TWI662506B - Method for distributing customer services based on question forecasts - Google Patents

Abstract

A customer service allocation method based on prediction problems includes: receiving a record of customer service questions from at least one first user; connecting to a social network site to collect the first website of the at least one first user on the social network site Behavior; establishing a prediction model based on the behavior of the first website of the at least one first user and customer service question records; receiving a customer service question request from a second user; collecting second users from one of the social network sites Website behavior; based on the second website behavior of the second user, using a predictive model to predict one of the question types of the customer service request; and according to the question type, assigning one of the plurality of responders to reply to the second user.

Description

Customer service allocation method based on prediction problem

The invention relates to a customer service allocation method, in particular to a customer service allocation method based on prediction problems.

At present, many enterprises have telephone voice customer service, and users can make calls to ask customer service staff for questions. However, its construction cost is high, and the phone is busy, that is, a customer service staff can only reply to one user at a time. If the number of users consulted is too large, it will cause users to spend a considerable amount of waiting time, and also increase the cost of telephone charges.

With the popularity of the Internet, there is also a web real-time customer service system that allows users to connect to customer service web pages and enter real-time conversations with customer service personnel by entering text. However, this method still faces the problem that when the number of consultations is too large, users will need to wait for a considerable amount of time.

In view of this, an embodiment of the present invention proposes a customer service allocation method based on prediction problems, including: receiving a record of customer service questions from at least one first user; connecting to a social networking site to collect the at least one first user A first website behavior on a community website; establishing a prediction model based on the first website behavior and customer service question records of the at least one first user; receiving a customer service question request from a second user; collecting a second User's behavior on the second website of one of the social networking sites; predicting one of the types of questions from the customer service request using a predictive model based on the behavior of the second website of the second user; Reply to the second user.

Based on this, through the behavior of the website before the user's consultation, the category of the question that the user will consult is predicted, and it is assigned to the matching respondent. Depending on the type of problem, the choice of a robot or a live customer service to reply to the user can reduce the workload of the live customer service, at the same time reduce the load of the inquiry queue, and shorten the user waiting time. In addition, it enables real customer service to focus on issues in their areas of focus, which can improve customer service processing efficiency and customer satisfaction.

1 and FIG. 2 are schematic diagrams of hardware and software architectures for implementing a customer service allocation method based on prediction problems according to an embodiment of the present invention, respectively. The software system is installed in one or more servers 100. The server 100 includes a processor 110, a memory 120, a network module 130, a storage medium 140, and the like. The software is stored in the storage medium 140 and executed by the processor 110. The memory 120 is used for storing temporary data required when the software is executed. The storage medium 140 stores a plurality of databases to record user data. For example, the database may be a member database to store user's member data. The database can also be a website behavior database, a customer service question record database, a response database, etc., which will be described in detail later. The network module 130 is for connecting to the Internet for users to connect.

As shown in FIG. 2, the software includes a protocol conversion interface 210, a user interface layer 220, a logical operation layer 230, a data service layer 240, a scheduler 250, a responder 260, a data access layer 270, a security layer 280, and a database. 290. The protocol conversion interface 210 includes an application programming interface (API) of a social platform (such as a social networking site or communication software) to interface with messages using a compatible protocol. In this way, the user terminal can use a web browser or a specific application program to send a message of a customer service question request to the server 100, and extract the question in the message through the protocol conversion interface 210. The user interface layer 220 can provide a graphical operation interface, so that after the customer service personnel logs in to the server 100, operations can be performed on the operation interface. The operation interface may include one or more dialog boxes. The dialog content in each dialog box is converted through the protocol conversion interface 210 and transmitted between the user terminal and the server 100 through the social platform. The logic operation layer 230 includes one or more calculation programs for establishing a prediction model for asking questions. The data service layer 240 is capable of managing the database 290 and accessing the data in the database 290 via the data access layer 270. The security items that the security layer 280 is responsible for include user authorization 281, data security 282, network security 283, system security 284, and application security 285. The scheduler 250 is used to assign users to a certain responder 260 so that the user can obtain the customer service of the matched responder 260. The respondent 260 may be a response robot using artificial intelligence, or a live customer service. The live customer service is logged in to the server through a web browser or application, and can receive the assignment from the dispatcher 250.

3 and FIG. 4 are flowcharts (1) and (2) of a method for allocating customer service based on prediction problems according to an embodiment of the present invention, respectively. First, in order to establish a prediction model for customer service questions, steps S310 to S330 are performed first.

Step S310: Receive the customer service question record of at least one first user via the social platform, so that the live customer service can reply online to solve the problem. After the end of each user questioning process, the user's question and the response from the live customer service will be recorded in the customer service question record database (step S401).

Step S320: Connect to a social network website to collect at least one first user's website behavior on the social network website (herein referred to as the first website behavior) (step S402). The first website behavior includes the browsing history, click history, and interaction behavior of the first user on the social networking site. The browsing history can record other user pages that the first user has browsed, and the dwell time during the browsing. The click record can record the first user's click items on the social networking site, such as the evaluation of a speech (such as like or dislike), a message, etc. Interactions can be, for example, friend lists, tracked users, and the like.

In one embodiment, the member database can record the connections of the same user's accounts on different social networking sites or communication software. For example, establish a related list, which records the account, mobile phone number, and / or email of the same user on different social networking sites or communication software. In this way, through the table lookup method, even if the first user uses a different communication software from the social networking site for customer service questions, it can also be connected to one or more social networking sites to collect the first user's website on the social networking site. behavior.

In an embodiment, the order of step S310 and step S320 may be interchanged.

Through the above steps, the customer service record of the first user and the behavior of the first website can be collected. Next, in step S330, a prediction model is established according to the first website behavior of the at least one first user and the customer service question record. As shown in FIG. 4, after the behavior of the first website is collected, feature extraction and selection are performed on the collected data, and then one or more features are found (step S403). For example: number of friends, number of clicks on ads, number of comments, frequency of following friends, interests, etc. After the customer service question record is recorded, the question can be counted to find out the frequently consulted categories (step S404). For example, for bankers, there can be credit card issues, financial card issues, and fund investment issues. Then, the consultation questions can be classified, that is, the collected consultation questions are classified into the frequent consultation category (step S405). Here, the mutual relationship (Mutual Information) method can be used to establish the characteristic relationship between the behavior of the first website and the record of customer service questions. Through the mutual information method, the correlation between the characteristics in the behavior of the first website and the category of customer service question records can be measured. If the amount of information is greater, the correlation is higher. After completing steps S403 and S405, model training may be performed according to the selected features and problem classification (step S406) to generate a prediction model (step S407). Here, the random forest method (Random Forest) can be used to train the prediction model.

After the second user initiates a customer service question request (step S408), that is, the server 100 receives the second user's customer service question request via the social platform (step S340), the server 100 can connect to the social network site to collect the first The website behavior of the second user on the social network site (herein referred to as the second website behavior) (steps S350, S409). Here, the second user's behavior of the second user can also be collected on the social website based on the connection of the same user's account on different social websites or communication software recorded in the member database. The second website behavior includes the second user's browsing history, click history, and interaction behavior on the social networking site.

Next, in step S360, according to the behavior of the second website of the second user, a prediction model is used to predict a question type of the customer service question request. As shown in FIG. 4, feature extraction is performed on the behavior of the second website, that is, corresponding features are extracted from the behavior of the second website according to the feature item selected and extracted in step S403 (step S410). Then, the extracted features are input into a prediction model, and the type of the problem is predicted (S411).

Finally, in step S370, one of the plurality of responders 260 may be allocated to reply to the second user according to the type of the question. Referring to FIG. 5, it is a flowchart (3) of a method for allocating customer service based on a prediction problem according to an embodiment of the present invention. One of the responders 260 in the foregoing step S370 may be a live customer service or a response robot. Based on the type of problem predicted, the scheduler 250 may choose to respond to the second user by a live customer service or response robot. If the predicted problem type has a higher probability of being recorded in the answer database, the answering robot searches for the question corresponding to the customer service question request according to the answer database to obtain and respond to the answer corresponding to the second user (step S371 ). If the predicted question type has a higher probability than a question in the response database, the scheduler 250 may choose to have the answering robot reply to the second user with a multi-level menu for the second user to select the business item they want to ask (step S372), interacting with the second user through a multi-layer option, which is suitable for situations such as event registration and query matters. Alternatively, the scheduler 250 may also be assigned to a live customer service, and the live customer service may reply to the second user (step S372). Here, you can decide which live customer service is assigned according to the professional field and workload of the live customer service, which can reduce the waiting time of the second user and obtain good service quality. In one embodiment, the allocation of live customer service can also be determined according to the membership level of the second user. If the member level is higher, it can be assigned to the live customer service first. In some embodiments, the scheduler 250 may also choose to enter one of steps S371 to S273 according to other factors, such as the second user queue load.

Here, the data service layer 240 periodically or irregularly counts common questions in the customer service question record database, and establishes common questions and corresponding responses in the answer database. The answer database also contains keywords for frequently asked questions. The logic operation layer 230 includes a semantic detection unit, which can check whether the second user's dialog contains keywords of common questions, and thereby determine whether the second user's questions are common questions, for the scheduler 250 to choose to proceed to step S371 ~ One of steps S273.

In one embodiment, the first user and the second user may be the same person. That is, the prediction model is established based on the historical customer service question requests and website behaviors, and adjusted according to the new customer service question requests or / and website behaviors.

In summary, a customer service allocation method based on a prediction problem according to an embodiment of the present invention can predict the category of the user to consult the problem through the behavior of the website before the user's consultation, and assign it to a matching respondent. Depending on the type of problem, the choice of a robot or a live customer service to reply to the user can reduce the workload of the live customer service, at the same time reduce the load of the inquiry queue, and shorten the user waiting time. In addition, it enables real customer service to focus on issues in their areas of focus, which can improve customer service processing efficiency and customer satisfaction.

100‧‧‧Server

110‧‧‧ processor

120‧‧‧Memory

130‧‧‧Network Module

140‧‧‧Storage media

210‧‧‧ protocol conversion interface

220‧‧‧user interface layer

230‧‧‧Logical Operation Layer

240‧‧‧Data Service Layer

250‧‧‧ Scheduler

260‧‧‧ Respondents

270‧‧‧Data Access Layer

280‧‧‧security layer

281‧‧‧User License

282‧‧‧Data Security

283‧‧‧ Cyber Security

284‧‧‧System Security

285‧‧‧Application Security

290‧‧‧Database

291‧‧‧Member Database

292‧‧‧Website database

293‧‧‧Customer Question Record Database

294‧‧‧response database

S310‧‧‧ Receive at least one first user customer service question record

S320‧‧‧ is connected to a social website to collect the behavior of the at least one first user on a first website on the social website

S330‧‧‧ establish a prediction model based on the first website behavior and customer service question records of the at least one first user

S340‧‧‧ receives a customer service question request from a second user

S350 ‧‧‧ Collect second user behavior on one of the social networking sites

S360‧‧‧ predicts one of the types of customer service requests based on a prediction model based on the behavior of the second website of the second user

S370‧‧‧ According to the type of question, assign one of the plurality of respondents to reply to the second user

S371‧‧‧The answering robot searches for a question corresponding to a customer service request according to a response database, so as to obtain and respond to the answer corresponding to the second user.

S372‧‧‧The response robot responds to the second user with a multi-layer menu for the second user to choose

S373‧‧‧Determines the distribution of live customer service according to the professional field and workload of live customer service

S401‧‧‧ Generate customer service question record

S402‧‧‧Collecting the First Website

S403‧‧‧ Feature extraction and selection

S404‧‧‧Identify common consultation categories

S405‧‧‧Classification consultation questions

S406‧‧‧Training model

S407‧‧‧Generating prediction model

S408‧‧‧Initiate a customer service question request

S409‧‧‧ Collection of Second Website

S410‧‧‧Capture feature

S411‧‧‧ Prediction Problem Type

S412‧‧‧Scheduled responder

[FIG. 1] A schematic diagram of a hardware architecture for performing a customer service allocation method based on a prediction problem according to an embodiment of the present invention. FIG. 2 is a schematic diagram of a software architecture for executing a customer service allocation method based on a prediction problem according to an embodiment of the present invention. [FIG. 3] A flowchart of a customer service allocation method based on a prediction problem according to an embodiment of the present invention (1). [Fig. 4] A flowchart of a customer service allocation method based on a prediction problem according to an embodiment of the present invention (II). [FIG. 5] A flowchart of a customer service allocation method based on a prediction problem according to an embodiment of the present invention (3).

Claims (8)

A customer service allocation method based on prediction problems includes: receiving a record of customer service questions from at least one first user; connecting to a social network site to collect the first at least one first user on the social network site; Website behavior; establishing a prediction model based on the first website behavior of the at least one first user and the customer service question record; receiving a customer service question request from a second user; collecting the second user in the community A second website behavior of one of the websites; using the prediction model to predict a question type of the customer service request based on the second website behavior of the second user; and assigning a live customer service among the plurality of responders according to the problem type Or a response robot responds to the second user, wherein when the question type is not a question in a response database, the live customer service is assigned to reply to the second user, and the assigned live customer service is based on the The professional field and workload of live customer service are determined. The method for allocating customer service based on prediction questions as described in claim 1, further comprising: the answering robot searches for a question corresponding to the customer service question request according to the response database to obtain and respond to the answer corresponding to the second user. The method for allocating customer service based on the prediction problem according to claim 1, further comprising: the response robot responds to the second user with a multi-layer menu for the second user to choose. The method for allocating customer service based on prediction questions as described in claim 1, wherein the professional field of the live customer service corresponds to the question type of the customer service question request. The customer service allocation method based on the prediction problem described in claim 1, wherein the allocation of the live customer service is also determined according to the membership level of the second user, and if the second user has a higher membership level, it is allocated to the real person Customer service. The method for allocating customer service based on a prediction problem as described in claim 1, wherein the behavior of the first website includes the browsing history, click history, and interaction behavior of the first user on the social website, and the behavior of the second website includes the first 2. The browsing history, click history and interaction behavior of users on the social networking site. The method for allocating customer service based on a prediction problem as described in claim 1, wherein the establishing a prediction model includes: using a mutual information method to establish a characteristic relationship between the behavior of the first website and the customer service question record. The method for assigning customer service based on a prediction problem according to claim 1, wherein the establishing a prediction model includes: training the prediction by a random forest method according to the behavior of the first website of the at least one first user and the customer service question record model.