CN111489095A

CN111489095A - Risk user management method and device, computer equipment and storage medium

Info

Publication number: CN111489095A
Application number: CN202010297207.5A
Authority: CN
Inventors: 陈强; 吴俊江; 雷植程
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-04-15
Filing date: 2020-04-15
Publication date: 2020-08-04
Anticipated expiration: 2040-04-15
Also published as: CN111489095B

Abstract

The invention discloses a risk user management method, a risk user management device, computer equipment and a storage medium, wherein user associated data can be acquired based on historical customer service records of a target user in a plurality of customer service channels of a target product; acquiring customer service emergency degree evaluation data on a plurality of customer service emergency degree evaluation dimensions from the user associated data; performing data processing on the evaluation data to obtain multi-dimensional customer service emergency degree evaluation characteristics; the risk attribute information of the target user is predicted based on the evaluation features, the customer service response scheme of the target user is determined based on the risk attribute information and the customer service channels, the data source for extracting the evaluation features comprises a plurality of customer service channels, the information in the features is rich, the service condition of the user to the customer service can be reflected better, the multi-dimension of the evaluation features is beneficial to analyzing the customer service emergency degree of the user from different angles, the risk prediction accuracy is improved, the customer service response scheme is set based on the risk attribute prediction result, customer service resource optimization is achieved, and the user experience is improved.

Description

Risk user management method and device, computer equipment and storage medium

Technical Field

The invention relates to the technical field of computers, in particular to a risk user management method, a risk user management device, computer equipment and a storage medium.

Background

In the field of customer service, a call center needs to process interactive behaviors such as consultation, suggestion, complaint and the like of massive users every day; however, due to the limitation of manual service capacity, it is difficult to manually service each user requesting customer service.

In the related technology, users can be accessed according to the order of customer service telephone dialing, but for customers using products, different customers generally have different degrees of urgency, and users with high degrees of urgency are generally regarded as risk users, once the users initiate customer service requests and cannot obtain timely service, a wind control event may be caused, so how to manage the risk users is a problem which needs to be solved currently.

Disclosure of Invention

The embodiment of the invention provides a risk user management method, a risk user management device, computer equipment and a storage medium, which can be used for analyzing risk attributes of a user from multiple dimensions by combining user associated data of multiple customer service channels, are beneficial to improving analysis accuracy and optimizing customer service resources.

The embodiment of the invention provides a risk user management method, which comprises the following steps:

acquiring user association data generated based on the target user's use of a target product based on historical customer service records of the target user in at least two customer service channels provided by the target product;

acquiring customer service emergency degree evaluation data of the target user in a plurality of customer service emergency degree evaluation dimensions from the user associated data;

performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of the plurality of customer service emergency degree evaluation dimensions;

predicting risk attribute information of the target user based on the customer service emergency degree evaluation features, wherein the risk attribute information is used for representing the customer service emergency degree of the target user;

determining a customer service response scheme for the target user based on the risk attribute information of the target user and the customer service channel of the target product.

The embodiment of the invention also provides a risk user management device, which comprises:

the system comprises a user associated data acquisition unit, a service management unit and a service management unit, wherein the user associated data acquisition unit is used for acquiring user associated data generated based on the use of a target product by a target user based on historical customer service records of the target user in at least two customer service channels provided by the target product;

an evaluation data obtaining unit, configured to obtain, from the user-related data, customer service emergency degree evaluation data of the target user in a plurality of customer service emergency degree evaluation dimensions;

the characteristic acquisition unit is used for carrying out data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of the customer service emergency degree evaluation dimensions;

a prediction unit, configured to predict risk attribute information of the target user based on the customer service emergency degree evaluation feature, where the risk attribute information is used to indicate a customer service emergency degree for the target user;

and the scheme management unit is used for determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product.

An embodiment of the present invention further provides a storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the risk user management method as described above.

An embodiment of the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned risk user management method when executing the computer program.

The embodiment of the invention provides a risk user management method, a risk user management device, computer equipment and a storage medium, which can acquire user association data generated based on the use of a target product by a target user based on historical customer service records of the target user in at least two customer service channels provided by the target product; acquiring customer service emergency degree evaluation data of a target user on a plurality of customer service emergency degree evaluation dimensions from the user associated data; performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions; therefore, the data source of the customer service emergency degree evaluation feature comprises a plurality of customer service channels of the target product, the information in the evaluation feature of each dimension is richer, the requirement of a user on customer service and the satisfaction degree of the user on the target product can be reflected better, the multi-dimension of the customer service emergency degree evaluation feature can reflect the customer service emergency degree of the user from a plurality of different angles, the prediction accuracy of the risk attribute information is favorably improved, after the feature is extracted, the risk attribute information of the target user can be predicted based on the customer service emergency degree evaluation feature, and then the customer service response scheme of the target user is determined based on the risk attribute information of the target user and the customer service channels of the target product; therefore, the embodiment can realize setting of corresponding customer service response schemes for target users with different customer service emergency degrees, realize customer service resource optimization, and is beneficial to improving user satisfaction and reducing user complaints.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a scene schematic diagram of a risk user management method according to an embodiment of the present invention;

fig. 2a is a flowchart of a method for risk user management according to an embodiment of the present invention;

fig. 2b is an architecture diagram of a risk user management method according to an embodiment of the present invention;

FIG. 3a is a flow chart of a method for training a risk identification model according to an embodiment of the present invention;

FIG. 3b is a schematic structural diagram of a risk identification model provided by an embodiment of the present invention;

FIG. 3c is a schematic structural diagram of an identification submodel in the risk identification model provided by the embodiment of the present invention;

fig. 4 is a schematic structural diagram of a risk user management device according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The embodiment of the invention provides a risk user management method, a risk user management device, computer equipment and a storage medium. In particular, the present embodiment provides a risk user management method applicable to a risk user management apparatus, which may be integrated in a computer device.

The computer device may be a terminal or other device, such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, or other device.

The computer device may also be a device such as a server, and the server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, middleware service, a domain name service, a security service, a CDN, and a big data and artificial intelligence platform, but is not limited thereto.

The risk user management method of this embodiment may be implemented by a terminal or a server, or may be implemented by both the terminal and the server.

The following describes a risk user management method by taking an example in which the terminal and the server implement the risk user management method together.

Referring to fig. 1, a risk user management system provided in an embodiment of the present invention includes a terminal 10, a server 20, and the like; the terminal 10 and the server 20 are connected via a network, for example, a wired or wireless network connection, wherein the risk user management device on the terminal side may be integrated in the terminal in the form of a client.

The server 20 may be configured to obtain user association data generated based on the target user's usage of a target product based on historical customer service records of the target user in at least two customer service channels provided by the target product; acquiring customer service emergency degree evaluation data of the target user on a plurality of customer service emergency degree evaluation dimensions from user associated data; performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of the plurality of customer service emergency degree evaluation dimensions; predicting risk attribute information of the target user based on the customer service emergency degree evaluation features, wherein the risk attribute information is used for representing the customer service emergency degree of the target user; determining a customer service response scheme for the target user based on the risk attribute information of the target user and the customer service channel of the target product.

The terminal 10 may be configured to send a customer service request to a server, and the server 20 may be further configured to receive a customer service request of a user for the target product, and compare a customer service channel of the customer service request initiated by the user with a service capability of a target customer channel of the user; and if the service capability of the target customer channel is relatively strong, switching to the target customer service channel to provide customer service corresponding to the customer service request for the user.

The following are detailed below. It should be noted that the following description of the embodiments is not intended to limit the preferred order of the embodiments.

The embodiments of the present invention will be described from the perspective of a risk user management device, which may be specifically integrated in a server. An embodiment of the present invention provides a method for managing a risk user, where the method may be executed by a processor of a server, and as shown in fig. 2a, a flow of the method for managing a risk user may be as follows:

201. acquiring user association data generated based on the target user's use of a target product based on historical customer service records of the target user in at least two customer service channels provided by the target product;

the product in this embodiment refers to anything that is provided to the market as a commodity, used and consumed by people, and can satisfy certain needs of people, including tangible articles, intangible services, organizations, ideas or combinations thereof. In one example, the product in this embodiment may be an internet product, i.e., a commodity produced for business in the internet field, which is an intangible carrier to meet the needs and desires of internet users. For example, the target product may be a web product including, but not limited to, a web page version search engine, a webgame product, and the like, and an application client including, but not limited to, a video client, an instant messenger client, a game client, and the like.

The customer service channel of the embodiment may be a channel for providing customer service to users of products, where the customer service includes, but is not limited to, consultation, advice, and complaint.

In this embodiment, different customer service channels are different from the user communication modes, and the customer service channels include but are not limited to: a telephone channel, an online reply channel, a forum channel, a mail channel, and an instant messaging service channel, etc., wherein the instant messaging service channel can provide customer service to users through instant messaging clients. Alternatively, in one example, different products may use the same customer service channel.

For example, the telephone channel may include an artificial channel, an ivr (interactive Voice response) channel, an interactive Voice response channel, and an ai (artificial intelligence) artificial intelligence response channel.

The user identification information may be different in different customer service channels, for example, in a telephone channel, the user identification information may include a telephone number, in an online reply channel, the user identification information may include account information registered by the user on a product, and in a mail channel, the user identification information may include mailbox information of the user.

In this embodiment, the identification information of the same user in different customer service channels can be determined first, then the historical customer service records of the user in each customer service channel are respectively obtained based on the identification information, and the historical customer service records of the different customer service channels are integrated into the historical customer service record of the target user on the target product; and then acquiring user associated data based on the historical customer service records.

Optionally, the step "obtaining user association data generated based on the target user's usage of the target product based on historical customer service records of the target user in at least two customer service channels provided by the target product" may include:

acquiring identity identification information of a target user in at least two customer service channels of a target product;

acquiring historical customer service records of target products of target users in each customer service channel based on the identity identification information;

and acquiring user association data generated based on the target product usage of the target user based on the historical customer service record.

In this embodiment, the required historical customer service record may be obtained from the customer service log corresponding to the customer service channel through the identification information of the target user and the name of the target product.

The user association data in this embodiment may include any data generated by the target user in the process of using the target product. For example, may include data associated with the target user itself, and/or data associated with a target product used by the target user.

In one embodiment, the user-associated data may be obtained from historical customer service records, and in another example, data associated with target products used by the target users may be obtained from historical customer service records, and data of the target users themselves, such as age, gender, and the like, may be obtained from a user database of the target products.

Optionally, the user-associated data is data that may be embodied in the risk management importance of the target user in the target product, where the risk management importance of the target user is related to multiple aspects, such as being related to the target product itself (for example, if the product relates to finance, or the user size is large, or the product income is large, the more important the product is, the higher the risk management importance of each user in the product is), or being related to the target user itself (for example, if the target user has performed a large amount of recharge in the product, or purchased VI P service of the product, etc., the more important the target user is to the target product, the higher the risk management importance is), or being related to customer service initiated by the target user for the target product, it can be understood that, if the more satisfied the product is by the target user, the less the number of customer service requests may be, the higher the satisfaction degree and tolerance of the target user for the target product are high, if the target user is not satisfied with the product, the number of customer service requests may be large, the satisfaction and tolerance of the target user to the target product are low, and in the latter case, the importance of risk management of the target user is high.

Further, the user association data includes, but is not limited to: data which can reflect the importance degree of the target product and/or data which can reflect the importance degree of the target product service used by the target user and/or data which can reflect the frequency degree of customer service requests of the target user for the target product and/or the use probability distribution of the target user for different customer service channels and/or the content of the customer service requests made by the target user using the customer service channels and/or data which can reflect the satisfaction degree of the target user using the customer service channels and/or data which can reflect the customer service use condition of the target user for the target product; and data reflecting the importance of the target user to the target product, and the like.

202. Acquiring customer service emergency degree evaluation data of a target user on a plurality of customer service emergency degree evaluation dimensions from the user associated data;

in this embodiment, the number of customer service urgency evaluation dimensions is not less than two, and optionally, the customer service urgency evaluation dimensions include but are not limited to: user attribute dimension, user behavior dimension and product service dimension. The data in the user attribute dimension is related to the attribute of a user (such as a target user), the data in the user behavior dimension is related to the behavior of the user in the customer service process, and the data in the product service dimension is related to the service access process of the user in the product.

Optionally, the step of obtaining, from the user-related data, customer service emergency degree evaluation data of the target user in a plurality of customer service emergency degree evaluation dimensions may include:

acquiring user attribute data of a target user in a target product from the user associated data, and taking the user attribute data as customer service emergency degree evaluation data on a user attribute dimension;

acquiring user behavior data generated by a target user by using a customer service channel from the user associated data, and taking the user behavior data as customer service emergency degree evaluation data on user behavior dimension;

and acquiring product service access associated data of the target product from the user associated data as customer service emergency degree evaluation data on product service dimensionality.

Optionally, the user attribute data of this embodiment may be user portrait information, where the user portrait information includes but is not limited to: gender, age, product use duration, VI P information, high consumption information, total consumption information and the like.

Optionally, in this embodiment, if the historical customer service record has the USER attribute data, the USER attribute data may be directly obtained from the historical customer service record, and if the historical customer service record does not have the USER attribute data, the identity information of the target USER, for example, a USER-ID (USER identification ID) may be obtained from the historical customer service record, and the USER attribute data is obtained from a database (for example, a USER wind control database) of the target product based on the USER-ID (USER identification ID).

Optionally, in this embodiment, the user behavior data generated by the target user using the customer service channel may be understood as user behavior data generated from the time when the user initiates the customer service request to the time when the customer service is finished.

Examples of user behavior data and product business access association data are shown in the table above.

For example, user behavior data includes, but is not limited to: customer service channel, incoming call time (for telephone channel), number of manual requests, number of service times, successful short message delivery, successful delivery of instant messaging client message, switching from IVR to AI (Artificial Intelligence), switching from IVR to AI and back to IVR, switching from IVR to asynchronous processing channel (for example, processing customer service request through instant messaging client), switching from IVR to asynchronous processing channel and back to IVR, and so on.

Wherein, the AI channel may be an intelligent robot service channel provided by a product. The AI technique is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. In this example, in the AI channel, a customer service simulating a human can be provided to the user.

Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making. The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like. In this embodiment, in the AI channel, for a dialog (in a voice or text manner) of a user, a corresponding answer may be given based on a natural language processing technique, so as to realize simulation of the AI on a customer service person.

Optionally, the product service access association data in this embodiment includes data related to the access of the target user to the product service, for example, the product business access related data includes attribute data of the product itself such as product name, product business range, product importance, whether the product relates to financial service, and the like, wherein the product business access association data may further include product-related data generated based on the business access of the product by the user, such as whether the target user is stolen in the assets (including but not limited to virtual assets, wherein account numbers and the like are also regarded as virtual assets) of the product, whether payment abnormity occurs, whether the target user purchases services in the product, whether the target user is a VIP customer of the target product, the top-up balance of the target user in the target product, the service life of the target user on the target product and the like.

203. Performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions;

in this embodiment, the data processing of the customer service emergency degree evaluation data includes, but is not limited to: and performing feature engineering processing on the customer service emergency degree evaluation data, and converting the customer service emergency degree evaluation data into customer service emergency degree evaluation features represented by vectors.

For example, different data in one customer service urgency evaluation dimension can be regarded as data of different sub-dimensions. For example, for the user attribute dimension, the user age may be considered as data in the user age dimension, the user gender may be considered as data in the user gender dimension, and so on.

In this embodiment, data cleaning may be performed on the data first, for example, deleting duplicate data and abnormal data, filling default values in the data, performing normalization operation on the data, and the like.

Optionally, the step of performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation features of a plurality of customer service emergency degree evaluation dimensions may include:

converting the data in the customer service emergency degree evaluation data into corresponding vectors based on a vector conversion mode corresponding to the data type;

and forming the customer service emergency degree evaluation characteristics on the customer service emergency degree evaluation dimension by using the vectors corresponding to the customer service emergency degree evaluation data on the same customer service emergency degree evaluation dimension to obtain the customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions.

In this embodiment, the data types of the customer service urgency evaluation data include, but are not limited to, a continuous type and a discrete type, for the continuous type, min-max normalized to an interval of 0-1 may be used, and for the discrete type, one-hot encoding or random vector may be used for representation.

In this embodiment, the feature obtained after processing the user attribute data is the user attribute feature, the feature obtained after processing the user behavior data is the user behavior feature, and the feature obtained after processing the product service access association data is the product service feature.

See, for example, the architecture diagram of the risk user management scheme of FIG. 2 b. In this embodiment, three types of data, i.e., a user history profile, user behavior data, and product service access correlation data, are first obtained from user correlation data of a target user, and then user history profile features (i.e., the user attribute features), the user behavior features, and the product service features are respectively extracted based on the three types of data, and are used as customer service emergency degree evaluation features of three dimensions. And inputting the customer service emergency degree evaluation characteristics of the three dimensions into a risk identification model to predict the risk attribute information of the user.

204. Predicting risk attribute information of the target user based on the customer service emergency degree evaluation characteristics, wherein the risk attribute information is used for expressing the customer service emergency degree aiming at the target user;

in this embodiment, a trained risk recognition model may be used to predict risk attribute information of a user based on customer service urgency assessment features, where the risk recognition model may be implemented based on machine learning (M L), such as Deep learning (Deep L earning), where machine learning is a multi-domain interdisciplinary, and involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory.

In one example, the risk identification model may adopt a GBDT (gradient boosting iterative decision tree) model, a classification model based on Collaborative Filtering (CF), L R (L logistic regression) model, an FM (Factor machine) model, or a model based on the GBDT + L R principle, and the like, which is not limited in this embodiment.

Optionally, the step of "predicting risk attribute information of the target user based on the customer service urgency assessment feature" may include:

acquiring a trained risk identification model, wherein the risk identification model comprises: the incidence relation between the customer service emergency degree evaluation characteristics of the user and the risk attribute information;

and predicting the risk attribute information of the target user based on the customer service emergency degree evaluation characteristics of the target user through a risk identification model.

The risk attribute information in this embodiment may include risk user identification information, risk user level information, and the like, where the risk user identification may be used to indicate whether the target user is a risk user (if the target user is a risk user, the customer service urgency degree of the target user may be "urgent", and if the target user is a non-risk user, the customer service urgency degree of the target user may be "non-urgent"), where the risk user level may be a division of the risk level of the user, and the specific number of levels may be determined based on the number of expected risk user levels in the sample label when the risk recognition model is trained. And different risk user grades have different corresponding customer service emergency degrees, the highest risk user grade has the highest customer service emergency degree, and the lower the risk user grade is, the lower the customer service emergency degree is.

In this embodiment, the incidence relation between the customer service urgency evaluation characteristics of the user and the risk attribute information described above may exist in the form of a weight matrix between neural networks in the risk identification model, where the weight matrix is learned by the risk identification model during training. The training process of the model is described in detail later.

The value of the risky user identifier may be two, for example, 0 and 1 respectively indicate that the corresponding user is a non-risky user and a risky user. The present embodiment does not limit this.

205. And determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product.

Optionally, the customer service response scheme in this embodiment includes, but is not limited to, a target customer service channel set for the target user, and/or a slowest response speed corresponding to the customer service request, and/or a return policy corresponding to the customer service request (whether return visit is required, a return visit mode (including a return visit customer service channel), and the like).

Optionally, the risk attribute information in this embodiment includes risk user identification information, where the risk user identification information is used to indicate whether a corresponding user is a risk user.

The step of determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product may include:

determining risk users in the target users based on the risk user identification information of the target users to obtain a risk user set;

acquiring user risk scores of risk users in a risk user set based on user association data of a target user and a preset risk score rule;

ranking the risk users in the risk user set based on the user risk scores;

and determining a customer service response scheme of the risk user based on the sequencing result and the service capacity of the customer service channel of the target product.

Optionally, in this embodiment, the preset risk scoring rule may be obtained by weighting and summing scores corresponding to specific data(s) in the user-associated data.

For example, the user risk score of the target user is >3 (corresponding score 200) + number of incoming calls 1+ number of requests 2+ number of times of entering AI channels 2+ number of times of entering asynchronous channels 3+ number of times of entering IMC channels 1+ age 1+ VIP (30) + high consumption (20) + stolen (50) + payment exception (60)

In other examples, the user risk score may also be calculated based on other specific data, which is not limited by the present embodiment.

For example, referring to the architecture diagram of the risk user management scheme of fig. 2b, after the risk identification model outputs the predicted risk attribute information, user risk scoring may be performed on the risk users among the target users through the scoring model, ranking the risk users based on the scoring, and then outputting the scoring and ranking results. Thereafter, customer service response scenario settings may be made based on the scoring and/or ranking results.

In this embodiment, the risk users in the risk user set may be sorted in descending order according to the user risk scores. After the ranking, the priority can be set for the risk users based on the ranking order of the risk users, and the customer service response scheme can be set based on the priority. The service capability of the customer service channel in this embodiment includes, but is not limited to, the response timeliness of the customer service request.

The step of determining a customer service response scheme of the risk user based on the sequencing result and the service capacity of the customer service channel of the target product comprises the following steps:

setting service priority for the risk users based on the ranking results of the risk users, wherein the higher the risk user score is, the higher the service priority of the risk users is;

and determining a customer service response scheme of the risk user based on the service priority of the risk user.

In this embodiment, the number of the service priority levels may be the same as the number of customer service channels of the target product, and one customer service channel may serve as a target service channel for a service priority user. The stronger the service capability of the customer service channel, the higher the service priority of the corresponding user.

For example, assuming that the number of the risk users is 1000, the number of the customer service channels of the target product is 3, after the risk users are arranged according to the descending order of the user risk scores, the 1000 risk users are divided into three service priority levels according to the change rule of the user risk scores in the ordering result, the score corresponding to the first service priority level is the highest, the corresponding target customer service channel is the service channel with the strongest service capability, such as a telephone channel, the score corresponding to the second service priority level is the medium, the corresponding target customer service channel is the service channel with the medium service capability, such as an instant messaging client channel, the score corresponding to the third service priority level is the lowest, and the corresponding target customer service channel is the service channel with the weakest service capability, such as an email channel and the like.

In another example, the step "determining a customer service response scheme of the risk user based on the ranking result and the service capability of the customer service channel of the target product" may include:

determining the risk users distributed by each customer service channel based on the service capacity of the customer service channel of the target product and the number and the sequencing result of the risk users;

and setting different service priorities for risk users distributed by different customer service channels according to the strength of the service capacity of each customer service channel.

It can be understood that, in the above example, the number of service priorities is the same as the number of customer service channels of the target product, the service priority is high, and the service capacity of the corresponding customer service channel is also high.

For example, assuming that the number of the risk users is 1000, the number of the customer service channels of the target product is 3, the risk users are arranged in a descending order according to the user risk scores, the customer service channels respectively include customer service channels 1,2,3, the service capabilities are from strong to weak, based on the timeliness of the customer service request responses in the service capabilities of the customer service channels 1,2,3 and the ratio of the number of the customers capable of being simultaneously served, the risk users arranged in the first 100, the 101 th-.

In addition to the preset risk scoring rules, the user risk scoring can be predicted in a model manner.

Optionally, the step "determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product" may further include:

predicting the user risk score of the risk user based on the customer service emergency degree evaluation characteristics of the risk user in the risk user set through the trained risk score model;

ranking the risk users in the risk user set based on the user risk scores;

and determining a customer service response scheme of the target user based on the sequencing result and the service capacity of the customer service channel of the target product.

The risk scoring model may be any available model, such as a regression model.

The training sample of the risk scoring model may include customer service urgency assessment features of the user, and the sample label may include a desired risk user score for the user (which score may be set manually).

Optionally, in this embodiment, the risk attribute information may further include risk user level information;

determining the number of target users under each risk user level based on the risk user level information of the target users;

and determining a customer service response scheme of the target user under each risk user level based on the number of the target users under each risk user level and the service capacity of a customer service channel of the target product.

The number of the risk user grades can be set according to actual needs, for example, three grades including a high risk grade, a medium risk grade and a low risk grade are set.

Wherein, the number of the levels of the risk user level can be equal to the number of the customer service channels. One customer service channel may correspond to a user responsible for one level of risk users. Namely, in the customer service response scheme of the target user, the target customer service channel is set as the customer service channel corresponding to the risk user level.

In an example, when the number of the risk users in the risk user level is large, the target customer service channels of some of the risk users may also be set as the customer service channels corresponding to other risk user levels.

Optionally, in this embodiment, the customer service response scheme includes a target customer service channel set for the target user;

after determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product, the method can further comprise the following steps:

if a customer service request of a user for a target product is received, comparing a customer service channel of the customer service request initiated by the user with the service capacity of a target customer channel of the user;

and if the service capability of the target customer channel is relatively strong, switching to the target customer service channel to provide customer service corresponding to the customer service request for the user.

The service channel of the customer service request initiated by the user is not limited, and may be, for example, a mail channel.

For example, if a customer service request of a user for a target product is received through a mail of the target product, but a target customer service channel of the user is a telephone channel, the user telephone number in the mail or the user telephone number in the history record can be used for dialing the user telephone number to respond to the customer service request of the target product.

In this embodiment, after the step "determining the customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product", the method may further include:

determining a return visit priority of a target user based on risk attribute information of the target user;

setting a return visit strategy for the target user based on the return visit priority;

and performing a return visit on the historical customer service for the target user based on a return visit strategy corresponding to the target user and the historical customer service request of the target user, wherein the return visit content comprises the opinion (such as satisfaction degree) of the target user on the historical customer service request.

The number of the levels of the return visit priority can be set according to actual situations, for example, the return visit priority of the risky user is higher than that of the non-risky user, and the return visit policy includes, but is not limited to, a customer service channel for the return visit, a return visit frequency, a return visit speed, and the like.

Wherein, the return visit can be carried out through mail, instant communication message, short message, outbound call and other customer service channels.

In this embodiment, before using the risk recognition model, the risk recognition model may be trained according to the training samples. The training process of the risk identification model is described in detail below in conjunction with FIGS. 3a-3 c.

Referring to fig. 3a, the training method of the risk recognition model includes:

301. obtaining a training sample, wherein the training sample comprises customer service emergency degree evaluation characteristics of a user of a product on a plurality of customer service emergency degree evaluation dimensions, and a label of the training sample comprises expected risk attribute information of the user;

the number of the products corresponding to the training sample may be more than one, and optionally, the product corresponding to the training sample includes the target product.

In this embodiment, the customer service urgency evaluation features in the training samples are obtained based on historical customer service records of the user in multiple customer service channels of the product.

Optionally, the step of "obtaining a training sample" may include:

based on a user of a product, obtaining user association data generated based on the use of the product by the user in historical customer service records in at least two customer service channels provided by the product;

acquiring customer service emergency degree evaluation data of a user on a plurality of customer service emergency degree evaluation dimensions from the user associated data;

performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions;

obtaining a training sample corresponding to the user based on the customer service emergency degree evaluation characteristics corresponding to the user;

and acquiring expected risk attribute information of a user to which the training sample belongs as a label of the training sample.

The customer service emergency degree evaluation dimension may include: a user attribute dimension, a user behavior dimension and a product service dimension;

the step of obtaining customer service emergency degree evaluation data of the user in a plurality of customer service emergency degree evaluation dimensions from the user associated data may include:

acquiring user attribute data of the user in the product from the user associated data, wherein the user attribute data is used as customer service emergency degree evaluation data on the user attribute dimension;

acquiring user behavior data generated by the user by using the customer service channel from the user associated data, wherein the user behavior data is used as customer service emergency degree evaluation data on the user behavior dimension;

and acquiring product service access associated data of the product from the user associated data as customer service emergency degree evaluation data on the product service dimension.

Correspondingly, the step of performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation features of a plurality of customer service emergency degree evaluation dimensions may include:

and forming customer service emergency degree evaluation characteristics on the customer service emergency degree evaluation dimension by using vectors corresponding to the customer service emergency degree evaluation data on the same customer service emergency degree evaluation dimension to obtain the customer service emergency degree evaluation characteristics of the plurality of customer service emergency degree evaluation dimensions.

The concept and the obtaining manner of the user-related data, the historical customer service record, the customer service urgency evaluation dimension, and the customer service urgency evaluation feature may refer to the description in the foregoing example, and are not described herein again.

302. Acquiring a risk identification model to be trained;

the risk identification model in this embodiment may be constructed based on a training sample.

303. Performing feature extraction on the training sample based on the risk identification model to obtain sample features;

in one example, the risk identification model includes a feature extraction module configured to perform feature extraction on the training sample to obtain sample features.

304. Constructing a loss function of the risk identification model based on the sample characteristics and the expected risk attribute information;

in one example, the risk identification model includes a classification module to obtain predicted risk attribute information for the sample based on the sample features. Alternatively, the loss function may be constructed based on the predicted risk attribute information and the expected risk attribute.

305. And performing iterative training on the risk identification model based on the loss function and the training sample until the training of the risk identification model is completed.

In one example, the type of training samples is a triplet of training samples, where a set of triplet of training samples includes an anchor sample, a positive sample, and a negative sample. Wherein the positive samples are of the same type as the anchor samples, and the negative samples are of a different type from the anchor samples.

The same type of understanding is related to the information in the sample label, for example, the risk attribute information in the label of the sample includes a risk user identifier, then the users corresponding to the sample are classified into a risk user and a non-risk user, the same type of the anchor sample of the positive sample is understood as the same risk user identifier in the label of the positive sample and the anchor sample, that is, the corresponding user type is the same, and the different type of the negative sample and the anchor sample is understood as similar.

For another example, the risk attribute information in the label of the sample includes risk user level information, and the types of the anchor samples of the positive sample are the same, which can be understood that the risk user level information in the labels of the positive sample and the anchor samples is the same, that is, the corresponding risk user levels are the same. The understanding of the difference in the types of negative and anchor samples is similar.

For example, the triplet training sample in this embodiment is represented as < a, P, N >, where a (anchor) represents an anchor sample, P (positive) represents a positive sample, and N (negative) represents a negative sample; in the present model, the users in the sample can be classified into only two categories, risk users and non-risk users.

In this embodiment, there may be more than one label of the sample, depending on the type of information included in the risk attribute information, for example, the risk attribute information includes only risk user identification information, the number of labels of the sample is 1, the label is a risk user identification label, the label may include risk user identification information, and the samples of the risk user and the non-risk user may be represented by risk user identification information 1 and 0, respectively.

If the risk attribute information only includes risk user identification information and risk user grade information, the number of the tags of the sample is 2, and the method further includes the following steps except for the risk user identification tags: a risk level label comprising risk user level information.

In one example, the risk recognition model is trained based on Triplet L oss (triple loss function).

After data processing, the obtained triplet training samples may be: a ═ a₀,a₁,...,a_M}、P＝{p₀,p₁,...,p_M}、N＝{n₀,n₁,...,n_MAnchor samples, positive samples and negative samples, respectively. And M is a dimension of the vector and is determined based on the total number of the customer service emergency degree evaluation features in all the customer service emergency degree evaluation dimensions.

During the training process, the risk recognition model comprises three identical (identical in structure and parameters) recognizer models, for example, the model structure described with reference to fig. 3b, and the risk recognition model comprises three

recognizer models

31, 32, 33. All three submodels include the same feature extraction module, such as the three-layer fully-connected layer structure in fig. 3 b.

Optionally, the step of performing feature extraction on the training sample based on the risk recognition model to obtain the sample feature may include:

and performing feature extraction on the anchor sample, the positive sample and the negative sample in the triple training sample based on the risk identification model to respectively obtain the anchor sample feature, the positive sample feature and the negative sample feature.

Correspondingly, the step of "constructing a loss function of the risk identification model based on the sample characteristics and the expected risk attribute information" may include:

respectively predicting risk attribute information based on the anchor sample characteristics, the positive sample characteristics and the negative sample characteristics through a risk identification model to obtain predicted risk attribute information corresponding to the anchor sample, the positive sample and the negative sample;

constructing a first loss function of a risk identification model based on the predicted risk attribute information and the expected risk attribute information of the same training sample;

constructing a second loss function of the risk identification model based on the difference between the positive sample characteristics and the anchor sample characteristics corresponding to the same triple training sample and the difference between the negative sample characteristics and the anchor sample characteristics;

and obtaining a loss function of the risk identification model based on the first loss function and the second loss function.

In this embodiment, the structure of the recognition submodel may be described with reference to fig. 3c, and includes a feature extraction module 301, a classification task module 302, and a semantic matching task module 303, where the classification task module and the semantic matching task module share sample features output by the feature extraction module.

In this embodiment, the calculation of the first loss function is implemented based on the classification task module 302, and the calculation of the second loss function is implemented based on the semantic matching task module 303.

In this embodiment, the feature extraction module may be implemented based on any existing network structure that can be used for extracting features, for example, a CNN (Convolutional Neural Networks) network or based on a full connection layer.

Optionally, in an example, as shown in fig. 3b, the feature extraction module (311 or 321 or 331) is configured by three full connected layers (FC), in this embodiment, the triple training samples < a, P, N > obtain respective sample feature vectors through the three full connected layers, and finally, the identification capability of the risky user is enhanced by using a triple loss and a multitask manner, that is, the classification task is used to reduce the intra-class distance of the sample features of the same class, the semantic matching task is used to enhance the distinction among samples of different classes, increase the inter-class distance between sample classes, and the extraction capability of the model on the key features of the customer service emergency degree evaluation features is enhanced by using the combined training of two tasks and the construction of multiple loss functions, so as to enhance the identification capability of the risk attribute of the user.

Specifically, the anchor sample a ═ { a } in the triplet training sample₀,a₁,...,a_MP ═ P }, positive sample P ═ P₀,p₁,...,p_MN ═ N } and negative samples₀,n₁,...,n_MThe feature extraction modules of different identifier models are respectively input, for example, the feature extraction modules 321,311 and 331 formed by full connection layers in fig. 3b are respectively input, and three encoding vectors (namely, sample features) P _ Emb ∈ R corresponding to A, P, N samples are obtained by the three feature extraction modules^d,A_Emb∈R^d,N_Emb∈R^dWhere d is the vector dimension.

In this embodiment, the fully-connected layer may be expressed as FC ═ σ (ω · X + b), where X is the input and σ, ω, and b are the parameters.

In this embodiment, the risk identification model adopts a siamese structure, and weights of feature extraction modules of the three sub-models are shared. I.e. after each weight update of the risk identification model, the weights of the feature extraction modules of the three sub-models (e.g. the three fully-connected layers FC of the respective sub-models in fig. 3 b) are still the same.

After the sample characteristics are obtained, different loss function calculations are performed through the two task modules.

In this embodiment, risk attribute information prediction may be performed by the classification task module based on the anchor sample characteristics, the positive sample characteristics, and the negative sample characteristics, respectively, to obtain predicted risk attribute information corresponding to the anchor sample, the positive sample, and the negative sample. A first loss function of the risk identification model is then constructed based on the predicted risk attribute information and the expected risk attribute information.

The first loss function comprises a positive sample loss function, an anchor sample loss function and a negative sample loss function, the positive sample loss function is constructed on the basis of the expected risk attribute information and the predicted risk attribute information of the positive sample, the negative sample loss function is constructed on the basis of the expected risk attribute information and the predicted risk attribute information of the negative sample, and the anchor sample loss function is constructed on the basis of the expected risk attribute information and the predicted risk attribute information of the anchor sample.

For example, referring to fig. 3b, classification loss (i.e., a first loss function) is correspondingly constructed for each of the three submodels to reduce the intra-class distance of the same sample and improve the feature extraction capability of the same class, and the first loss function can be calculated by using a Cross entropy loss function Cross entry L _ C, whose structure is shown below.

Where T is represented as the number of category labels (for the risky user identification label, the number is 2, for the user risk level label, the number is the number of risky user levels), a_iValue of ith neuron of output layer of task module for classification, s_iIs the normalized probability value of the ith neuron, y_iThe softmax loss of three samples A, P, N is obtained via Cross entry as its true tag value, L _ CA, &lTtT translation = L "&gTt L &lTt/T &gTt _ CP, &lTtTtranslation = L" &gTt L &/T &gTt _ CN.

In one example, the second loss function may be the first sub-loss function, or the second sub-loss function, or the first sub-loss function and the second sub-loss function may be weighted sum results.

The calculation scheme of the first sub-loss function, or the second sub-loss function, may refer to the following description.

Optionally, the step of "constructing a second loss function of the risk identification model based on the difference between the positive sample feature and the anchor sample feature corresponding to the same triplet training sample and the difference between the negative sample feature and the anchor sample feature" may include:

taking a positive sample and an anchor sample in the same triple training sample as a positive sample pair, and taking a negative sample and the anchor sample as a negative sample pair;

calculating a first similarity of the positive sample features and the anchor sample features in the positive sample pairs and a second similarity of the negative sample features and the anchor sample features in the negative sample pairs;

constructing a first sub-loss function of the risk identification model based on the difference between the first similarity and the second similarity;

analyzing the interactivity of the positive sample characteristic and the anchor sample characteristic in the positive sample pair and the interactivity of the negative sample characteristic and the anchor sample characteristic in the negative sample pair based on a risk identification model to respectively obtain a positive sample pair interactive vector and a negative sample pair interactive vector;

predicting the similarity of the risk attribute information of the two samples in the positive sample pair based on the interaction vector of the positive sample pair through a risk identification model, and predicting the similarity of the risk attribute information of the two samples in the negative sample pair based on the interaction vector of the negative sample pair;

constructing a second sub-loss function of the risk identification model based on the prediction results of the positive sample pair and the negative sample pair;

and obtaining a second loss function of the risk identification model based on the first sub-loss function and the second sub-loss function.

In this embodiment, the calculation of the first sub-loss function and the second sub-loss function is performed based on the semantic matching task module.

For example, referring to fig. 3b again, based on the triplet training sample < a, P, N >, L oss of Tripletloss (triplet loss function, i.e. the first sub-loss function mentioned above) and Match loss (matching loss function, i.e. the second sub-loss function mentioned above) are constructed in the risk recognition model.

Wherein the first sub-loss function L _ T is:

L_T＝Max(d(A_Emb,P_Emb)-d(A_Emb,N_Emb)+m arg in,0)

wherein d (·, ·) is an euclidean distance, which is used to calculate a distance between vectors, and may represent a similarity between the vectors, a _ Emb is an anchor sample feature, P _ Emb is a positive sample feature, N _ Emb is a negative sample feature, d (a _ Emb, P _ Emb) is the first similarity, and d (a _ Emb, N _ Emb) is the second similarity.

The triple loss aims at reducing the distance between A _ Emb and P _ Emb and increasing the distance between A _ Emb and N _ Emb, so that the vectors of the same type of samples are more similar, and the vectors of different types of samples are distinguished, for example, the feature vectors of positive samples are more similar, the feature vectors of negative samples are more similar, and the feature vectors of positive samples and negative samples are more different.

The Triplet loss is to optimize the vector and weight distribution from the positive and negative samples as a whole, Match loss can optimize the vector and weight distribution from the positive and negative samples locally, each two sample pairs < a _ Emb, P _ Emb >, < a _ Emb, N _ Emb > respectively construct its Match loss, and since the vector distance is directly calculated and optimized in Triplet, we only use M L P (Multi layer per Perceptron) to construct the second sub-loss function in Match L oss.

For example, the interactivity of the positive sample feature and the anchor sample feature in the positive sample pair is analyzed based on a multi-layer perception mechanism to obtain a positive sample pair interaction vector, for example:

Pos_Emb＝[A_Emb:P_Emb:A_Emb-P_Emb:A_Emb*P_Emb]

for example, based on the interactivity of the negative sample feature and the anchor sample feature in the negative sample pair of the multi-layer perception mechanism, a negative sample pair interaction vector is obtained, for example:

Neg_Emb＝[A_Emb:N_Emb:A_Emb-N_Emb:A_Emb*N_Emb]

the classified submodule then predicts the similarity (e.g., similarity probability) of the risk attribute information for the two samples in the positive sample pair and the similarity (e.g., similarity probability) of the risk attribute information for the two samples in the negative sample pair, respectively, based on Pos _ Emb and Neg _ Emb.

A second sub-loss function is then constructed based on the predicted similarity of the sample pairs and the true similarity (the probability of similarity of positive sample pairs to true is 1, and the probability of similarity of negative sample pairs to true is 0).

The classification submodule can be realized based on a structure of a full connection layer.

The second sub-loss functions of this embodiment may be constructed based on Cross entry, and the second sub-loss functions corresponding to the positive and negative sample pairs are L _ MP, &lttttranslation = L "&gtt L &/ttt &gtt _ MN, respectively.

Wherein, a_iTo classify the value of the ith neuron of the output layer of the submodule, s_iIs the normalized probability value of the ith neuron, y_iFor example, the similarity probability of a positive sample to a true sample is 1, and the similarity probability of a negative sample to a true sample is 0.

In this embodiment, the first sub-loss function and the second sub-loss function may be weighted and summed to obtain the second loss function. And weighting and summing the first loss function and the second loss function to obtain a loss function of the risk identification model. In one example, the loss function of the risk identification model may be:

A_Loss＝w₀((L_CA+L_CP+L_CN)/3)+w₁(L_T)+w₂((L_MP+L_MN)/2)

wherein, w₀，w₁，w₂For weighting, in this embodiment, the risk identification model may be optimized based on any available optimization algorithm, such as SGD (stochastic Gradient Descent) optimization algorithm, GD (Gradient Descent) optimization algorithm, Adam (Adaptive motion Estimation) optimization algorithm, and the like. In this embodiment, after each weight parameter is adjusted, the weight parameters of the three feature extraction modules in the risk identification model are kept the same.

After the risk identification model to be trained is trained, a sub-model can be taken out from the risk identification model training, the semantic matching task module in the sub-model is removed, the classification task module is reserved, and the risk identification model for practical use is obtained.

In one example, the conditions under which the risk recognition model training is complete include, but are not limited to: and the training times of the risk identification model are not lower than a preset training time threshold, the loss value convergence value of the loss function of the risk identification model is below a preset maximum loss value, and the like.

By adopting the method of the embodiment, the user associated data can be collected based on a plurality of customer service channels; the method has the advantages that multi-dimensional feature analysis is carried out on users, users with wind control features are screened out from massive users, the users are processed preferentially, and for some users who are not processed timely, wind control events can be effectively prevented from occurring through calling out afterwards. In addition, the embodiment can also set a corresponding customer service response scheme based on the risk attribute information of the user, which is beneficial to distributing users with different customer service emergency degrees to a proper customer service channel, realizing resource optimization configuration, and being beneficial to reducing user complaints, improving customer service quality and improving user satisfaction.

Furthermore, the risk identification model in the embodiment is a Triplet L oss-based model, and the model does not need to manually construct a complex feature function, enhances the extraction capability of the key features of the sample, and improves the identification accuracy of the risk attributes of the user.

In order to better implement the method, correspondingly, the embodiment of the invention also provides a risk user management device, and the risk user management device can be specifically integrated in a terminal or a server.

Referring to fig. 4, the apparatus includes:

a user associated data obtaining unit 401, configured to obtain, based on historical customer service records of a target user in at least two customer service channels provided by a target product, user associated data generated based on usage of the target product by the target user;

an evaluation data obtaining unit 402, configured to obtain, from the user-related data, customer service emergency degree evaluation data of the target user in multiple customer service emergency degree evaluation dimensions;

a characteristic obtaining unit 403, configured to perform data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of multiple customer service emergency degree evaluation dimensions;

a prediction unit 404, configured to predict risk attribute information of the target user based on the customer service emergency degree evaluation feature, where the risk attribute information is used to indicate a customer service emergency degree for the target user;

and the plan management unit 405 is configured to determine a customer service response plan of the target user based on the risk attribute information of the target user and the customer service channel of the target product.

Optionally, the customer service urgency evaluation dimension includes: user attribute dimension, user behavior dimension and product service dimension.

An evaluation data obtaining unit 402, configured to obtain, from the user-related data, user attribute data of the target user in the target product, as customer service emergency degree evaluation data in a user attribute dimension; acquiring user behavior data generated by a target user by using a customer service channel from the user associated data, and taking the user behavior data as customer service emergency degree evaluation data on user behavior dimension; and acquiring product service access associated data of the target product from the user associated data as customer service emergency degree evaluation data on product service dimensionality.

Optionally, the feature obtaining unit 403 is configured to convert data in the customer service emergency degree evaluation data into a corresponding vector based on a vector conversion manner corresponding to the data type; and forming the customer service emergency degree evaluation characteristics on the customer service emergency degree evaluation dimension by using the vectors corresponding to the customer service emergency degree evaluation data on the same customer service emergency degree evaluation dimension to obtain the customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions.

Optionally, the predicting unit is configured to obtain a trained risk identification model, where the risk identification model includes: the incidence relation between the customer service emergency degree evaluation characteristics of the user and the risk attribute information; and predicting the risk attribute information of the target user based on the customer service emergency degree evaluation characteristics of the target user through a risk identification model.

Optionally, the apparatus of this embodiment further includes: the training unit is used for acquiring a training sample before the prediction unit predicts the risk attribute information of the target user based on the customer service emergency degree evaluation features, wherein the training sample comprises the customer service emergency degree evaluation features of the user of the product on a plurality of customer service emergency degree evaluation dimensions, and the label of the training sample comprises expected risk attribute information of the user; acquiring a risk identification model to be trained; performing feature extraction on the training sample based on the risk identification model to obtain sample features; constructing a loss function of the risk identification model based on the sample characteristics and the expected risk attribute information; and performing iterative training on the risk identification model based on the loss function and the training sample until the training of the risk identification model is completed.

Optionally, the training unit is specifically configured to obtain, based on a user of the product, user association data generated based on use of the product by the user in historical customer service records in at least two customer service channels provided by the product; acquiring customer service emergency degree evaluation data of a user on a plurality of customer service emergency degree evaluation dimensions from the user associated data; performing data processing on the customer service emergency degree evaluation data to obtain customer service emergency degree evaluation characteristics of a plurality of customer service emergency degree evaluation dimensions; obtaining a training sample corresponding to the user based on the customer service emergency degree evaluation characteristics corresponding to the user; and acquiring expected risk attribute information of a user to which the training sample belongs as a label of the training sample.

Optionally, the type of the training sample is a triple training sample, and a group of triple training samples includes an anchor sample, a positive sample and a negative sample;

optionally, the training unit is specifically configured to perform feature extraction on an anchor sample, a positive sample and a negative sample in the triple training sample based on the risk identification model, so as to obtain an anchor sample feature, a positive sample feature and a negative sample feature respectively; constructing a loss function of a risk identification model based on the sample characteristics and the expected risk attribute information, and constructing a first loss function of the risk identification model based on the predicted risk attribute information and the expected risk attribute information of the same training sample; constructing a second loss function of the risk identification model based on the difference between the positive sample characteristics and the anchor sample characteristics corresponding to the same triple training sample and the difference between the negative sample characteristics and the anchor sample characteristics; and obtaining a loss function of the risk identification model based on the first loss function and the second loss function.

Optionally, the training unit is specifically configured to use a positive sample and an anchor sample in the same triple training sample as a positive sample pair, and use a negative sample and the anchor sample as a negative sample pair; calculating a first similarity of the positive sample features and the anchor sample features in the positive sample pairs and a second similarity of the negative sample features and the anchor sample features in the negative sample pairs; constructing a first sub-loss function of the risk identification model based on the difference between the first similarity and the second similarity; analyzing the interactivity of the positive sample characteristic and the anchor sample characteristic in the positive sample pair and the interactivity of the negative sample characteristic and the anchor sample characteristic in the negative sample pair based on a risk identification model to respectively obtain a positive sample pair interactive vector and a negative sample pair interactive vector; predicting the similarity of the risk attribute information of the two samples in the positive sample pair based on the interaction vector of the positive sample pair through a risk identification model, and predicting the similarity of the risk attribute information of the two samples in the negative sample pair based on the interaction vector of the negative sample pair;

constructing a second sub-loss function of the risk identification model based on the similarity prediction results of the positive sample pair and the negative sample pair;

Optionally, the risk attribute information includes risk user identification information, and the risk user identification information is used to indicate whether a corresponding user is a risk user.

The scheme management unit is used for determining risk users in the target users based on the risk user identification information of the target users to obtain a risk user set; acquiring user risk scores of risk users in a risk user set based on user association data of a target user and a preset risk score rule; ranking the risk users in the risk user set based on the user risk scores; and determining a customer service response scheme of the risk user based on the sequencing result and the service capacity of the customer service channel of the target product.

Optionally, the scheme management unit is configured to determine a risk user in the target user based on the risk user identification information of the target user, so as to obtain a risk user set; predicting the user risk score of the risk user based on the customer service emergency degree evaluation characteristics of the risk user in the risk user set through the trained risk score model; ranking the risk users in the risk user set based on the user risk scores; and determining a customer service response scheme of the target user based on the sequencing result and the service capacity of the customer service channel of the target product.

Optionally, the risk attribute information includes risk user level information; the scheme management unit is used for determining the number of target users under each risk user level based on the risk user level information of the target users; and determining a customer service response scheme of the target user under each risk user level based on the number of the target users under each risk user level and the service capacity of a customer service channel of the target product.

In an example, the customer service response scheme includes a target customer service channel set for the target user, and the apparatus of this embodiment further includes a response unit, configured to compare, after the plan management unit determines the customer service response scheme for the target user based on the risk attribute information of the target user and the customer service channel of the target product, if a customer service request for the target product is received from the user, a service capability of the customer service channel from which the customer service request is initiated by the user with a service capability of the target customer channel of the user; and if the service capability of the target customer channel is relatively strong, switching to the target customer service channel to provide customer service corresponding to the customer service request for the user.

By adopting the device of the embodiment, the multi-dimensional characteristic analysis can be performed on the user based on the user associated data of the plurality of customer service channels, the risk attribute information of the user is determined, and the corresponding customer service response scheme is set, so that the users with different customer service emergency degrees can be distributed to the appropriate customer service channels, the occurrence probability of the pneumatic control event is reduced, the resource optimization configuration is realized, the complaint of the user is reduced, the customer service quality is improved, and the satisfaction degree of the user is improved.

In addition, an embodiment of the present invention further provides a computer device, where the computer device may be a terminal or a server, as shown in fig. 5, which shows a schematic structural diagram of the computer device according to the embodiment of the present invention, and specifically:

the computer device may include components such as a processor 501 of one or more processing cores, memory 502 of one or more computer-readable storage media, a power supply 503, and an input unit 504. Those skilled in the art will appreciate that the computer device configuration illustrated in FIG. 5 does not constitute a limitation of computer devices, and may include more or fewer components than those illustrated, or some components may be combined, or a different arrangement of components. Wherein:

the processor 501 is a control center of the computer device, connects various parts of the entire computer device by using various interfaces and lines, and performs various functions of the computer device and processes data by running or executing software programs and/or modules stored in the memory 502 and calling data stored in the memory 502, thereby monitoring the computer device as a whole. Optionally, processor 501 may include one or more processing cores; preferably, the processor 501 may integrate an application processor, which mainly handles operating systems, user interfaces, application programs, etc., and a modem processor, which mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 501.

The memory 502 may be used to store software programs and modules, and the processor 501 executes various functional applications and data processing by operating the software programs and modules stored in the memory 502. The memory 502 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to use of the computer device, and the like. Further, the memory 502 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 502 may also include a memory controller to provide the processor 501 with access to the memory 502.

The computer device further comprises a power supply 503 for supplying power to the various components, and preferably, the power supply 503 may be logically connected to the processor 501 through a power management system, so that functions of managing charging, discharging, power consumption, and the like are realized through the power management system. The power supply 503 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The computer device may also include an input unit 504, and the input unit 504 may be used to receive input numeric or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

Although not shown, the computer device may further include a display unit and the like, which are not described in detail herein. Specifically, in this embodiment, the processor 501 in the computer device loads the executable file corresponding to the process of one or more application programs into the memory 502 according to the following instructions, and the processor 501 runs the application programs stored in the memory 502, so as to implement various functions as follows:

The above operations can be implemented in the foregoing embodiments, and are not described in detail herein.

It will be understood by those skilled in the art that all or part of the steps of the methods of the above embodiments may be performed by instructions or by associated hardware controlled by the instructions, which may be stored in a computer readable storage medium and loaded and executed by a processor.

To this end, an embodiment of the present invention further provides a storage medium, where a plurality of instructions are stored, and the instructions can be loaded by a processor to execute the method for risk user management provided in the embodiment of the present invention.

Wherein the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

Since the instructions stored in the storage medium can execute the steps in the risk user management method provided in the embodiment of the present invention, beneficial effects that can be achieved by the risk user management method provided in the embodiment of the present invention can be achieved, which are detailed in the foregoing embodiments and will not be described herein again.

The method, the apparatus, the computer device and the storage medium for risk user management provided by the embodiments of the present invention are described in detail above, and a specific example is applied in the present disclosure to explain the principle and the implementation of the present invention, and the description of the above embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for those skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims

1. A method for risk user management, comprising:

2. The risk user management method of claim 1, wherein the customer service urgency assessment dimension comprises: a user attribute dimension, a user behavior dimension and a product service dimension;

the step of obtaining the customer service emergency degree evaluation data of the target user in a plurality of customer service emergency degree evaluation dimensions from the user associated data comprises the following steps:

acquiring user attribute data of the target user in the target product from the user associated data, wherein the user attribute data is used as customer service emergency degree evaluation data on the user attribute dimension;

acquiring user behavior data generated by the target user by using the customer service channel from the user associated data, wherein the user behavior data is used as customer service emergency degree evaluation data on the user behavior dimension;

and acquiring product service access associated data of the target product from the user associated data as customer service emergency degree evaluation data on the product service dimension.

3. The method for risk user management according to claim 1, wherein the performing data processing on the customer service urgency evaluation data to obtain customer service urgency evaluation characteristics of the plurality of customer service urgency evaluation dimensions comprises:

4. The method for risk user management according to claim 1, wherein the predicting risk attribute information of the target user based on the customer service urgency assessment characteristics comprises:

and predicting the risk attribute information of the target user based on the customer service emergency degree evaluation characteristics of the target user through the risk identification model.

5. The method of claim 4, further comprising, prior to predicting the risk attribute information of the target user based on the customer service urgency assessment characteristics:

obtaining a training sample, wherein the training sample comprises customer service emergency degree evaluation characteristics of a user of a product in a plurality of customer service emergency degree evaluation dimensions, and a label of the training sample comprises expected risk attribute information of the user;

acquiring a risk identification model to be trained;

performing feature extraction on the training sample based on the risk identification model to obtain sample features;

constructing a loss function of the risk identification model based on the sample features and the expected risk attribute information;

and iteratively training the risk identification model based on the loss function and the training sample until the risk identification model training is completed.

6. The at risk user management method of claim 5, wherein the obtaining training samples comprises:

acquiring customer service emergency degree evaluation data of the user on a plurality of customer service emergency degree evaluation dimensions from the user associated data;

7. The method of risk user management according to claim 5, wherein the type of the training samples is a triplet of training samples, a set of triplet of training samples comprising an anchor sample, a positive sample and a negative sample;

the extracting features of the training sample based on the risk recognition model to obtain sample features comprises:

performing feature extraction on an anchor sample, a positive sample and a negative sample in the triple training samples based on the risk identification model to respectively obtain an anchor sample feature, a positive sample feature and a negative sample feature;

constructing a loss function of the risk identification model based on the sample features and the expected risk attribute information, comprising:

respectively predicting risk attribute information based on the anchor sample characteristics, the positive sample characteristics and the negative sample characteristics through the risk identification model to obtain predicted risk attribute information corresponding to the anchor sample, the positive sample and the negative sample;

constructing a first loss function of the risk identification model based on the predicted risk attribute information and the expected risk attribute information of the same training sample;

8. The method for risk user management according to claim 7, wherein the constructing the second loss function of the risk identification model based on the difference between the positive sample feature and the anchor sample feature and the difference between the negative sample feature and the anchor sample feature corresponding to the same triplet training sample comprises:

analyzing the interactivity of the positive sample characteristic and the anchor sample characteristic in the positive sample pair and the interactivity of the negative sample characteristic and the anchor sample characteristic in the negative sample pair based on the risk identification model to respectively obtain a positive sample pair interactive vector and a negative sample pair interactive vector;

predicting, by the risk identification model, similarity of risk attribute information of the two samples in the positive sample pair based on the positive sample pair interaction vector, and predicting similarity of risk attribute information of the two samples in the negative sample pair based on the negative sample pair interaction vector;

9. The method according to any one of claims 1 to 8, wherein the risk attribute information includes risk user identification information, and the risk user identification information is used to indicate whether a corresponding user is a risk user;

determining a customer service response scheme for the target user based on the risk attribute information of the target user and the customer service channel of the target product, comprising:

acquiring a user risk score of a risk user in the risk user set based on the user association data of the target user and a preset risk score rule;

ranking the risky users in the set of risky users based on the user risk scores;

10. The method according to any one of claims 1 to 8, wherein the risk attribute information includes risk user identification information, and the risk user identification information is used to indicate whether a corresponding user is a risk user;

ranking the risky users in the set of risky users based on the user risk score;

11. The method according to any one of claims 1 to 8, wherein the risk attribute information includes risk user rating information;

12. The risk user management method according to any of claims 1-8, wherein the customer service response scheme comprises a target customer service channel set for the target user;

after determining a customer service response scheme of the target user based on the risk attribute information of the target user and the customer service channel of the target product, the method further comprises:

if a customer service request of a user for the target product is received, comparing a customer service channel of the customer service request initiated by the user with the service capacity of a target customer channel of the user;

13. An apparatus for risk user management, comprising:

14. A storage medium having a computer program stored thereon, wherein the computer program when executed by a processor implements the steps of the method according to any of claims 1-12.

15. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the steps of the method according to any of claims 1-12 are implemented when the computer program is executed by the processor.