Prediction method and device for collection style information, equipment, storage medium and program product

Abstract

The invention relates to a method and device for predicting collection style information, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring first basic feature data of a predicted user; the first basic feature data is input into a prediction model, a prediction result output by the prediction model is obtained, and the prediction result comprises target collection style information corresponding to a prediction user; the target collection style information is determined based on the target verbal skill data; the prediction model stores a mapping relationship between the target verbal skill data and the customer group type; the mapping relation is determined based on the second basic feature data of the sample user and the initial verbal skill data in each scene; the initial verbal skill data in each scene is output based on a scene classification model, and the scene classification model is obtained based on second basic feature data training. By adopting the method, the adaptability to customers and the resource utilization rate of the intelligent robot can be improved while the flexibility of the collection robot is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting collection style information. Background Art

[0002] With the development of artificial intelligence technology, intelligent collection robots have emerged. Intelligent collection robots can automatically execute the collection process, including automatic outbound calls, automatic text messages, automatic emails, etc. Through automation, manual operations can be greatly reduced and collection efficiency can be improved.

[0003] In traditional technology, the collection style, collection process and collection language of intelligent collection robots are rigid and single, and they operate through fixed collection processes and collection language.

[0004] However, the current collection methods or traditional methods have a single collection style and low user adaptability, resulting in poor flexibility in collection style, leading to low resource utilization of intelligent robots. Summary of the invention

[0005] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for predicting collection style information that can improve the flexibility and resource utilization of intelligent robots in response to the above-mentioned technical problems.

[0006] In a first aspect, the present application provides a method for predicting collection style information, the method comprising:

[0007] Acquire first basic characteristic data of the predicted user;

[0008] The first basic feature data is input into the prediction model to obtain the prediction result output by the prediction model, wherein the prediction result includes the target collection style information corresponding to the predicted user; the target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is output based on the scenario classification model, and the scenario classification model is used to classify the collection manual interaction text.

[0009] In one embodiment, the mapping relationship is generated by:

[0010] For each of the initial speech data in the scenario, determining the second basic feature data associated with the scenario; wherein the initial speech data includes initial speech content and an initial speech sequence corresponding to the initial speech content;

[0011] Determining the customer group type according to the second basic characteristic data;

[0012] The initial speech content and the initial speech order in each scenario are sorted and superimposed to determine each target speech content and the target speech order corresponding to the target speech content;

[0013] The target speech content, the target speech sequence and the customer group type are associated to determine a mapping relationship between the target speech data and the customer group type, wherein the target speech data includes the target speech content and the target speech sequence.

[0014] In one embodiment, the method for determining the initial speech data includes:

[0015] Obtaining a target collection manual interaction text for a target time period; wherein the target collection manual interaction text indicates a text containing user repayment data within a preset time after the collection manual interaction;

[0016] Performing semantic understanding on the target collection manual interaction text, obtaining semantics corresponding to each text in the target collection manual interaction text, and counting the number of the semantics;

[0017] According to the number of the semantics, word embedding encoding is performed on the semantics corresponding to each text to determine the vector representation of the target collection manual interaction text;

[0018] Performing intent labeling on the vector representation according to the sample labeling;

[0019] The vector representations after the intent annotation are classified to obtain the initial speech data in each scenario.

[0020] In one embodiment, obtaining the target collection manual interaction text in the target time period includes:

[0021] Obtain historical collection manual interaction texts and historical user repayment data for the target time period;

[0022] For each target user corresponding to the historical user repayment data, determining the historical collection manual interaction text corresponding to the target user;

[0023] The historical collection manual interaction text corresponding to the target user is used as the target collection manual interaction text in the target time period.

[0024] In one embodiment, the method further comprises:

[0025] Determining a customer group type according to the first basic characteristic data of the predicted user;

[0026] Determining a feature vector according to the customer group type;

[0027] The feature vector is processed through the mapping relationship between the target speech data and the customer group type to obtain the target speech data corresponding to the predicted user;

[0028] Target collection style information is determined based on the target speech data.

[0029] In one embodiment, determining the feature vector according to the customer group type includes:

[0030] Determine the feature field according to the customer group type;

[0031] Acquire a first feature field that is a null value from the feature fields, and predict the first feature field to obtain a fitting value of the first feature field;

[0032] The fitted value of the first feature field and the actual value of the second feature field are encoded, normalized, and vectorized to obtain the feature vector; wherein the second feature field indicates a feature field that is not a null value.

[0033] In a second aspect, the present application provides a device for predicting collection style information, the device comprising:

[0034] An acquisition module, used to acquire first basic characteristic data of the predicted user;

[0035] An input-output module is used to input the first basic feature data into a prediction model to obtain a prediction result output by the prediction model, wherein the prediction result includes target collection style information corresponding to the predicted user; the target collection style information is determined based on target speech data; the prediction model stores a mapping relationship between target speech data and customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is output based on a scenario classification model, and the scenario classification model is used to classify collection manual interaction texts.

[0036] In a third aspect, the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0038] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the steps of the above method when executed by a processor.

[0039] The above-mentioned prediction method, device, computer equipment, computer-readable storage medium and computer program product of collection style information firstly outputs the initial speech data in each scenario through the scenario classification model, and secondly, determines the mapping relationship between the target speech data and the customer type according to the initial speech data in each scenario and the second basic feature data of the sample user, matches different target speech data for different customer types, and stores the mapping relationship in the prediction model. Finally, the first basic feature data of the predicted user is input into the prediction model to obtain the target collection style information corresponding to the predicted user output by the prediction model, and the first basic feature data of the predicted user is processed by the prediction model to output the collection style information of the customer type corresponding to the predicted user. Outputting different collection style information for different predicted users avoids the problem of single collection style information in traditional technology, improves the flexibility of the collection robot while also improving its adaptability to customers, and improves the resource utilization of the intelligent robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A diagram showing an application environment of a method for predicting collection style information in an embodiment;

[0042] Figure 2 A schematic diagram of a flow chart of a method for predicting collection style information in one embodiment;

[0043] Figure 3 is a flow chart of a method for generating a mapping relationship in an embodiment;

[0044] Figure 4 A flowchart of a method for determining initial speech data in one embodiment;

[0045] Figure 5 A schematic diagram of a process of obtaining a target collection manual interaction text for a target time period in one embodiment;

[0046] Figure 6 A flowchart of a method for processing a prediction model in one embodiment;

[0047] Figure 7 A schematic diagram of a process for determining a feature vector according to a customer group type in one embodiment;

[0048] Figure 8 is a structural block diagram of a device for predicting collection style information in one embodiment;

[0049] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] The method for predicting collection style information provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The server 104 obtains the first basic feature data of the predicted user; inputs the first basic feature data into the prediction model, and obtains the prediction result output by the prediction model, wherein the prediction result includes the target collection style information corresponding to the predicted user; the target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer group type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scene; the initial speech data in each scene is output based on the scene classification model, and the scene classification model is trained based on the second basic feature data. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head mounted device may be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0052] In an exemplary embodiment, Figure 2 As shown, a method for predicting collection style information is provided, and the method is applied to Figure 1The server in the example is used to illustrate, including the following steps S202 to S204. Among them:

[0053] Step S202, obtaining first basic characteristic data of the predicted user.

[0054] The first basic feature data is the basic feature data of the predicted user, such as name, gender, card number, address and the like.

[0055] Optionally, the server obtains the predicted user, and obtains corresponding first basic feature data, such as name, gender, card number, address, etc., based on the predicted user.

[0056] Step S204: input the first basic feature data into the prediction model to obtain the prediction result output by the prediction model.

[0057] Among them, the prediction results include the target collection style information corresponding to the predicted user; the target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is based on the output of the scenario classification model, and the scenario classification model is used to classify the collection manual interaction text.

[0058] Optionally, before inputting the first basic feature data into the prediction model, the server inputs the second basic feature data, user repayment data, and collection manual interaction text into the trained classification model, and obtains the initial speech data for each scenario output by the classification model.

[0059] Optionally, before inputting the first basic feature data into the prediction model, the server trains the initial prediction model through the second basic feature data of the sample users and the initial speech data in each scenario output by the scenario classification model, and determines the mapping relationship between the target speech data and the customer type through the second basic feature data of the sample users and the initial speech data in each scenario, and stores the mapping relationship in the prediction model to obtain a trained prediction model.

[0060] Optionally, the server inputs the first basic feature data into a trained prediction model, and obtains a prediction result output by the prediction model, where the prediction result includes target collection style information corresponding to the predicted user.

[0061] In the above-mentioned prediction method of collection style information, first, the initial speech data for each scenario is output by the scenario classification model. Secondly, the mapping relationship between the target speech data and the customer type is determined based on the initial speech data for each scenario and the second basic feature data of the sample user, and different target speech data are matched for different customer types, and the mapping relationship is stored in the prediction model. Finally, the first basic feature data of the predicted user is input into the prediction model to obtain the target collection style information corresponding to the predicted user output by the prediction model, and the first basic feature data of the predicted user is processed by the prediction model to output the collection style information of the customer type corresponding to the predicted user. Outputting different collection style information for different predicted users avoids the problem of single collection style information in traditional technologies, improves the flexibility of the collection robot, and also improves its adaptability to customers, thereby improving the resource utilization of the intelligent robot.

[0062] In the above embodiment, Figure 3 As shown, the method for generating the mapping relationship includes steps S302 to S308. Among them:

[0063] Step S302, for each scene's initial speech data, determine the second basic feature data associated with the scene.

[0064] The initial speech data includes the initial speech content and the initial speech order corresponding to the initial speech content.

[0065] Optionally, the server determines the sample users associated with each scenario for the initial speech data in each scenario, and determines the corresponding second basic feature data according to the sample users. For example, the second basic feature data of sample user A associated with scenario 1; the second basic feature data of sample user B associated with scenario 2, and so on. For example, in 100 scenarios, there are 100 associated sample users.

[0066] Step S304: Determine the customer group type according to the second basic characteristic data.

[0067] Optionally, the server screens and obtains common characteristic data according to the second basic characteristic data of each sample user, for example, the age of 100 sample users is between 20-30, the income is between xx-xx, etc., and determines that the customer type of the 100 sample users is one customer type.

[0068] Step S306, sort and superimpose the initial speech content and the initial speech order in each scene, and determine the target speech content and the target speech order corresponding to the target speech content.

[0069] Optionally, the server sorts and superimposes the initial speech content and initial speech order in each scenario. For example, the initial speech content and initial speech order of scenario 1 are abcde; the initial speech content and initial speech order of scenario 2 are abch; the initial speech content and initial speech order of scenario 3 are abegi... The initial speech content and initial speech order of 100 scenarios are sorted and superimposed, and the sorted and superimposed initial speech data are used as the target speech content and target speech order.

[0070] It should be noted that there is no limitation on the content and order of the initial dialogue in each scenario.

[0071] Step S308, associating the target speech content, target speech sequence and customer group type, and determining the mapping relationship between the target speech data and the customer group type, wherein the target speech data includes the target speech content and the target speech sequence.

[0072] Optionally, the server processes the initial speech content and initial speech sequence data of several scenes to obtain target speech content and target speech sequence; the server processes the second basic feature data of several sample users to obtain the customer group type. The server associates the target speech content, target speech sequence and customer group type to determine the mapping relationship between the target speech data and the customer group type.

[0073] In this embodiment, the target speech content and target speech sequence of the corresponding customer type are formed by associating and matching the user feature data with the initial speech content and initial speech sequence of each scene output by the corresponding classification model, and stored in the prediction model as the basis for the subsequent prediction model to process the first basic feature data of the predicted user.

[0074] In the previous embodiment, Figure 4 As shown, the method for determining the initial speech data includes steps S402 to S410. Among them:

[0075] Step S402, obtaining the target collection manual interaction text in the target time period.

[0076] The target collection manual interaction text indicates the text containing the user repayment data within the preset time after the collection manual interaction. For example, if the user repayment data is newly added within 2 hours after the collection manual interaction, the collection manual interaction text is used as the target text, and the target collection manual interaction text is used as the interactive text with excellent performance. The collection style information of the target collection manual interaction text is mined to increase the anthropomorphism of the collection robot.

[0077] Optionally, the server obtains target collection manual interaction texts in a target time period through a classification model, such as several target collection manual interaction texts in December 2024.

[0078] Step S404, semantic understanding is performed on the target collection manual interaction text to obtain the semantics corresponding to each text in the target collection manual interaction text, and the number of semantics is counted.

[0079] Optionally, the server performs semantic understanding on each target collection manual interaction text through a classification model to obtain the semantics corresponding to each text in the target collection manual interaction text, and counts the number of semantics. For example, there are 200 groups of dialogue texts in one of the target collection manual interaction texts. The server performs semantic understanding on each text to obtain the corresponding semantics. Assume that one of the same semantics appears in 3 groups of dialogue texts and another of the same semantics appears in 10 groups of dialogue texts. The number of corresponding semantics in the target collection manual interaction text may be any number less than the number of texts.

[0080] Furthermore, the server performs semantic understanding on each text, and outputs three semantics for the same text WB. For example, the probability that the text WB corresponds to semantic 1 is 80%, the probability that it corresponds to semantic 2 is 70%, and the probability that it corresponds to semantic 3 is 60%. The semantic with the highest probability is selected as the semantic of the text.

[0081] Step S406, based on the number of semantics, word embedding encoding is performed on the semantics corresponding to each text to determine the vector representation of the target collection manual interaction text.

[0082] Optionally, the server encodes each semantics through a classification model to obtain a vector of fixed dimension, which is used to represent the semantics. The number of rows corresponds to the number of words in the input text, that is, the number of semantics, and the number of columns is the dimension of the encoding. The server performs word embedding encoding on the semantics corresponding to each text to determine the vector representation of the target collection manual interaction text.

[0083] Step S408: perform intent labeling on the vector representation according to the sample labeling.

[0084] Among them, the sample annotation is a reference annotation, which can be a manually completed annotation.

[0085] Optionally, the server uses a classification model to label a large number of vector representations according to sample labels, such that the intent of vector representation 1 is labeled as A, the intent of vector representation 2 is labeled as A, the intent of vector representation 3 is labeled as B, and so on.

[0086] Step S410, classify the vector representations after the intent annotation is completed to obtain the initial speech data in each scenario.

[0087] Optionally, the server classifies the vector representations after the intent annotations are completed according to the scenarios through a classification model to obtain initial speech data for each scenario.

[0088] In this embodiment, the manual collection interaction texts with excellent performance are processed through the classification model and classified according to the scenarios, so that the initial speech data in each scenario can be obtained. The initial speech data in each scenario is used as a kind of training data to train the prediction model. When using the collection robot for collection, the style of manual collection is absorbed to increase the humanity of the collection robot.

[0089] Continuing with the above embodiment, Figure 5 As shown, obtaining the target collection manual interaction text in the target time period includes steps S502 to S506. Among them:

[0090] Step S502, obtaining historical collection manual interaction texts and historical user repayment data for the target time period.

[0091] Optionally, the server obtains historical collection manual interaction texts and historical user repayment data for the target time period. For example, the server obtains historical collection manual interaction texts and historical user repayment data for December 2024. In this case, the historical collection manual interaction texts and historical user repayment data are two independent parts of data.

[0092] Step S504: for each target user corresponding to the historical user repayment data, determine the historical collection manual interaction text corresponding to the target user.

[0093] Optionally, for each target user corresponding to the historical user repayment data, the historical collection manual interaction text corresponding to the target user is matched from the historical collection manual interaction text.

[0094] Step S506: Use the historical collection manual interaction text corresponding to the target user as the target collection manual interaction text for the target time period.

[0095] Optionally, the server uses the historical collection manual interaction text corresponding to the target user as the target collection manual interaction text for the target time period.

[0096] In this embodiment, by screening the historical user repayment data to obtain the corresponding historical collection manual interaction texts, and using them as the target collection manual interaction texts for the target time period, it is possible to screen out the collection manual interaction texts that have collection effects, and use them as training data for the classification model and the prediction model to enrich the customer segmentation rules of the intelligent collection robot and enhance the collection robot's adaptability to changes in the market collection environment.

[0097] In an exemplary embodiment, Figure 6As shown, the processing process of the prediction model includes steps S602 to S608. Among them:

[0098] Step S602: Determine the customer group type based on the first basic characteristic data of the predicted user.

[0099] Optionally, the server analyzes the first basic characteristic data of the predicted user through a prediction model to determine the customer group type, such as characteristic fields included in the customer group type.

[0100] Step S604: determining a feature vector according to the customer group type.

[0101] Optionally, the server determines the corresponding different feature fields according to the customer type through the prediction model. For example, customer type 1 corresponds to feature field 1, feature field 2, and feature field 3; customer type 2 corresponds to feature field 1, feature field 2, and feature field 4, etc. The server performs encoding, normalization, and vector processing on each feature field to obtain a feature vector.

[0102] Step S606, processing the feature vector through the mapping relationship between the target speech data and the customer group type to obtain the target speech data corresponding to the predicted user.

[0103] Optionally, the prediction model includes a hidden layer and an output layer; the server calculates the feature vector through the hidden layer of the prediction model based on the mapping relationship between the target speech data and the customer type, and passes the calculated feature vector to the output layer. The Softmax activation function in the output layer converts the output value into a probability. For example, the probability of the output target speech data 1 is 75%, the probability of the target speech data 2 is 95%, and the probability of the target speech data 3 is 60%. The target speech data with the highest probability value is obtained as the target speech data corresponding to the predicted user.

[0104] Step S608, determining the target collection style information according to the target speech data.

[0105] Optionally, the server determines the target collection style information based on the target speech data through the prediction model. For example, the server determines the target collection style information that will produce a good collection effect on the predicted user based on the target speech content and target speech sequence in the target speech data. For example, for a certain customer type, a tough and decisive target collection style information will have a good collection effect; the target speech content and target speech sequence corresponding to the tough and decisive target collection style information are used to collect from the user.

[0106] In this embodiment, the target collection style information and target speech data of the predicted user are determined by the prediction model. Compared with the collection method with fixed scenarios and fixed communication speech, it is more targeted, flexible and more suitable for the collection user.

[0107] In an exemplary embodiment, Figure 7 As shown, determining the feature vector according to the customer group type includes steps S702 to S706. Among them:

[0108] Step S702: Determine the feature field according to the customer group type.

[0109] Optionally, the server determines corresponding different feature fields according to the customer group type. For example, customer group type 1 corresponds to feature field 1, feature field 2, and feature field 3; customer group type 2 corresponds to feature field 1, feature field 2, and feature field 4, and so on.

[0110] Step S704: Acquire a first feature field that is a null value from the feature fields, and predict the first feature field to obtain a fitting value of the first feature field.

[0111] Among them, the characteristic field with an empty value is used as the first characteristic field, and the characteristic field with a non-empty value is used as the second characteristic field.

[0112] Optionally, the server obtains a first feature field that is an empty value from the feature field, performs fitting and completion using different calculation methods (statistics, algorithm prediction, etc.), and determines to obtain a fitting value of the first feature field.

[0113] Step S706: Encode, normalize and vectorize the fitted value of the first feature field and the actual value of the second feature field to obtain a feature vector.

[0114] The second characteristic field indicates a characteristic field that is not a null value.

[0115] Optionally, the server performs encoding processing, normalization processing, and vector processing on the fitted value of the first feature field and the actual value of the second feature field to obtain a feature vector.

[0116] Optionally, encoding, normalizing and vectorizing the fitted value of the first feature field and the actual value of the second feature field to obtain a feature vector includes: extracting the first feature field and the second feature field from the first basic feature data, encoding the first feature field, the fitted value of the first feature field, the second feature field and the actual value of the second feature field; converting the encoded fitted value of the first feature field and the encoded actual value of the second feature field into dimensionless data, and normalizing the dimensionless data; determining the feature vector corresponding to the first basic feature data based on the normalized dimensionless data. The server mathematically encodes different feature fields of the existing data, including the first feature field and the second feature field, such as occupation, risk level, whether overdue and other non-digital data, according to their values, and converts the original data into dimensionless data through data standardization, so that the value range of each indicator is at the same quantitative level, eliminating the differences (dimensions) between the features, and treating all dimensionally normalized feature data as vectors of multiple dimensions (n ​​dimensions), and combining them into an m*n matrix according to a fixed number of rows (m rows).

[0117] In this embodiment, by encoding, normalizing and vectorizing the feature field to obtain a feature vector or matrix, the computational efficiency of the prediction model can be improved and the training time of the prediction model can be reduced.

[0118] In an exemplary embodiment, in the first stage: the server inputs the second basic feature data, user repayment data, and collection manual interaction text into the trained classification model, and obtains the initial speech data for each scenario output by the classification model. The server trains the initial prediction model through the second basic feature data of the sample users and the initial speech data for each scenario output by the scenario classification model, determines the mapping relationship between the target speech data and the customer type through the second basic feature data of the sample users and the initial speech data for each scenario, and stores the mapping relationship in the prediction model to obtain a trained prediction model.

[0119] The second stage: The server obtains the predicted user, and obtains the corresponding first basic feature data based on the predicted user, such as name, gender, card number, address, etc. The server inputs the first basic feature data into the trained prediction model, and obtains the prediction result output by the prediction model. The prediction result includes the target collection style information corresponding to the predicted user. The target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is based on the output of the scene classification model, and the scene classification model is used to classify the collection manual interaction text.

[0120] Among them, the mapping relationship is generated in the following way: the server determines the sample users associated with each scenario for the initial speech data in each scenario, and determines the corresponding second basic feature data based on the sample users. For example, the second basic feature data of sample user A associated with scenario 1; the second basic feature data of sample user B associated with scenario 2, and so on. For example, in 100 scenarios, there are 100 associated sample users. The server screens and obtains common feature data based on the second basic feature data of each sample user. For example, the age of the 100 sample users is between 20-30, and the income is between xx-xx, etc., and determines that the customer type of the 100 sample users is one customer type. The server sorts and superimposes the initial speech content and initial speech order in each scene. For example, the initial speech content and initial speech order of scene 1 are abcde; the initial speech content and initial speech order of scene 2 are abch; the initial speech content and initial speech order of scene 3 are abegi... The initial speech content and initial speech order of 100 scenes are sorted and superimposed, and the sorted and superimposed initial speech data are used as the target speech content and target speech order. The server processes the initial speech content and initial speech order data of several scenes to obtain the target speech content and target speech order; the server processes the second basic feature data of several sample users to obtain the customer group type. The server associates the target speech content, target speech order and customer group type to determine the mapping relationship between the target speech data and the customer group type.

[0121] How to determine the initial speech data:

[0122] The server obtains the historical collection manual interaction texts and historical user repayment data for the target time period. For example, the historical collection manual interaction texts and historical user repayment data for December 2024 are obtained. At this time, the historical collection manual interaction texts and historical user repayment data are two independent parts of data. For each target user corresponding to the historical user repayment data, the historical collection manual interaction texts corresponding to the target user are matched from the historical collection manual interaction texts. The server uses the historical collection manual interaction texts corresponding to the target user as the target collection manual interaction texts for the target time period. The server performs semantic understanding on each target collection manual interaction text through a classification model, obtains the semantics corresponding to each text in the target collection manual interaction text, and counts the number of semantics. For example, there are 200 groups of dialogue texts in one of the target collection manual interaction texts. The server performs semantic understanding on each text and obtains the corresponding semantics. Assume that one of the same semantics appears in 3 groups of dialogue texts, and another of the same semantics appears in 10 groups of dialogue texts. The number of corresponding semantics in the target collection manual interaction text may be any number less than the number of texts.

[0123] Furthermore, the server performs semantic understanding on each text, and outputs three semantics for the same text WB. For example, the probability of the text WB corresponding to semantic 1 is 80%, the probability of the corresponding semantic 2 is 70%, and the probability of the corresponding semantic 3 is 60%. The semantic with the highest probability is selected as the semantic of the text. The server encodes each semantic through the classification model to obtain a vector of fixed dimension, which is used to represent this semantic. The number of rows corresponds to the number of words in the input text, that is, the number of semantics, and the number of columns is the dimension of the encoding. The server performs word embedding encoding on the semantics corresponding to each text to determine the vector representation of the target collection manual interaction text. The server uses the classification model to annotate a large number of vector representations according to sample annotations. The intent of vector representation 1 is annotated as A; the intent of vector representation 2 is annotated as A; the intent of vector representation 3 is annotated as B, and so on. The server classifies the vector representations with completed intent annotation according to the scene through the classification model to obtain the initial speech data in each scene.

[0124] The processing process of the prediction model includes: the server analyzes the first basic feature data of the predicted user through the prediction model to determine the customer group type, such as the feature fields included in the customer group type. The server determines the corresponding different feature fields according to the customer group type. For example, customer group type 1 corresponds to feature field 1, feature field 2 and feature field 3; customer group type 2 corresponds to feature field 1, feature field 2 and feature field 4, etc. The server obtains the first feature field with a null value from the feature field, and uses different calculation methods (statistics, algorithm prediction, etc.) to perform fitting and completion to determine the fitting value of the first feature field. The first feature field and the second feature field are extracted from the first basic feature data, and the first feature field, the fitting value of the first feature field, the second feature field and the actual value of the second feature field are encoded; the encoded fitting value of the first feature field and the encoded actual value of the second feature field are converted into dimensionless data, and the dimensionless data is normalized; based on the normalized dimensionless data, the feature vector corresponding to the first basic feature data is determined. The prediction model includes a hidden layer and an output layer; the server calculates the feature vector based on the mapping relationship between the target speech data and the customer type through the hidden layer of the prediction model, and passes the calculated feature vector to the output layer. The Softmax activation function in the output layer converts the output value into a probability. For example, the probability of the output target speech data 1 is 75%, the probability of the target speech data 2 is 95%, and the probability of the target speech data 3 is 60%. The target speech data with the highest probability value is obtained as the target speech data corresponding to the predicted user. The server determines the target collection style information based on the target speech data through the prediction model. For example, the server determines the target collection style information that has a good collection effect on the predicted user based on the target speech content and target speech sequence in the target speech data. For example, for a certain customer type, a tough and decisive target collection style information will have a good collection effect; the target speech content and target speech sequence corresponding to the tough and decisive target collection style information are used to collect debts from users.

[0125] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0126] Based on the same inventive concept, the embodiment of the present application also provides a device for predicting collection style information for implementing the above-mentioned method for predicting collection style information. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiments of one or more devices for predicting collection style information provided below can refer to the limitations of the method for predicting collection style information above, and will not be repeated here.

[0127] In an exemplary embodiment, Figure 8 As shown, a device for predicting collection style information is provided, including: an acquisition module 801 and an input and output module 802, wherein:

[0128] An acquisition module 801 is used to acquire first basic characteristic data of a predicted user;

[0129] The input-output module 802 is used to input the first basic feature data into the prediction model to obtain the prediction result output by the prediction model, wherein the prediction result includes the target collection style information corresponding to the predicted user; the target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is output based on the scenario classification model, and the scenario classification model is used to classify the collection manual interaction text.

[0130] In an exemplary embodiment, a device for predicting collection style information further includes:

[0131] A mapping relationship generation module is used to determine the second basic characteristic data associated with the scene for the initial speech data in each scene; wherein the initial speech data includes the initial speech content and the initial speech order corresponding to the initial speech content; determine the customer type based on the second basic characteristic data; sort and superimpose the initial speech content and the initial speech order in each scene to determine each target speech content and the target speech order corresponding to the target speech content; associate the target speech content, the target speech order and the customer type to determine the mapping relationship between the target speech data and the customer type, wherein the target speech data includes the target speech content and the target speech order.

[0132] In an exemplary embodiment, a mapping relationship generation module includes an initial speech data determination unit, which is used to obtain a target collection manual interaction text in a target time period; wherein the target collection manual interaction text indicates a text containing user repayment data within a preset time after the collection manual interaction; semantic understanding is performed on the target collection manual interaction text to obtain the semantics corresponding to each text in the target collection manual interaction text, and the number of semantics is counted; based on the number of semantics, word embedding encoding is performed on the semantics corresponding to each text to determine a vector representation of the target collection manual interaction text; intent annotation is performed on the vector representation based on the sample annotation; and the vector representation with completed intent annotation is classified to obtain initial speech data for each scenario.

[0133] In an exemplary embodiment, the initial speech data determination unit includes a target collection text acquisition subunit, which is used to obtain historical collection manual interaction texts and historical user repayment data for a target time period; for each target user corresponding to the historical user repayment data, determine the historical collection manual interaction text corresponding to the target user; and use the historical collection manual interaction text corresponding to the target user as the target collection manual interaction text for the target time period.

[0134] In an exemplary embodiment, a device for predicting collection style information further includes:

[0135] The prediction model processing module is used to determine the customer type based on the first basic characteristic data of the predicted user; determine the feature vector based on the customer type; process the feature vector through the mapping relationship between the target speech data and the customer type to obtain the target speech data corresponding to the predicted user; and determine the target collection style information based on the target speech data.

[0136] In an exemplary embodiment, the prediction model processing module includes a feature vector determination unit, which is used to determine the feature field according to the customer type; obtain a first feature field with an empty value from the feature field, and predict the first feature field to obtain a fitting value of the first feature field; encode, normalize and vectorize the fitting value of the first feature field and the actual value of the second feature field to obtain a feature vector; wherein the second feature field indicates a feature field that is not an empty value.

[0137] Each module in the above-mentioned device for predicting collection style information can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0138] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig. 9 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store basic feature data and collection style information. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting collection style information is implemented.

[0139] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0140] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0141] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0142] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0143] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0144] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0145] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0146] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting collection style information, characterized in that: The method comprises: Acquire first basic characteristic data of the predicted user; The first basic feature data is input into the prediction model to obtain the prediction result output by the prediction model, wherein the prediction result includes the target collection style information corresponding to the predicted user; the target collection style information is determined based on the target speech data; the prediction model stores the mapping relationship between the target speech data and the customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is output based on the scenario classification model, and the scenario classification model is used to classify the collection manual interaction text.

2. The method according to claim 1, characterized in that The generation method of the mapping relationship includes: For each of the initial speech data in the scenario, determining the second basic feature data associated with the scenario; wherein the initial speech data includes initial speech content and an initial speech sequence corresponding to the initial speech content; Determining the customer group type according to the second basic characteristic data; The initial speech content and the initial speech order in each scenario are sorted and superimposed to determine the target speech content and the target speech order corresponding to the target speech content; The target speech content, the target speech sequence and the customer group type are associated to determine a mapping relationship between the target speech data and the customer group type, wherein the target speech data includes the target speech content and the target speech sequence.

3. The method according to claim 2, characterized in that The method for determining the initial speech data includes: Obtaining a target collection manual interaction text for a target time period; wherein the target collection manual interaction text indicates a text containing user repayment data within a preset time after the collection manual interaction; Performing semantic understanding on the target collection manual interaction text, obtaining semantics corresponding to each text in the target collection manual interaction text, and counting the number of the semantics; According to the number of the semantics, word embedding encoding is performed on the semantics corresponding to each text to determine the vector representation of the target collection manual interaction text; Performing intent labeling on the vector representation according to the sample labeling; The vector representations after the intent annotation are classified to obtain the initial speech data in each scenario.

4. The method according to claim 3, characterized in that The target collection manual interaction text for the target time period is obtained, including: Obtain historical collection manual interaction texts and historical user repayment data for the target time period; For each target user corresponding to the historical user repayment data, determining the historical collection manual interaction text corresponding to the target user; The historical collection manual interaction text corresponding to the target user is used as the target collection manual interaction text in the target time period.

5. The method according to claim 1, characterized in that The method further comprises: Determining a customer group type according to the first basic characteristic data of the predicted user; Determining a feature vector according to the customer group type; The feature vector is processed through the mapping relationship between the target speech data and the customer group type to obtain the target speech data corresponding to the predicted user; Target collection style information is determined based on the target speech data.

6. The method according to claim 5, characterized in that The step of determining a feature vector according to the customer group type includes: Determine the feature field according to the customer group type; Acquire a first feature field that is a null value from the feature fields, and predict the first feature field to obtain a fitting value of the first feature field; The fitted value of the first feature field and the actual value of the second feature field are encoded, normalized, and vectorized to obtain the feature vector; wherein the second feature field indicates a feature field that is not a null value.

7. A device for predicting collection style information, characterized in that: The device comprises: An acquisition module, used to acquire first basic characteristic data of the predicted user; An input-output module is used to input the first basic feature data into a prediction model to obtain a prediction result output by the prediction model, wherein the prediction result includes target collection style information corresponding to the predicted user; the target collection style information is determined based on target speech data; the prediction model stores a mapping relationship between target speech data and customer type; the mapping relationship is determined based on the second basic feature data of the sample user and the initial speech data in each scenario; the initial speech data in each scenario is output based on a scenario classification model, and the scenario classification model is used to classify collection manual interaction texts.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images