adoption of online transactions, more and more users are getting used to handling business via telephone. However, because customer service representatives cannot show users the on-screen interface during phone calls, users may experience operational failures and be unable to complete their transactions, leading to repeated calls from users.

[0003] The solutions in related technologies can only analyze repeat calls from users to determine the reasons for repeat calls, but they cannot predict in advance whether users they have already communicated with will call again based on the user's call history. In other words, the related technologies have the problem of not being able to predict whether users they have already communicated with will call again. Summary of the Invention

[0004] This application provides a method and apparatus for predicting repeat calls, which can solve the problem in related technologies that it is impossible to predict whether a user who has already been contacted by phone will call again.

[0005] In a first aspect, embodiments of this application provide a method for predicting repeat calls, the method comprising:

[0006] Obtain the target user's historical call text;

[0007] Based on the historical call text, first data and second data of the target user are determined; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has call abnormalities, and the third feature is used to indicate whether the target user has comprehension abnormalities.

[0008] The first data and the second data are input into a pre-trained target machine learning model to predict repeat calls, thereby obtaining a prediction result for the target user, which includes whether the target user will make repeat calls.

[0009] Secondly, embodiments of this application provide an apparatus for predicting repeat calls, the apparatus comprising:

[0010] The acquisition module is used to acquire the historical call text of the target user;

[0011] The determining module is used to determine first data and second data of the target user based on the historical call text; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has call abnormalities, and the third feature is used to indicate whether the target user has comprehension abnormalities.

[0012] The processing module is used to input the first data and the second data into a pre-trained target machine learning model to predict repeat calls, and to obtain the prediction result of the target user, the prediction result including whether the target user will make repeat calls.

[0013] Thirdly, embodiments of this application provide an electronic device including a processor and a memory, wherein the memory stores a program or instructions that, when executed by the processor, implement the steps of the method described in the first aspect.

[0014] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0015] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0016] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the method described in the first aspect.

[0017] In this embodiment, the historical call text of a target user is obtained; based on the historical call text, first data and second data of the target user are determined; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, where the second feature is used to indicate whether the target user has call abnormalities, and the third feature is used to indicate whether the target user has comprehension abnormalities; the first data and the second data are input into a pre-trained target machine learning model to predict repeat calls, and a prediction result for the target user is obtained, including whether the target user will call again. Thus, the historical call text of the target user can be used to determine whether the target user's initial call purpose has been met, whether the target user has call abnormalities, and / or whether the target user has comprehension abnormalities. By combining the features of the initial call purpose, call abnormalities, and / or comprehension abnormalities, and using a pre-trained target machine learning model to predict repeat calls, it is possible to quickly predict whether a target user with whom a phone call has already been made will call again. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a method for predicting repeat calls provided in an embodiment of this application;

[0019] Figure 2 This is a schematic diagram of another method for predicting repeat calls provided in an embodiment of this application;

[0020] Figure 3 This is a flowchart of a method for predicting repeat calls provided in an embodiment of this application;

[0021] Figure 4 This is a flowchart of another method for predicting repeat calls provided in an embodiment of this application;

[0022] Figure 5 This is a flowchart of another method for predicting repeat calls provided in an embodiment of this application;

[0023] Figure 6 This is a structural block diagram of a repeat call prediction device provided in an embodiment of this application;

[0024] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0027] In related technologies, when analyzing a user's historical calls, it is only possible to determine the reason for the call based on the content of the historical calls, but it is impossible to predict whether a user who has already been contacted by phone will call again.

[0028] To address this issue, this application proposes a method for predicting repeat calls. Figure 1 This is a schematic diagram of a method for predicting repeat calls provided in an embodiment of this application. Figure 1 As shown, in the process of predicting whether a target user will call repeatedly, this embodiment first determines whether the target user's initial purpose for calling has been met, whether there are any call abnormalities, and whether there are any comprehension abnormalities. Then, the features of these three aspects—initial purpose, call abnormalities, and comprehension abnormalities—are input into the target machine learning model to obtain the prediction result of repeated calls. It is understandable that the reasons for repeated calls can generally be categorized as follows: 1. The user's initial purpose for calling has not been met. For example, the service the user hoped to handle via telephone was not successfully completed. 2. Call abnormalities. For example, during a telephone conversation, the user has other matters to attend to and interrupts the communication. 3. Comprehension abnormalities. For example, the customer service representative's explanation is not logically clear, and the user says they do not understand. Therefore, by comprehensively considering the features of the three aspects—initial purpose, call abnormalities, and comprehension abnormalities—and taking into account various factors that may lead to repeated calls, the accuracy of the repeated call prediction result can be improved.

[0029] Thus, the repeat call prediction method provided in this application embodiment, on the one hand, eliminates the need for manual prediction, quickly predicting whether a target user will call repeatedly through a target detection model; on the other hand, it comprehensively considers the initial purpose of the call, as well as characteristics of call abnormalities and comprehension abnormalities, taking into account various factors that may lead to repeat calls from target users, thereby making the prediction results more accurate. The repeat call prediction results of this application embodiment can help business operators focus on target users who may call repeatedly and assign experienced customer service representatives to answer these calls, avoiding complaints from target users. Of course, the repeat call prediction results of this application embodiment can also help business operators study how to reduce repeat calls from customers and alleviate the workload of customer service representatives.

[0030] It should be noted that, Figure 1 This is merely one example of the present application and should not be considered a limitation thereof. In practice, embodiments of the present application can predict whether a target user will call repeatedly based solely on the initial purpose of the call and characteristics of call anomalies. Of course, embodiments of the present application can also predict whether a user will call repeatedly based solely on the initial purpose of the call and characteristics of call anomalies.

[0031] Furthermore, the embodiments of this application further illustrate how to determine whether the target user's initial intention for the call has been met, whether the target user has any call abnormalities, and whether the target user has any misunderstandings. Figure 2 This is a schematic diagram of another method for predicting repeat calls provided in an embodiment of this application. For example... Figure 2 As shown, the repeat call prediction method provided in this application embodiment can use six pre-trained machine learning models. In this application embodiment, the target user's historical call text can be input into the intent recognition model to obtain intent information, and the caller's initial intention can be determined based on the intent information. Simultaneously, the target call text from the historical call text can be input into a first detection model and a second detection model. The first detection model can detect the target call text sentence by sentence and output the call anomaly identification result for each sentence. The output of the first detection model can be used as input to the second detection model. Based on the output of the second detection model, it can be determined whether the target user has call anomalies. The third detection model can detect the target call text sentence by sentence and output the comprehension anomaly identification result for each sentence. The output of the third detection model can be used as input to the fourth detection model. Based on the output of the fourth detection model, it can be determined whether the target user has comprehension anomalies.

[0032] It should be noted that, Figure 2This is merely an example of this application and should not be considered as a limitation thereof. In practice, embodiments of this application may use more or fewer machine learning models than six to predict whether a user will call repeatedly, or other processing methods such as calculation formulas may be used instead of machine learning models.

[0033] Furthermore, this application also proposes to integrate the intent nodes corresponding to historical call texts to form a historical call intent path, and compare the historical call intent path with the standard path of the target service to determine whether the historical call intent path covers the standard path, thereby determining whether the target user's initial call intention has been met. Since the standard path of the target service is an essential part of meeting the target user's initial call intention, by comparing the historical call intent path with the standard path of the target service, it is possible to accurately determine whether the target user's initial call intention has been met, thereby improving the accuracy of repeat call prediction results to a certain extent.

[0034] Furthermore, it should be noted that this application also proposes determining a second feature by combining the number of times single-sentence call anomalies occur in the target call text and the location of each single-sentence call anomaly, and determining a third feature by combining the number of times single-sentence comprehension anomalies occur in the target call text and the location of each single-sentence comprehension anomaly. Since the number of anomalies and the location of anomalies can reflect whether anomalies in a call have been resolved, combining these two factors can accurately determine whether there are anomalies in the entire call text, avoiding the input of resolved anomalies as features into the target machine learning model, thereby improving the accuracy of repeat call prediction results to a certain extent.

[0035] The following detailed description, in conjunction with the accompanying drawings, of the repeat call prediction method provided in this application through specific embodiments and application scenarios, will illustrate this method in detail.

[0036] Figure 3 This is a flowchart of a method for predicting repeat calls provided in an embodiment of this application. Figure 3 As shown, the method for predicting repeat calls provided in this application embodiment may include:

[0037] Step 310: Obtain the target user's historical call texts;

[0038] Historical call texts can include the texts of the target user's most recent calls (e.g., three, five, ten, etc.). These call texts can be obtained by converting historical calls using speech-to-text technology. The target user's historical call texts can be retrieved by the call time and the corresponding user ID.

[0039] Step 320: Based on the historical call text, determine the first data and the second data of the target user; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has a call abnormality, and the third feature is used to indicate whether the target user has a comprehension abnormality.

[0040] The initial purpose of a call from a target user reflects their caller intent. This intent could be, for example, inquiring about business information, applying for early repayment, or applying for a loan. Meeting this initial purpose indicates successful business processing or an effective answer to the user's question. Failure to meet this initial purpose can occur in several situations: 1. The initial purpose is not met before the call ends. For example, regarding a specific service (e.g., penalty reduction), the current customer service representative may have limited authority and cannot directly reply to the user, requiring a report to a superior who will call back within a certain timeframe, such as 72 hours. 2. During the call, the user expresses doubt about the specific elements or procedures explained by the customer service representative. For example, some types of business operations are complex, involving many elements, and the user may have many questions, potentially leading to failure and repeat calls. 3. The customer service representative's explanation is incorrect and fails to effectively solve the user's problem. For example, the customer service representative explains that transferring money requires logging into a specific bank's system, but the user finds that the specified bank cannot be found during the actual operation. 4. Conflicts in specific procedures. For example, when a user's bank card is automatically deducted, it is not possible to specify a particular period for repayments that are still being processed.

[0041] Call anomalies reflect abnormal factors affecting the call. These anomalies can include the following situations: 1. The target user indicates they are unavailable to answer the phone. 2. The target user does not answer a callback from customer service in a timely manner. 3. The call is interrupted due to signal issues or other matters. 4. The caller's volume is too low to be heard due to equipment or environmental factors.

[0042] Anomalies in understanding reflect unresolved questions from the target user during communication. These anomalies can include the following scenarios: 1. Unclear explanations from customer service representatives, leading to difficulty in user comprehension. 2. Users repeatedly asking the same question. 3. Users explicitly stating they do not understand and do not know how to proceed. 4. Users call multiple times, with the most recent call being to a newly assigned customer service representative who is unfamiliar with the user's needs, resulting in communication breakdowns.

[0043] In some situations, such as when the historical call text does not involve the scenarios described above, the first feature can be used to indicate that the target user's initial call purpose has been met, the second feature can be used to indicate that there is no call abnormality, and the third feature can be used to indicate that there is no comprehension abnormality. It is understood that when the first feature indicates that the target user's initial call purpose has been met, and the second feature indicates that there is no call abnormality and / or the third feature indicates that there is no comprehension abnormality, the target user may not call again. In other situations, the target user may call again.

[0044] In the embodiments of this application, different machine learning models can be used to process historical call texts to determine the first and second data of the target user; alternatively, the same machine learning model can be used to process historical call texts in different stages to determine the first and second data of the target user in sequence.

[0045] Step 330: Input the first data and the second data into the pre-trained target machine learning model to predict repeat calls, and obtain the prediction result of the target user, the prediction result including whether the target user will make repeat calls.

[0046] The input to the target machine learning model can be a first feature, a second feature, and / or a third feature. The output of the target machine learning model can be a prediction result. The target machine learning model can be a highly interpretable model, such as a decision tree or a Bayesian model. The prediction result can include at least one of whether the target user will call again or the probability of the target user calling again.

[0047] In one implementation, in addition to the first and second data, the user profile of the target user (e.g., whether they like repeated calls, the number of repeated calls, or the frequency of repeated calls) can be input into the target machine learning model, so that the target machine learning model can make predictions based on the user profile, the first feature, the second feature, and / or the third feature, making the prediction results more in line with the habits of the target user.

[0048] The repeat call prediction method provided in this application can determine whether the target user's initial call purpose has been met, whether the target user has call abnormalities, and / or whether the target user has comprehension abnormalities by using the target user's historical call text. It combines multiple features and uses a trained target machine learning model to predict repeat calls, thereby quickly predicting whether a target user who has already been contacted by phone will call again.

[0049] Optionally, in one embodiment, the first data of the target user can be determined based on the historical call text by performing steps A1-A5:

[0050] Step A1: Input the historical call text into the trained intent recognition model for intent recognition to obtain at least one intent message;

[0051] The intent recognition model can process the input historical call text sentence by sentence and output the intent information corresponding to each sentence in the historical call text. For example, if the first sentence in the target user's historical call text is "Hello, I want to repay the loan early", then the intent recognition model can output the intent information: the target user intends to repay the loan early.

[0052] Step A2: Based on the at least one intent information, a historical call intent path is formed. The historical call intent path includes multiple intent nodes, and each intent node corresponds to one intent information in the at least one intent information.

[0053] Based on at least one intent message, forming a historical call intent path can include: determining an intent node for each intent message, arranging the intent nodes chronologically to form a historical call intent path. Alternatively, the intent node and its corresponding intent information can be combined to form the historical call intent path. It is understood that, to standardize customer service procedures, business units can develop node maps, which can contain standard paths for each business, where each standard path can consist of necessary nodes. Customer service representatives can accurately guide users based on the node map. In this embodiment, the intent recognition model can simultaneously output at least one intent message and the intent node corresponding to each intent message based on the historical call text and the node map. For example, if the intent node corresponding to the early repayment intent is 1, and the first sentence in the target user's historical call text is "Hello, I want to repay early," the intent recognition model can output the intent information and intent node: 1 (Target user has an early repayment intent).

[0054] Step A3: Compare and match the historical call intent path with the standard path of the target service;

[0055] The standard path for a target service can include all necessary nodes that the customer service representative and the user might encounter when the user initiates the transaction, based on the user's initial intention. These necessary nodes can be arranged in a sequential order. For example, if the target service is early repayment, the standard path could be: 1-3-5-6-7, where: 1: The user intends to repay early; 3: Customer service informs the user that the repayment conditions are met; 5: Customer service asks the user which repayment method they prefer; 6: The user chooses WeChat Pay; 7: Customer service processes the repayment for the user. Nodes 1, 3, 5, 6, and 7 are considered necessary nodes.

[0056] After obtaining the historical call intent path, the historical call intent path can be compared and matched with the standard path of the target service to determine whether the historical call intent path is consistent with the standard path of the target service.

[0057] Step A4: If the historical call intent path covers the standard path, determine the first data of the target user as a first feature indicating that the target user's initial call intent has been satisfied.

[0058] In this embodiment, the historical call intent path covering the standard path can include the following two scenarios: 1. The historical call intent path is completely consistent with the standard path; 2. The historical call intent path includes the standard path and other extended intent nodes outside the standard path. For example, taking the early repayment service described above as an example, if the historical call intent path is 1-3-5-6-7, or 1-3-5-6-7-8, 8: Customer service recommends automatic repayment when the balance is sufficient, then the historical call intent path covers the standard path, and the target user's first data can be the first feature indicating that the target user's initial call intention has been met.

[0059] Step A5: If the historical call intent path does not cover the standard path, determine the first data of the target user as a first feature indicating that the target user's call intention has not been met.

[0060] In this embodiment, the historical call intent path not covering the standard path may indicate that the historical call intent path does not contain all the necessary nodes in the standard path, or that the order of the same nodes in the historical call intent path and the standard path is inconsistent. For example, taking the early repayment service described above as an example, if the historical call intent path is 1-3-5-6 or 1-3-7-6, then the historical call path does not cover the standard path, and the first data of the target user can be a first feature used to indicate that the target user's initial call intention has not been met.

[0061] In this embodiment of the application, by inputting historical call text into the intent recognition model for intent recognition, intent information is obtained, and the historical call intent path corresponding to the intent information is compared with the standard path, which can accurately determine whether the initial intention of the target user's call has been met, which is beneficial to improving the accuracy of repeat call prediction results.

[0062] As discussed above, the historical call text in this application embodiment may include the call text of the target user's most recent calls. In determining the target user's second data, second data for each call can be obtained on a per-call basis. In one embodiment, the historical call text may include the call text of at least one call. Step 320, determining the target user's second data based on the historical call text, specifically may include: obtaining the target call text of the target user from the call text of the at least one call, where the target call text is the call text of one call; and determining the target user's second data based on the target call text.

[0063] In this embodiment of the application, by obtaining the target user's target call text from historical call texts and determining the target user's second data based solely on the target call text, the workload can be reduced to a certain extent, time can be saved, and the efficiency of repeat call prediction can be improved.

[0064] Figure 4 This is another method for predicting repeat calls provided in the embodiments of this application. For example... Figure 4 As shown, the method for predicting repeat calls provided in this application embodiment may include the following steps:

[0065] Step 410: Obtain the target user's historical call text, which includes the call text of at least one call.

[0066] Step 420: Determine the first data of the target user based on the call text of the at least one call;

[0067] The first data includes a first feature, which is used to indicate whether the target user's initial intention for the call has been met.

[0068] Step 430: Obtain the target call text of the target user from the call text of the at least one call, wherein the target call text is the call text of one call;

[0069] Step 440: Perform single-sentence anomaly identification on the target call text sentence by sentence to obtain the anomaly identification result for each sentence in the target call text;

[0070] The anomaly detection result can include whether a single sentence in the target call text contains call anomalies and / or comprehension anomalies. The anomaly detection result can be represented in the form of a feature value (e.g., 0 or 1). For example, a feature value of 0 or 1 indicates that the single sentence text does not contain anomalies, while a value of 1 indicates that the single sentence text contains anomalies. In this embodiment, the anomaly detection result is specific to each sentence of text; that is, an anomaly detection result can be obtained for each sentence in the target call text.

[0071] Step 450: Based on the anomaly identification results of each sentence in the target call text, determine the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly; wherein, the number of times a single sentence anomaly occurs is the number of anomalies obtained in the target call text with one sentence as the statistical unit;

[0072] The number of times a single sentence anomaly occurs in the target call text can be the sum of the anomaly identification results corresponding to the anomaly in that single sentence. The location of each single sentence anomaly can be the specific position of the sentence with the anomaly in the target call text. For example, if the anomaly identification results for the third and seventh sentences in the target call text are that anomalies exist, while the identification results for other sentences are that no anomalies exist, then the number of times a single sentence anomaly occurs in the target call text can be 2, and the location of the single sentence anomaly can be the third and seventh sentences of the target call text.

[0073] Step 460: Based on the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly, determine the second data of the target user;

[0074] The second data includes a second feature and / or a third feature, wherein the second feature is used to indicate whether the target user has a call abnormality, and the third feature is used to indicate whether the target user has a comprehension abnormality.

[0075] The second feature in the second data can be used to reflect the overall anomaly situation of the target call text. In the embodiments of this application, a weighting formula or a machine learning model for anomaly identification can be used to determine the second data of the target user. For example, when using a weighting formula, different weights can be assigned to the number of times a single sentence anomaly occurs and the location of each single sentence call anomaly, and the second data is calculated by combining the number of anomalies and the location. At this time, the second data can be equivalent to the second feature and / or the third feature. When the second data meets a certain condition, such as the statistical result of a certain dimension of the second data being greater than or equal to a threshold, it can be determined that the target user has a call anomaly or a comprehension anomaly. When the second data does not meet a certain condition, such as the statistical result of a certain dimension of the second data being less than a threshold, it is determined that the target user does not have a call anomaly or a comprehension anomaly. When using a machine learning model for anomaly identification, the number of times a single sentence anomaly occurs and the location of each single sentence anomaly occur can be input into the machine learning model for anomaly identification to obtain the second data.

[0076] Step 470: Input the first data and the second data into the pre-trained target machine learning model to predict repeat calls, and obtain the prediction result of the target user, the prediction result including whether the target user will make repeat calls.

[0077] The input to the target machine learning model can be a first feature, a second feature, and / or a third feature. The output of the target machine learning model can be a prediction result. The target machine learning model can be a highly interpretable model, such as a decision tree or a Bayesian model. The prediction result can include at least one of whether the target user will call again or the probability of the target user calling again.

[0078] The repeat call prediction method provided in this application determines the anomaly identification result of each sentence in the target call text, and further determines the second data based on the number and location of the anomalies in a single sentence. This can obtain features that can accurately reflect whether there are call anomalies and / or comprehension anomalies, thereby improving the accuracy of repeat call prediction results to a certain extent.

[0079] Since the text of the most recent call among the target user's recent calls better reflects whether the target user's needs have been met, using the text of the most recent call to predict repeat calls can achieve better results. In one embodiment, the target call text is the text of the target user's most recent call; step 440, which involves performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result for each sentence, may specifically include: inputting the most recent call text into a pre-trained first detection model, and using the first detection model to perform sentence-by-sentence anomaly identification on the most recent call text to obtain the anomaly identification result for each sentence; step 460, which involves determining the target user's second data based on the number of times anomalies occur in the most recent call text and the location of each anomaly, may specifically include: inputting the number of times a single sentence anomaly occurs in the most recent call text and the location of each single sentence anomaly into a pre-trained second detection model to obtain the target user's second data.

[0080] In this embodiment, the first detection model can be used for single-sentence call anomaly detection and / or comprehension anomaly detection. The second detection model can be used to determine whether there are call anomalies and / or comprehension anomalies in the most recent call text.

[0081] In this embodiment of the application, by determining the text of the target user's most recent call as the target call text, the second data of the target user can be obtained more effectively, avoiding the introduction of unnecessary features. At the same time, the efficiency can be improved to a certain extent by performing anomaly identification based on the first detection model and determining whether there are any anomalies in the text of the most recent call based on the second detection model.

[0082] The second data in the embodiments of this application may include a second feature and / or a third feature. The following discussion addresses the cases where the second data includes the second feature and the cases where the second data includes the third feature.

[0083] In one embodiment, the second data includes a second feature indicating whether the target user has a call anomaly. In this case, step 440, which involves performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result for each sentence, may include: inputting the target call text into a pre-trained first detection model, and using the first detection model to perform sentence-by-sentence call anomaly identification on the target call text to obtain the call anomaly identification result for each sentence. Correspondingly, step 450, which involves determining the number of occurrences of a single-sentence anomaly in the target call text and the location of each occurrence of a single-sentence anomaly based on the anomaly identification result for each sentence, may include: determining the number of occurrences of a single-sentence call anomaly in the target call text and the location of each occurrence of a single-sentence call anomaly based on the call anomaly identification result for each sentence. Step 460, which involves determining the second data of the target user based on the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly, may include: inputting the number of times a single sentence call anomaly occurs in the target call text and the location of each single sentence call anomaly into a pre-trained second detection model to obtain the second feature of the target user, wherein the second feature is used to indicate whether there is a call anomaly in the entire call of the target user.

[0084] In this embodiment, a first detection model is used to identify single-sentence call anomalies, and a second detection model is used to determine whether there are call anomalies in the entire call of the target user. This can improve the efficiency and accuracy of the second data acquisition to a certain extent, facilitate the rapid acquisition of the second feature, and further improve the efficiency of repeat call prediction.

[0085] In another embodiment, the second data includes a third feature indicating whether the target user has comprehension anomalies. In this case, step 440, which involves performing sentence-by-sentence anomaly identification on the target call text to obtain anomaly identification results for each sentence, may include: inputting the target call text into a pre-trained third detection model, and using the third detection model to perform sentence-by-sentence comprehension anomaly identification on the target call text to obtain comprehension anomaly identification results for each sentence. Step 450, which involves determining the number of occurrences of sentence-by-sentence anomalies and the location of each occurrence based on the anomaly identification results for each sentence in the target call text, may include: determining the number of occurrences of sentence-by-sentence comprehension anomalies and the location of each occurrence based on the comprehension anomaly identification results for each sentence in the target call text. The step 460, which determines the second data of the target user based on the number of times the single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly, may include: inputting the number of times the single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly occurs into a pre-trained fourth detection model to obtain the third feature of the target user, wherein the third feature is used to indicate whether there is a comprehension anomaly in the entire call of the target user.

[0086] In this embodiment, a third detection model is used to identify single-sentence comprehension anomalies, and a fourth detection model is used to determine whether there are comprehension anomalies in the entire call of the target user. This can improve the efficiency and accuracy of the second data acquisition to a certain extent, facilitate the rapid acquisition of the third feature, and further improve the efficiency of repeat call prediction.

[0087] The foregoing has described the specific process of predicting repeat calls using a target machine learning model based on the first and second data. It should be understood that the repeat call prediction method provided in this application can use multiple machine learning models. For example, a first detection model, a second detection model, a third detection model, a fourth detection model, an intent recognition model, and a target machine learning model. Specifically, the first detection model can be used to detect call anomalies in the call text; the second detection model can be used to detect whether there are call anomalies in the entire call; the third detection model can be used to detect comprehension anomalies in the call text; the fourth detection model can be used to detect comprehension anomalies in the entire call; the intent recognition model can be used to identify intent in the call text and determine intent information; and the target machine learning model can be used to predict whether a user will receive repeat calls based on the outputs of the intent recognition model, the second detection model, and / or the fourth detection model.

[0088] The training process of the various machine learning models mentioned above will be discussed further below.

[0089] In one embodiment of this application, the training process of the first detection model may include: acquiring M first training samples, each first training sample corresponding to a text sentence with a pre-annotated single-sentence call anomaly, where M is an integer greater than 2; training a first pre-trained model using the M first training samples to obtain the first detection model; the training process of the second detection model includes: acquiring N second training samples, each second training sample including the number of times a single-sentence call anomaly occurs in the text of a call and the position of each single-sentence call anomaly, where N is an integer greater than 2; training a second pre-trained model using the N second training samples to obtain the second detection model.

[0090] In this embodiment, the first detection model can detect call anomalies for each individual sentence in the call text. The input to the first detection model can be the call text. The output of the first detection model can include the number of times an anomaly occurs in a single sentence of the call text and the location of each anomaly. The second detection model can be used to detect whether there are any call anomalies in the entire call.

[0091] In this embodiment, the first pre-trained model is trained using M single-sentence texts labeled with single-sentence call anomalies, enabling the trained first detection model to accurately identify single-sentence call anomalies. Meanwhile, the second pre-trained model is trained using the number of times single-sentence call anomalies occur and the location of each single-sentence call anomaly, enabling the trained second detection model to accurately predict whether there are call anomalies in the entire call.

[0092] Optionally, in one embodiment of this application, the training process of the third detection model includes: obtaining P third training samples, each third training sample corresponding to a text with pre-annotated single-sentence comprehension anomalies, where P is an integer greater than 2; training the third pre-trained model using the P third training samples to obtain the third detection model; the training process of the fourth detection model includes: obtaining Q fourth training samples, each fourth training sample including the number of times single-sentence comprehension anomalies occur in the text of a call and the position of each single-sentence comprehension anomaly, where Q is an integer greater than 2; training the fourth pre-trained model using the Q fourth training samples to obtain the fourth detection model.

[0093] In this embodiment, the third detection model can detect comprehension anomalies for each individual sentence in the call text. The input to the third detection model can be the call text. The output of the third detection model can include the number of times comprehension anomalies occur in individual sentences in a call text and the location of each comprehension anomaly. The fourth detection model can be used to detect whether there are comprehension anomalies in the entire call.

[0094] In this embodiment, the third pre-trained model is trained using P single-sentence texts labeled with single-sentence comprehension anomalies, enabling the trained third detection model to accurately identify single-sentence comprehension anomalies. Meanwhile, the fourth pre-trained model is trained using the number of occurrences of single-sentence comprehension anomalies and the location of each occurrence, enabling the trained fourth detection model to accurately predict whether there are comprehension anomalies in the entire call.

[0095] Optionally, in one embodiment of this application, the second data includes a second feature and a third feature, and the training process of the target machine learning model includes: acquiring L fifth training samples, each fifth training sample including first indication information, second indication information, third indication information, and fourth indication information; the first indication information is used to indicate whether the initial purpose of the call from the target user is met, the second indication information is used to indicate whether the entire call of the target user has a call abnormality, the third indication information is used to indicate whether the entire call of the target user has a comprehension abnormality, and the fourth indication information is used to indicate whether the target user has repeated calls; the fifth pre-trained model is trained using the L fifth training samples to obtain the target machine learning model.

[0096] The fourth instruction information can be information determined based on the target user's user profile (e.g., whether they like repeated calls, the number of repeated calls, or the frequency of repeated calls).

[0097] In this embodiment of the application, the fifth pre-trained model is trained by combining multiple factors such as the target user's initial call intention, call abnormality, comprehension abnormality, and repeated calls, which enables the trained target machine learning model to have a better prediction accuracy.

[0098] Optionally, in one embodiment of this application, the training process of the intent recognition model includes: acquiring K target training samples, each target training sample corresponding to a sentence of text with an labeled intent; and training a sixth pre-trained model using the K target training samples to obtain the intent recognition model.

[0099] In this embodiment, the sixth pre-trained model is trained using K pre-annotated single-sentence texts to obtain an intent recognition model that can accurately identify text intent, thereby improving the accuracy of repeat call prediction to a certain extent.

[0100] Figure 5 This is a flowchart of another method for predicting repeat calls provided in an embodiment of this application. For example... Figure 5 As shown, the method for predicting repeat calls provided in this application embodiment may include the following steps:

[0101] Step 505: Obtain the target user's historical call text, which includes the call text of at least one call.

[0102] Step 510: Determine the first data of the target user based on the call text of the at least one call;

[0103] The first data includes a first feature, which is used to indicate whether the target user's initial intention for the call has been met.

[0104] Step 515: Obtain the target call text of the target user from the call text of the at least one call, wherein the target call text is the call text of one call;

[0105] Step 520: Input the target call text into the pre-trained first detection model, and use the first detection model to perform single-sentence call anomaly identification on the target call text sentence by sentence to obtain the call anomaly identification result for each sentence in the target call text;

[0106] Step 525: Based on the call anomaly identification results of each sentence in the target call text, determine the number of times a single sentence call anomaly occurs in the target call text and the location of each single sentence call anomaly.

[0107] The number of times a single-sentence call anomaly occurs can be the number of call anomalies obtained from the target call text, with each sentence as the statistical unit.

[0108] Step 530: Input the number of times a single-sentence call anomaly occurs in the target call text and the location of each single-sentence call anomaly into the trained second detection model to obtain the second feature of the target user. The second feature is used to indicate whether there is a call anomaly in the entire call of the target user.

[0109] Step 535: Input the target call text into the trained third detection model, and use the third detection model to perform single-sentence comprehension anomaly identification on the target call text sentence by sentence to obtain the comprehension anomaly identification result for each sentence in the target call text;

[0110] Step 540: Based on the anomaly recognition results of each sentence in the target call text, determine the number of times an anomaly in the single sentence in the target call text occurs and the location of each anomaly in the single sentence.

[0111] The number of times a single sentence comprehension anomaly occurs can be the number of comprehension anomalies obtained by taking a single sentence as the statistical unit in the target call text.

[0112] Step 545: Input the number of times single-sentence comprehension anomalies occur in the target call text and the location of each single-sentence comprehension anomaly into the trained fourth detection model to obtain the third feature of the target user. The third feature is used to indicate whether there are comprehension anomalies in the entire call of the target user.

[0113] Step 550: Input the first feature, the second feature and the third feature into the pre-trained target machine learning model to predict repeat calls, and obtain the prediction result of the target user, the prediction result including whether the target user will make repeat calls.

[0114] In this embodiment, the system can determine whether the target user's initial call intention has been met, whether there are call abnormalities and / or comprehension abnormalities by analyzing the target user's historical call text. Combining the characteristics of the initial call intention, call abnormalities and / or comprehension abnormalities, the system can use a trained target machine learning model to predict repeat calls, thereby quickly predicting whether a target user with whom a phone call has already been made will call again.

[0115] Figure 6 This is a structural block diagram of a repeat call prediction device provided in an embodiment of this application. As shown in Figure 6, the repeat call prediction device 600 provided in this embodiment of the application may include: an acquisition module 610, a determination module 620, and a processing module 630;

[0116] The acquisition module 610 is used to acquire the target user's historical call text;

[0117] The determining module 620 is used to determine first data and second data of the target user based on the historical call text; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has a call abnormality, and the third feature is used to indicate whether the target user has a comprehension abnormality.

[0118] The processing module 630 is used to input the first data and the second data into a pre-trained target machine learning model to predict repeat calls, and to obtain a prediction result for the target user, the prediction result including whether the target user will make repeat calls.

[0119] The repeat call prediction device provided in this application embodiment can determine whether the target user's initial call intention has been met, whether the target user has call abnormalities, and / or whether the target user has comprehension abnormalities by using the target user's historical call text. Combining the characteristics of the initial call intention, call abnormalities, and / or comprehension abnormalities, the device uses a trained target machine learning model to predict repeat calls, thereby quickly predicting whether a target user with whom a phone call has already been made will call again.

[0120] Optionally, in one embodiment, during the process of determining the first data of the target user based on the historical call text, the determining module 620 is specifically configured to: input the historical call text into a trained intent recognition model for intent recognition to obtain at least one intent information; form a historical call intent path based on the at least one intent information, the historical call intent path including multiple intent nodes, each intent node corresponding to one intent information in the at least one intent information; compare and match the historical call intent path with the standard path of the target service; if the historical call intent path covers the standard path, determine the first data of the target user as a first feature indicating that the initial intention of the target user's call has been satisfied; if the historical call intent path does not cover the standard path, determine the first data of the target user as a first feature indicating that the initial intention of the target user's call has not been satisfied. Thus, by inputting the historical call text into the intent recognition model for intent recognition to obtain intent information, and comparing the historical call intent path corresponding to the intent information with the standard path, it is possible to accurately determine whether the initial intention of the target user's call has been satisfied, which helps to improve the accuracy of repeat call prediction results.

[0121] Optionally, in one embodiment, the historical call text includes the call text of at least one call; in the process of determining the second data of the target user based on the historical call text, the determining module 620 is specifically used to: obtain the target call text of the target user from the call text of the at least one call, wherein the target call text is the call text of one call; and determine the second data of the target user based on the target call text. Thus, by obtaining the target call text of the target user from the historical call text, and determining the second data of the target user solely based on the target call text, the workload can be reduced to a certain extent, time can be saved, and the efficiency of repeat call prediction can be improved.

[0122] Optionally, in one embodiment, during the process of determining the second data of the target user based on the target call text, the determining module 620 is specifically configured to: perform single-sentence anomaly identification on each sentence of the target call text to obtain the anomaly identification result for each sentence in the target call text; determine the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly based on the anomaly identification result for each sentence in the target call text; and determine the second data of the target user based on the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly. Thus, by determining the anomaly identification result for each sentence in the target call text and further determining the second data based on the number and location of single sentence anomalies, features that accurately reflect whether call anomalies and / or comprehension anomalies exist can be obtained, thereby improving the accuracy of repeat call prediction results to a certain extent.

[0123] Optionally, in one embodiment, the second data includes a second feature indicating whether the target user has a call anomaly. In the process of performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result for each sentence in the target call text, the determining module 620 is specifically configured to: input the target call text into a pre-trained first detection model; perform sentence-by-sentence call anomaly identification on the target call text using the first detection model to obtain the call anomaly identification result for each sentence in the target call text; and, based on the anomaly identification result for each sentence in the target call text, determine the number of times a single-sentence anomaly occurs in the target call text and the number of times each single-sentence anomaly occurs. In the process of determining the location, the determining module 620 is specifically used to: determine the number of times a single-sentence call anomaly occurs in the target call text and the location of each single-sentence call anomaly based on the call anomaly identification result of each sentence in the target call text; in the process of determining the second data of the target user based on the number of times a single-sentence call anomaly occurs in the target call text and the location of each single-sentence call anomaly, the determining module 620 is specifically used to: input the number of times a single-sentence call anomaly occurs in the target call text and the location of each single-sentence call anomaly into the pre-trained second detection model to obtain the second feature of the target user, the second feature being used to indicate whether there is a call anomaly in the entire call of the target user. Thus, using the first detection model for single-sentence call anomaly identification and using the second detection model to determine whether there is a call anomaly in the entire call of the target user can improve the efficiency and accuracy of the second data acquisition to a certain extent, facilitate the rapid acquisition of the second feature, and further improve the efficiency of repeat call prediction.

[0124] Optionally, in one embodiment, the second data includes a third feature indicating whether the target user has a comprehension abnormality. In the process of performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result for each sentence in the target call text, the determining module 620 is specifically configured to: input the target call text into a pre-trained third detection model, perform sentence-by-sentence comprehension anomaly identification on the target call text using the third detection model, and obtain the comprehension anomaly identification result for each sentence in the target call text; and, based on the anomaly identification result for each sentence in the target call text, determine the number of times a single sentence anomaly occurs in the target call text and the number of times each single sentence anomaly occurs. In the process of determining the location, the determining module 620 is specifically used to: determine the number of times a single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly based on the comprehension anomaly identification result of each sentence in the target call text; in the process of determining the second data of the target user based on the number of times a single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly, the determining module 620 is specifically used to: input the number of times a single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly into the pre-trained fourth detection model to obtain the third feature of the target user, the third feature being used to indicate whether there is a comprehension anomaly in the entire call of the target user. Thus, using the third detection model for single sentence comprehension anomaly identification and using the fourth detection model to determine whether there is a comprehension anomaly in the entire call of the target user can improve the efficiency and accuracy of second data acquisition to a certain extent, facilitate the rapid acquisition of the third feature, and further improve the efficiency of repeat call prediction.

[0125] Optionally, in one embodiment, the target call text is the text of the target user's most recent call. In the process of performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result for each sentence, the determining module 620 is specifically used to: input the most recent call text into a pre-trained first detection model, and perform sentence-by-sentence anomaly identification on the most recent call text using the first detection model to obtain the anomaly identification result for each sentence. In the process of determining the target user's second data based on the number of times anomalies occur and the position of each anomaly in the most recent call text, the determining module 620 is specifically used to: input the number of times a single-sentence anomaly occurs and the position of each single-sentence anomaly in the most recent call text into a pre-trained second detection model to obtain the target user's second data. Thus, by determining the target user's most recent call text as the target call text, the target user's second data can be obtained more specifically, avoiding the introduction of unnecessary features. Simultaneously, performing anomaly identification based on the first detection model and determining whether there are anomalies in the most recent call text based on the second detection model can improve efficiency to a certain extent.

[0126] Optionally, in one embodiment, the repeat call prediction device 600 further includes a first sub-acquisition module, a first sub-training module, a second sub-acquisition module, and a second sub-training module; the first sub-acquisition module is used to acquire M first training samples, each first training sample corresponding to a text sentence with a pre-annotated single-sentence call anomaly, where M is an integer greater than 2; the first sub-training module is used to train a first pre-trained model using the M first training samples to obtain the first detection model; the second sub-acquisition module is used to acquire N second training samples, each second training sample including the number of times a single-sentence call anomaly occurs in the text of a call and the position of each single-sentence call anomaly, where N is an integer greater than 2; the second sub-training module is used to train a second pre-trained model using the N second training samples to obtain the second detection model. Thus, by training the first pre-trained model with M pre-annotated single-sentence call anomalies, the trained first detection model can accurately identify single-sentence call anomalies. At the same time, by training the second pre-trained model with the number of times single-sentence call anomalies occur and the location of each single-sentence call anomaly, the trained second detection model can accurately predict whether there are call anomalies in the entire call.

[0127] Optionally, in one embodiment, the repeat call prediction device 600 further includes a third sub-acquisition module, a third sub-training module, a fourth sub-acquisition module, and a fourth sub-training module; the third sub-acquisition module is used to acquire P third training samples, each third training sample corresponding to a text sentence with pre-annotated single-sentence comprehension anomalies, where P is an integer greater than 2; the third sub-training module is used to train a third pre-trained model using the P third training samples to obtain the third detection model; the fourth sub-acquisition module is used to acquire Q fourth training samples, each fourth training sample including the number of times single-sentence comprehension anomalies occur in the text of a call and the position of each single-sentence comprehension anomaly, where Q is an integer greater than 2; the fourth sub-training module is used to train a fourth pre-trained model using the Q fourth training samples to obtain the fourth detection model. Thus, by training the third pre-trained model with P single-sentence texts that have been labeled with single-sentence comprehension anomalies, the trained third detection model can accurately identify single-sentence comprehension anomalies. At the same time, by training the fourth pre-trained model with the number of times single-sentence comprehension anomalies occur and the location of each single-sentence comprehension anomaly, the trained fourth detection model can accurately predict whether there are comprehension anomalies in the entire call.

[0128] Optionally, in one embodiment, the second data includes a second feature and a third feature, and the repeat call prediction device 600 further includes a fifth sub-acquisition module and a fifth sub-training module; the fifth sub-acquisition module is used to acquire L fifth training samples, each fifth training sample including first indication information, second indication information, third indication information, and fourth indication information; the first indication information is used to indicate whether the target user's initial call intention is met, the second indication information is used to indicate whether the target user's entire call has call abnormalities, the third indication information is used to indicate whether the target user's entire call has comprehension abnormalities, and the fourth indication information is used to indicate whether the target user has repeated calls; the fifth sub-training module is used to train the fifth pre-trained model using the L fifth training samples to obtain the target machine learning model. Thus, by combining multiple factors such as the target user's initial call intention, call abnormalities, comprehension abnormalities, and repeat calls to train the fifth pre-trained model, the trained target machine learning model can achieve better prediction accuracy.

[0129] Optionally, in one embodiment, the repeat call prediction device 600 further includes a sixth sub-acquisition module and a sixth sub-training module; the sixth sub-acquisition module is used to acquire K target training samples, each target training sample corresponding to a sentence of text with pre-annotated intent; the sixth sub-training module is used to train a sixth pre-trained model using the K target training samples to obtain the intent recognition model. Thus, by using K pre-trained sentences of text with pre-annotated intent to train the sixth pre-trained model, an intent recognition model that can accurately identify text intent can be obtained, thereby improving the accuracy of repeat call prediction to a certain extent.

[0130] It should be noted that the repeat call prediction device provided in this application corresponds to the repeat call prediction method mentioned above. Related details can be found in the description of the repeat call prediction method above, and will not be repeated here.

[0131] In addition, such as Figure 7 As shown in the illustration, this application also provides an electronic device 700, which can be various types of computers, etc. The electronic device 700 includes a processor 710 and a memory 720. The memory 720 stores programs or instructions, which, when executed by the processor 710, implement the steps of any of the methods described above. For example, when the program is executed by the processor 720, it implements the following process: acquiring the historical call text of a target user; determining first data and second data of the target user based on the historical call text; wherein the first data includes a first feature, which indicates whether the initial purpose of the target user's call has been met; the second data includes a second feature and / or a third feature, where the second feature indicates whether the target user has a call abnormality, and the third feature indicates whether the target user has a comprehension abnormality; and inputting the first data and the second data into a pre-trained target machine learning model to predict repeated calls, obtaining a prediction result for the target user, the prediction result including whether the target user will make repeated calls. In this way, the target user's original intention for calling can be determined by the target user's historical call text, whether the target user has call abnormalities and / or comprehension abnormalities, and combined with the characteristics of the original intention for calling, call abnormalities and / or comprehension abnormalities, the trained target machine learning model can be used to predict repeat calls, so as to quickly predict whether the target user who has been communicated by phone will call again.

[0132] This application embodiment also provides a readable storage medium storing a program or instructions that, when executed by the processor 710, implement the steps of any of the methods described above. For example, when the program is executed by the processor 710, it performs the following process: Obtaining the historical call text of a target user. Based on the historical call text, determining first data and second data of the target user. The first data includes a first feature indicating whether the target user's initial call intention was met. The second data includes a second feature and / or a third feature, where the second feature indicates whether the target user has call abnormalities, and the third feature indicates whether the target user has comprehension abnormalities. The first data and the second data are input into a pre-trained target machine learning model for repeat call prediction, obtaining a prediction result for the target user, the prediction result including whether the target user will make repeat calls. In this way, the target user's original intention for calling can be determined by the target user's historical call text, whether the target user has call abnormalities and / or comprehension abnormalities, and combined with the characteristics of the original intention for calling, call abnormalities and / or comprehension abnormalities, the trained target machine learning model can be used to predict repeat calls, so as to quickly predict whether the target user who has been communicated by phone will call again.

[0133] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0138] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0139] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0140] It should also be noted that 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 limitation, 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.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The above description is merely an embodiment of this application and is not intended to limit the scope 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 the claims of this application.

Claims

1. A method for predicting repeat calls, characterized in that, include: Obtain the target user's historical call text; Based on the historical call text, first data and second data of the target user are determined; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has call abnormalities, and the third feature is used to indicate whether the target user has comprehension abnormalities. The first data and the second data are input into a pre-trained target machine learning model to predict repeat calls, thereby obtaining a prediction result for the target user, which includes whether the target user will make repeat calls.

2. The prediction method according to claim 1, characterized in that, The first data for determining the target user based on the historical call text includes: The historical call text is input into a trained intent recognition model for intent recognition to obtain at least one intent message. Based on the at least one intent information, a historical call intent path is formed, the historical call intent path including multiple intent nodes, each intent node corresponding to one intent information in the at least one intent information; The historical call intent path is compared and matched with the standard path of the target service; When the historical call intent path covers the standard path, the first data of the target user is determined to be a first feature indicating that the target user's initial call intent has been satisfied; If the historical call intent path does not cover the standard path, the first data of the target user is determined to be a first feature indicating that the target user's call intention has not been met.

3. The prediction method according to claim 1, characterized in that, The historical call text includes the call text of at least one call; The second data for determining the target user based on the historical call text includes: Obtain the target call text of the target user from the call text of the at least one call, wherein the target call text is the call text of one call; Based on the target call text, second data of the target user is determined.

4. The prediction method according to claim 3, characterized in that, The step of determining the second data of the target user based on the target call text includes: Sentence-by-sentence anomaly identification is performed on the target call text to obtain the anomaly identification result for each sentence in the target call text; Based on the anomaly identification results of each sentence in the target call text, the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly occurrence are determined; wherein, the number of times a single sentence anomaly occurs is the number of anomalies obtained in the target call text with each sentence as the statistical unit; The second data of the target user is determined based on the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly.

5. The prediction method according to claim 4, characterized in that, The second data includes a second feature used to indicate whether the target user has a call anomaly. The step of performing single-sentence anomaly identification on the target call text sentence by sentence to obtain the anomaly identification result of each sentence in the target call text includes: inputting the target call text into a pre-trained first detection model, and performing single-sentence call anomaly identification on the target call text sentence by sentence through the first detection model to obtain the call anomaly identification result of each sentence in the target call text; The step of determining the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly based on the anomaly identification result of each sentence in the target call text includes: determining the number of times a single sentence call anomaly occurs in the target call text and the location of each single sentence call anomaly based on the call anomaly identification result of each sentence in the target call text. The step of determining the second data of the target user based on the number of times the single sentence anomaly occurs in the target call text and the location of each single sentence anomaly occurs includes: inputting the number of times the single sentence call anomaly occurs in the target call text and the location of each single sentence call anomaly occurs into a pre-trained second detection model to obtain the second feature of the target user, the second feature being used to indicate whether there is a call anomaly in the entire call of the target user.

6. The prediction method according to claim 4, characterized in that, The second data includes a third feature to indicate whether the target user has a comprehension disorder. The step of performing sentence-by-sentence anomaly identification on the target call text to obtain the anomaly identification result of each sentence in the target call text includes: inputting the target call text into a pre-trained third detection model, and performing sentence-by-sentence comprehension anomaly identification on the target call text through the third detection model to obtain the comprehension anomaly identification result of each sentence in the target call text. The step of determining the number of times a single sentence anomaly occurs in the target call text and the location of each single sentence anomaly based on the anomaly identification result of each sentence in the target call text includes: determining the number of times a single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly based on the comprehension anomaly identification result of each sentence in the target call text. The step of determining the second data of the target user based on the number of times the single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly occurs includes: inputting the number of times the single sentence comprehension anomaly occurs in the target call text and the location of each single sentence comprehension anomaly occurs into a pre-trained fourth detection model to obtain the third feature of the target user, the third feature being used to indicate whether there is a comprehension anomaly in the entire call of the target user.

7. The prediction method according to claim 5, characterized in that, The training process of the first detection model includes: Obtain M first training samples, each of which corresponds to a text that has been labeled with a single sentence of call anomaly, where M is an integer greater than 2; The first detection model is obtained by training the first pre-trained model using the M first training samples; The training process of the second detection model includes: Obtain N second training samples. Each second training sample includes the number of times a single sentence call anomaly occurs in the call text of a call and the position of each single sentence call anomaly. N is an integer greater than 2. The second detection model is obtained by training the second pre-trained model using the N second training samples.

8. The prediction method according to claim 6, characterized in that, The training process of the third detection model includes: Obtain P third training samples, each of which corresponds to a text sentence with pre-annotated single-sentence comprehension anomalies, where P is an integer greater than 2; The third detection model is obtained by training the third pre-trained model with the P third training samples; The training process of the fourth detection model includes: Obtain Q fourth training samples. Each fourth training sample includes the number of times single-sentence comprehension anomalies occur in the text of a call and the location of each single-sentence comprehension anomaly. Q is an integer greater than 2. The fourth detection model is obtained by training the fourth pre-trained model using the Q fourth training samples.

9. The prediction method according to claim 1, characterized in that, The second data includes a second feature and a third feature, and the training process of the target machine learning model includes: Obtain L fifth training samples, each fifth training sample including first indication information, second indication information, third indication information and fourth indication information; the first indication information is used to indicate whether the initial purpose of the call of the target user is met, the second indication information is used to indicate whether the entire call of the target user has call abnormality, the third indication information is used to indicate whether the entire call of the target user has comprehension abnormality, and the fourth indication information is used to indicate whether the target user has repeated calls. The target machine learning model is obtained by training the fifth pre-trained model using the L fifth training samples.

10. A device for predicting repeat calls, characterized in that, include: The acquisition module is used to acquire the historical call text of the target user; The determining module is used to determine first data and second data of the target user based on the historical call text; wherein, the first data includes a first feature, which is used to indicate whether the target user's initial call purpose has been met; the second data includes a second feature and / or a third feature, whereby the second feature is used to indicate whether the target user has call abnormalities, and the third feature is used to indicate whether the target user has comprehension abnormalities. The processing module is used to input the first data and the second data into a pre-trained target machine learning model to predict repeat calls, and to obtain the prediction result of the target user, the prediction result including whether the target user will make repeat calls.

11. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that, when executed by the processor, implement the steps of the method as described in any one of claims 1-9.

12. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1-9.

Images