Question guiding method and device for intelligent customer service, medium and equipment

Abstract

The invention provides a question guiding method and device for intelligent customer service, a medium and equipment. In the method, a server can determine a corresponding target problem node from a preset user problem transfer graph by using a historical problem sent by a user, so as to plan a problem planning direction chain based on the target problem node, thereby generating a highly personalized and targeted recommendation problem for the user according to the problem planning direction chain. Therefore, the recommended question can be sent to the user to guide the user to define the direction of the question, meanwhile, accurate and compliant replies are quickly generated based on the real-time question of the user, and it is ensured that user requirements are effectively met.

Description

Technical Field

[0001] One or more embodiments of this specification relate to the field of artificial intelligence technology, and in particular, to a method, device, medium, and equipment for guiding questions in intelligent customer service. Background Art

[0002] With the development of social economy and the enhancement of people's risk awareness, more and more customers begin to pay attention to insurance products, resulting in a significant increase in the demand for insurance agent services. However, the shortage of insurance agent resources seriously affects the speed and quality of solving problems for customers. Based on this, the intelligent customer service system has gradually replaced some manual services and can automatically handle most simple problems of customers, thus alleviating the pressure on manual customer service, enabling manual insurance agents to concentrate on handling more complex customer needs and providing more accurate and effective services.

[0003] However, users often face problems such as unclear expression, inaccurate problem description, or not knowing how to ask questions when interacting with the intelligent customer service system. This may lead to the intelligent customer service system being unable to accurately understand the user's intention, and further result in a low interaction efficiency between the intelligent customer service system and the user.

[0004] Therefore, how to improve the interaction efficiency between the intelligent customer service system and the user is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, one or more embodiments of this specification provide the following technical solutions:

[0006] According to the first aspect of one or more embodiments of this specification, a method for guiding questions in intelligent customer service is proposed, including:

[0007] Obtain the user's historical questions;

[0008] Determine the node corresponding to the historical question from a preset user question transfer graph as the target question node, and determine at least one path containing the target question node as the question planning direction chain;

[0009] Input the historical question, the historical reply information of the historical question, and the question planning direction chain into a preset generation model to obtain the first type of alternative questions;

[0010] Determine the recommended questions according to the first type of alternative questions and send the recommended questions to the user.

[0011] According to the second aspect of one or more embodiments of this specification, a device for guiding questions in intelligent customer service is proposed, including:

[0012] An acquisition module for acquiring the user's historical questions;

[0013] A determination module for determining, from a preset user question transfer graph, the node corresponding to the historical question as the target question node, and determining at least one path containing the target question node as the question planning direction chain;

[0014] A question generation module for inputting the historical question, the historical reply information of the historical question, and the question planning direction chain into a preset generation model to obtain a first type of alternative questions;

[0015] A recommendation module for determining recommended questions according to the first type of alternative questions and sending the recommended questions to the user.

[0016] According to a third aspect of one or more embodiments of the present specification, an electronic device is provided, including: a processor; a memory for storing processor-executable instructions; wherein, the processor realizes the steps of the question guiding method for intelligent customer service as described above by running the executable instructions.

[0017] According to a fourth aspect of one or more embodiments of the present specification, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the question guiding method for intelligent customer service as described above are realized.

[0018] According to a fifth aspect of one or more embodiments of the present specification, a computer program product is provided, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the question guiding method for intelligent customer service as described above are realized.

[0019] As can be seen from the above embodiments, the present specification acquires the user's historical questions, determines the node corresponding to the historical question from a preset user question transfer graph as the target question node, and determines at least one path containing the target question node as the question planning direction chain. Furthermore, the historical question, the historical reply information of the historical question, and the question planning direction chain are input into a preset generation model to obtain a first type of alternative questions. According to the first type of alternative questions, recommended questions are determined and sent to the user.

[0020] In this method, the corresponding target question node can be determined from a preset user question transfer graph by using the historical questions sent by the user, so as to plan a question planning direction chain based on the target question node. Thus, highly personalized and targeted recommended questions can be generated for the user according to the question planning direction chain. Furthermore, the recommended questions can be sent to the user to guide them to clarify the question direction, thereby improving the interaction efficiency with the user. Brief Description of the Drawings

[0021] Figure 1 This is a flowchart of a question guiding method for intelligent customer service provided by an exemplary embodiment;

[0022] Figure 2 is a schematic diagram of an automated problem handling system provided by an exemplary embodiment;

[0023] Figure 3 is a schematic diagram of a process for generating a first round of recommendation questions provided by an exemplary embodiment;

[0024] Figure 4 A schematic diagram of a process for determining a takeover opportunity provided by an exemplary embodiment;

[0025] Figure 5 is a user problem transfer diagram provided by an exemplary embodiment;

[0026] Figure 6 is a schematic diagram of a determination process of a recommendation question provided by an exemplary embodiment;

[0027] Figure 7 is a structural schematic diagram of a device provided by an exemplary embodiment;

[0028] Figure 8 It is a block diagram of a question guiding device for intelligent customer service provided by an exemplary embodiment. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0030] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0031] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0032] Figure 1It is a schematic flow chart of a question guiding method for an intelligent customer service provided by an exemplary embodiment, including:

[0033] S100: Obtain the user's historical questions.

[0034] In this specification, when a user executes an insurance product business process, they may encounter problems such as: not knowing the calculation rules of the insurance product's premium, not knowing what the specific exemption clauses of the insurance product refer to, not knowing what the waiting period of the insurance product means, not knowing how to determine the compensation ratio of the insurance product, not knowing who can be designated as the beneficiary of the insurance product, etc. At this time, the user can input questions through the interaction interface provided by the business platform to seek help from the automated question receiving system set in the business platform.

[0035] Furthermore, the automated question receiving system of the business platform can obtain the user's questions, analyze the questions sent by the user, generate a reply message to send to the user, and display the reply message in the interaction interface to the user. Among them, the above-mentioned automated question receiving system is as Figure 2 shown.

[0036] Figure 2 It is a schematic diagram of an automated question receiving system provided by an exemplary embodiment.

[0037] In Figure 2 , the above-mentioned automated question receiving system may include: a timing judgment module for taking over, a starting phrase taking over module, a first-round guessing question module, a question recommendation module, and a multi-round question taking over module. The following will combine Figure 2 to explain in detail the process of the business platform performing text replies through the automated question receiving system.

[0038] In an actual application scenario, when the user opens the interaction interface, they may not know how to ask questions, resulting in a low interaction efficiency between the user and the automated question receiving system. Therefore, in order to improve the interaction efficiency between the user and the automated question receiving system, when the business platform monitors that the user enters the line for consultation, it can generate a first-round dialogue taking over message through the automated question receiving system and send it to the user, so that the user can conduct consultations based on the first-round dialogue taking over message, thereby improving the interaction efficiency with the user.

[0039] Among them, the above-mentioned first-round dialogue taking over message may refer to personalized taking over content highly relevant to the user's intention, which is mainly used to effectively stimulate the user's consultation interest to improve the success rate of dialogue initiation. Here, the first-round dialogue taking over message may include: a starting phrase and a first-round recommended question.

[0040] Specifically, when the business platform monitors the user's online consultation, it can obtain the user's user characteristics through the opening words acceptance module of the automated question acceptance system, and select the target opening words template from the preset opening words templates based on the user's user characteristics, and then input the user's user characteristics and the target opening words template into the preset opening words generation model to obtain the opening words.

[0041] The above-mentioned user characteristics include: user attribute characteristics and user behavior characteristics. The user attribute characteristics here refer to static information used to reflect the user's identity, financial status, and preferences, such as: age, occupation, place of residence, income level, marital status, education level, insurance product purchase record, health status, etc. The user behavior characteristics here refer to information used to reflect the user's dynamic interactive behavior in the business platform, such as: the frequency of users' online consultations, users' product browsing records, users' purchasing behavior, and users' jump paths in different pages or functional modules provided by the business platform.

[0042] The above opening phrase template is pre-built based on the historical conversation records between the manual customer service and the user. The opening phrase template here is used to specify the content structure and format of the opening phrase. For example, the opening phrase template defines the greeting, self-introduction and conversation guide that need to be included in the beginning of the opening phrase. Another example: the opening phrase template defines the inappropriate or sensitive words that need to be avoided in the main part of the opening phrase. Another example: the opening phrase template defines the format of the recommendation phrase in the opening phrase for products that the user may be interested in.

[0043] The above-mentioned method of selecting a target opening phrase template from various preset opening phrase templates according to the user characteristics of the user may be to predict the user's call intention according to the user characteristics of the user, and select a starting phrase template that matches the user's call intention from various preset opening phrase templates according to the call intention as the target opening phrase template. The call intention here may refer to the specific needs of the user that the user may want to obtain answers or help through the automated question acceptance system as reflected by the user characteristics of the user. The call intention here may include: product consultation intent for understanding a certain product, claims-related intent for understanding the specific process of claims, service support intent for operations such as querying the policy status and handling policy changes, product recommendation intent for understanding how to select a combination of insurance products, etc.

[0044] From the above content, it can be seen that after monitoring the user's online consultation, the business platform can send a personalized opening sentence to the user to stimulate the user's consultation interest, thereby improving the success rate of starting a conversation. In addition, the business platform can also send the first round of recommended questions to the user after sending the opening sentence to the user, so as to improve the interaction efficiency with the user. Of course, the business platform can also directly send the first round of recommended questions to the user after monitoring the user's online consultation, and this manual does not limit this. For ease of understanding, the following is combined with Figure 3 The generation process of the first round of recommendation questions in the above content is explained in detail.

[0045] Figure 3 It is a schematic diagram of a process of generating first-round recommendation questions provided by an exemplary embodiment.

[0046] Combination Figure 3 It can be seen that the business platform can obtain the user characteristics of the user, and determine the user characteristic node corresponding to the user characteristic of the user from the preset user characteristic and user question relationship graph as the target user characteristic node, and then determine the question corresponding to the question node connected to the target user characteristic node by an edge from the preset user characteristic and user question relationship graph as the basic recommendation question, and then query the questions matching the basic recommendation question from the preset question library as the first round of recommendation questions.

[0047] It should be noted that the above-mentioned user feature and user question relationship diagram is constructed based on the historical conversation records between the automated question acceptance system and different users stored in the business platform, and the above-mentioned user feature and user question relationship diagram includes user feature nodes and question nodes. Among them, different user feature nodes are used to represent different user features that users have before online consultation, and different question nodes are used to represent different questions raised by users after consultation. In addition, the edge between the user feature node and the question node is used to represent that there is a user with the user feature corresponding to the user feature node connected to the edge, and after the online consultation, the question corresponding to the question node connected to the edge is raised. In other words, if there is user A, and user A has user feature a, when user A raises question b after online consultation, then an edge can be used in the user feature and user question relationship diagram to connect the user feature node used to represent user feature a with the question node used to represent question b.

[0048] Among them, the above-mentioned problem node connected to the target user feature node by an edge may refer to a problem node directly connected to the target user feature node by an edge, or may refer to a problem node indirectly connected to the target user feature node through an intermediate node, and the number of intermediate nodes here may be less than a preset number threshold.

[0049] For example, if the above-mentioned quantity threshold is 3, for the user feature node a, there is a problem node b directly connected to it, and there is a user feature node c directly connected to the problem node b. Finally, there is a problem node d directly connected to the user feature node c and indirectly connected to the user feature node a through the problem node b and the user feature node c. At this time, it can be known that the problem node d is indirectly connected to the user feature node a through two intermediate nodes (i.e., the problem node b and the user feature node c), and the number of intermediate nodes between the problem node d and the user feature node a is less than 3. At this time, it can be considered that both the problem node b and the problem node d are problem nodes connected to the user feature node a by an edge.

[0050] It should be noted that the business platform can directly use the basic recommended questions determined from the preset user feature and user problem relationship graph as the first-round recommended questions. However, since in actual application scenarios, there may be many variants for the same problem, after the business platform obtains the basic recommended questions, it can also query the preset question library to obtain questions that match the basic recommended questions as the first-round recommended questions.

[0051] Among them, the above-mentioned question library can be constructed based on the historical questions sent by the users of the business platform. Specifically, the business platform can obtain the historical questions sent by each user of the business platform and hand over each historical question to be rewritten by an artificial person to obtain the rewritten questions, and then can construct a question library based on the rewritten questions.

[0052] The method of querying the preset question library to obtain questions that match the basic recommended questions can be that if it is determined that the similarity between the question and the basic recommended question is higher than the preset similarity threshold, then it can be considered that the question matches the basic recommended question.

[0053] In the field of insurance product recommendation, since different products may involve different questions. For example, the user questions that frequently appear during the consultation process of one product do not exist during the consultation process of other products. Therefore, in order to improve the accuracy of the first-round recommended questions recommended to users, the business platform can also determine the basic recommended questions from the preset user feature and user problem relationship graph according to the user's features, and determine the products involved in the user's historical conversation records as the target products. Then, it can query the preset question library to obtain questions that match the basic recommended questions and the target products as the first-round recommended questions.

[0054] In this specification, in order to further improve the accuracy of the determined first-round recommended questions, the business platform can also query questions that match the basic recommended questions and the target product from a preset question library as alternative first-round recommended questions, and then select the first-round recommended questions from each alternative first-round recommended question.

[0055] Specifically, there are three methods for the business platform to select the first-round recommended questions from each alternative first-round recommended question. The following will explain these three methods in detail respectively.

[0056] The first method for selecting the first-round recommended questions from each alternative first-round recommended question can be to generate a prompt based on the user's behavior characteristics and each alternative first-round recommended question, and input the prompt into a preset sorting model to sort each alternative first-round recommended question through the sorting model to obtain a sorting result. Then, the first-round recommended questions can be selected from each alternative first-round recommended question according to the sorting result.

[0057] Among them, the above-mentioned prompt is used to instruct the sorting model to determine the intensity of the user's inquiry intention for each alternative first-round recommended question according to the user's behavior characteristics, and sort each alternative first-round recommended question according to the intensity of the user's inquiry intention for each alternative first-round recommended question.

[0058] The second method for selecting the first-round recommended questions from each alternative first-round recommended question can be that the business platform can also select the first-round recommended questions from each alternative first-round recommended question according to the real-time click exposure volume of the user corresponding to each alternative first-round recommended question.

[0059] Among them, the above-mentioned real-time click exposure volume of the user can refer to the number of times an alternative first-round recommended question is recommended to different users and the number of clicks of different users on this alternative first-round recommended question (i.e., the number of times different users adopt this alternative first-round recommended question) within a specific time period (such as: within the most recent hour, within the most recent day, etc.).

[0060] The third method for selecting the first-round recommended questions from each alternative first-round recommended question can be that the business platform can also select the first-round recommended questions from each alternative first-round recommended question according to the user characteristics.

[0061] It should be noted that the above three methods can be used alone, or two or more methods can be used simultaneously. The following takes the simultaneous use of the three methods as an example to explain in detail the method of selecting the first-round recommended questions from each alternative first-round recommended question.

[0062] Specifically, the business platform can sort the various alternative first-round recommended questions through a sorting model to obtain a sorting result, and then select at least some of the alternative first-round recommended questions from the various alternative first-round recommended questions based on the sorting result as the initially selected first-round recommended questions, and then select the first-round recommended questions from the various alternative first-round recommended questions based on the real-time user click exposure and user characteristics corresponding to each alternative first-round recommended question.

[0063] In actual application scenarios, the interactive interface that users visit each time they consult online is often a fixed interactive interface, and users usually do not create new interactive interfaces on their own. Therefore, the interactive interface used by users each time they consult online often contains historical conversation records of the last conversation. In addition, in the process of conversing with users, users may temporarily cut out the interactive page with the automated problem acceptance system due to misoperation, secondary browsing of product pages, etc., that is, users may be temporarily offline after consulting online, and when the user is back online, if it is not certain whether the previous conversation with the user has been completed, each time the user goes online may be considered to need to reopen the conversation (that is, the messages saved in the interactive interface are considered to be the interactive content of the previous conversation), so the above-mentioned first round of conversation acceptance information is sent to the user again, which leads to a reduction in user experience and may reduce the efficiency of interaction with the user.

[0064] Based on this, after monitoring the user's online consultation, the business platform can also determine whether the conversation with the user is completed through the acceptance timing judgment module in the automated question acceptance system, and choose to send the content of the conversation acceptance to the user according to the completion of the conversation with the user. Figure 4 shown.

[0065] Figure 4 A schematic diagram of a process for determining a takeover opportunity provided by an exemplary embodiment.

[0066] Combination Figure 4 It can be seen that after monitoring the user's online consultation, the business platform can also obtain the user's historical conversation records through the automated problem acceptance system, and when it is determined that the previous conversation with the user has ended based on the historical conversation records, the conversation time of the historical conversation records (the sending time of the last message sent by the user in the historical conversation record can be used as the conversation time of the historical conversation record) and the current time, the pre-generated first-round conversation acceptance information is sent to the user to reopen the conversation with the user.

[0067] In addition, when it is determined that the previous conversation with the user is not completed based on the historical conversation records, the conversation time of the historical conversation records and the current time, the user's historical questions are obtained from the historical conversation records, and the preset question recommendation module is used to determine the recommended questions based on the user's historical questions and provide them to the user, so that the user can send the target question based on the recommended question, and then the preset multi-round question acceptance module can be used to generate reply information and reply to the user based on the conversation record of the current conversation with the user.

[0068] The above-mentioned historical questions may refer to the questions with the latest sending time obtained from the historical conversation records. In other words, when it is determined that the conversation with the user is not completed, the questions sent by the user in the last round may be obtained as historical questions.

[0069] It should be noted that the above questions can be stored in a variety of different formats, such as text format, image format, audio format, etc.

[0070] In this specification, the execution entity used to implement the question-guiding method for intelligent customer service may refer to a designated device such as a server set up in a business platform, or may refer to a terminal device such as a desktop computer or a laptop computer. For the sake of ease of description, the following only takes the server as an example of the execution entity to illustrate the question-guiding method for intelligent customer service provided in this specification.

[0071] S102: Determine the node corresponding to the historical problem from the preset user problem transfer graph as the target problem node, and determine at least one path including the target problem node as the problem planning direction chain.

[0072] After obtaining the user's historical questions, the server can determine the node corresponding to the historical question from the preset user question transfer graph as the target question node, and determine at least one path including the target question node as the question planning direction chain, as follows: Figure 5 shown.

[0073] Figure 5 It is a user problem transition diagram provided by an exemplary embodiment.

[0074] Combination Figure 5 It can be seen that in the above user question transfer graph, different nodes are used to represent different questions sent by users. For any two nodes, the edge between the two nodes is used to represent that the questions corresponding to the two nodes appear in the adjacent time sequence. In other words, for any question, if the user asks another question after asking the question, the nodes corresponding to the two questions can be connected by an edge.

[0075] Furthermore, after determining the target problem node, the server may perform a graph traversal starting from the target problem node to obtain a path including the target problem node as a problem planning direction chain, where the problem planning direction chain may consist of at least two nodes.

[0076] For example: Figure 5 The target problem node shown in , together with the node corresponding to problem A, the node corresponding to problem B, and the node corresponding to problem C, constitute a problem planning direction chain.

[0077] Another example: Figure 5 The target problem node in , the node corresponding to problem E, and the node corresponding to problem F together form another problem planning direction chain.

[0078] Another example: Figure 5 The target problem node in , together with the node corresponding to problem A, the node corresponding to problem E, and the node corresponding to problem F, constitute another problem planning direction chain.

[0079] It should be noted that the above-mentioned different problem planning direction chains can reflect the different logical transition relationships between problems of different users in the consultation process.

[0080] For example: After asking “How do I buy insurance?”, users will usually ask “Which insurance is suitable for me?” The automated question-taking system can accurately recommend follow-up questions based on this chain, thereby improving the targeted nature of the recommended questions.

[0081] For example, after asking about the “insurance claim process”, the user may further ask “what materials are needed for claim settlement?” The automated question-taking system can predict the user’s deeper needs based on this chain, thereby improving the comprehensiveness of the recommended questions.

[0082] S104: Input the historical questions, the historical response information of the historical questions and the question planning direction chain into a preset generation model to obtain a first type of candidate questions.

[0083] S106: Determine a recommended question based on the first category of candidate questions, and send the recommended question to the user.

[0084] Furthermore, after determining the problem planning direction chain, the server can obtain the first type of candidate questions according to the problem planning direction chain, as follows: Figure 6 shown.

[0085] Figure 6 It is a schematic diagram of a process of determining a recommendation question provided by an exemplary embodiment.

[0086] Combination Figure 6It can be seen that the server can input historical questions, historical reply information of historical questions, and the problem planning direction chain into a preset generation model to obtain the first type of alternative questions. Furthermore, the server can determine recommended questions based on the first type of alternative questions and send the recommended questions to the user to obtain the target questions sent by the user according to the recommended questions.

[0087] Among them, the method for the server to determine recommended questions based on the first type of alternative questions can be to sort each first type of alternative question according to user characteristics, and select recommended questions from each first type of alternative question according to the sorting result.

[0088] In addition, the server can also query each question from a preset question library according to the user's historical conversation record as the recalled questions. Furthermore, the server can select the second type of alternative questions from the recalled questions according to the user's characteristics, and then select at least some of the alternative questions from the first type of alternative questions and the second type of alternative questions as the recommended questions.

[0089] Among them, the method for the server to select the second type of alternative questions from the recalled questions according to the user's characteristics can be to input the user's characteristics and each recalled question into a preset selection model to determine, through the selection model, the recalled questions whose user inquiry intention intensity meets the preset conditions as the second type of alternative questions.

[0090] In addition, the server can also determine the process node that matches the user's historical conversation record and the user's characteristics from a preset product business interaction process chain as the target process node, and obtain the third type of alternative questions according to the target process node and the mapping diagram between the process node and the question. Thus, at least some of the alternative questions can be selected from the first type of alternative questions and the third type of alternative questions as the recommended questions.

[0091] Among them, different process nodes in the above product business interaction process chain are used to represent different interaction stages included in the process of interacting with the user.

[0092] The above interaction stage can refer to a specific link or step with a clear goal in the process of interacting with the user, such as: user information collection stage, demand analysis stage, insurance product portfolio design stage, product recommendation stage, after-sales service stage, etc.

[0093] The mapping diagram between the above process node and the question is used to represent the mapping relationship between different process nodes and different questions.

[0094] From the above content, it can be seen that in addition to generating alternative questions based on historical questions, historical response information of historical questions, and question planning direction chain through the generation model, the server can also obtain alternative questions through two other methods. In actual application scenarios, these three methods can also be used simultaneously, that is, the server can obtain the first category of alternative questions, the second category of alternative questions, and the third category of alternative questions through the above three methods respectively, and then can select at least some of the alternative questions from the first category of alternative questions, the second category of alternative questions, and the third category of alternative questions as recommended questions.

[0095] Furthermore, after obtaining the target question sent by the user, the server can generate reply information according to the target question and reply to the user.

[0096] Among them, there are many ways for the server to generate reply information based on the target question. For example, the server can query questions that match the target question from a preset question library based on the target question, and then use the reply information of the questions that match the target question as the reply information of the target question.

[0097] In addition, the server can also obtain the user's historical conversation records, and based on the user's historical conversation records and target questions, obtain various reference reply information from preset question libraries of different types, and then input the target questions and various reference reply information into a preset reply model to generate reply information through the reply model.

[0098] Among them, the method for the server to obtain various reference response information from preset question libraries of different types based on the user's historical conversation records and target questions can be to input the user's historical conversation records and target questions into a preset intent recognition model, so as to obtain various reference response information from preset question libraries of different types through the preset intent recognition model.

[0099] In the above content, different types of question libraries include: a question library pre-built by the server based on historical questions and historical response information contained in historical conversation records of users of the business platform, an open library storing a large number of open questions and their answers, etc.

[0100] In addition, in order to improve the accuracy of the generated reply information, the server can also be equipped with a quality detection module corresponding to each module in the automated problem acceptance system, and the neural network model used by the module and the output result of the module can be detected through the quality detection module corresponding to the module.

[0101] Specifically, for each module in the automated problem acceptance system, the server can construct a detection sample set through the quality detection module corresponding to the module, so as to perform offline detection on the accuracy of the neural network model used by the module through the detection sample set; when it is determined that the accuracy of the neural network model used by the module is lower than the preset threshold, an alarm message can be generated.

[0102] In addition, the server can also detect the output results of the neural network model used by the module through the quality detection module corresponding to the module to obtain the detection results of the output results of the neural network model used by the module, and generate an alarm message when it is determined that the output results of the neural network model used by the module are abnormal according to the above detection results.

[0103] In addition, for each module in the automated problem acceptance system, after obtaining the output result of the neural network model used by the module, the server can also input the output result of the neural network model used by the module into the above-mentioned quality inspection module, so as to obtain the detection result of the output result of the neural network model used by the module through the above-mentioned quality inspection module, and generate an alarm message when it is determined that the output result of the neural network model used by the module is abnormal according to the above-mentioned detection result.

[0104] In the above content, the detection result of the output result of the neural network model used by the module is used to characterize whether the output result of the neural network model used by the module has problems such as: not meeting the preset format requirements, involving unqualified content, and semantically unsmoothness.

[0105] In addition, the server can also perform iterative optimization and adjustment alternately on the acceptance timing judgment module, the starting words acceptance module, the first round guessing module, the question recommendation module, the multi-round question acceptance module and the quality detection modules used to detect the acceptance timing judgment module, the starting words acceptance module, the first round guessing module, the question recommendation module and the multi-round question acceptance module contained in the question acceptance system at specified time intervals, so as to obtain an adjusted acceptance timing judgment module, an adjusted starting words acceptance module, an adjusted first round guessing module, an adjusted question recommendation module, an adjusted multi-round question acceptance module and adjusted quality detection modules.

[0106] In addition, after each module undertakes the user's questions, the server can also collect the question text sent by the user and the reply text generated for the question text sent by the user through the above-mentioned automated question-answering system, and organize and summarize the question text sent by the user and the reply text generated for the question text sent by the user to obtain a corresponding requirement analysis report for the user. Furthermore, the requirement analysis report is pushed to professional insurance agents so that the insurance agents can reply to the question text sent by the user in a targeted manner.

[0107] Among them, when the server determines that the user is not satisfied with the automated reception and requests the service of an insurance agent, it can send the analysis report to the insurance agent so that the insurance agent can timely solve the user's problem according to the sorted customer requirements.

[0108] Of course, when the user still has concerns about purchasing a product and fails to complete an order within a specified time period after consultation, the server can also send the analysis report to the insurance agent so that the insurance agent can formulate a targeted plan based on the user's needs and concerns and actively solve the user's doubts.

[0109] As can be seen from the above content, the server can use the user's historical questions and interaction data to determine the target question nodes, plan the question planning direction chain, and provide highly personalized and targeted alternative questions for the user. Furthermore, it can actively screen and send recommended questions to the user to guide them to clarify the question direction, and at the same time quickly generate accurate and compliant replies based on the user's real-time questions to ensure effective solution of the user's needs. The whole process adopts a dynamic optimization mechanism, adjusts the dialogue strategy according to the user's behavior and feedback, provides a smooth and natural interaction experience, enhances the user's trust and satisfaction, and realizes the intelligence and efficiency of the whole process from question reception.

[0110] Figure 7 It is a schematic structural diagram of a device provided by an exemplary embodiment. Please refer to Figure 7 , at the hardware level, the device includes a processor 702, an internal bus 704, a network interface 706, a memory 708, and a non-volatile memory 710. Of course, it may also include other hardware required for other functions. One or more embodiments of this specification can be implemented in software. For example, the processor 702 reads the corresponding computer program from the non-volatile memory 710 into the memory 708 and then runs it. Of course, in addition to the software implementation method, one or more embodiments of this specification do not exclude other implementation methods, such as logical devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logical unit, and can also be hardware or a logical device.

[0111] Please refer to Figure 8 , the question guiding device for intelligent customer service can be applied to such asFigure 7 The device shown in the figure is used to implement the technical solution of this specification. Among them, the question guiding device for intelligent customer service may include:

[0112] The acquisition module 801 is used to acquire the user's historical questions;

[0113] The determination module 802 is used to determine the node corresponding to the historical problem from the preset user problem transfer graph as the target problem node, and determine at least one path including the target problem node as the problem planning direction chain;

[0114] The question generation module 803 is used to input the historical question, the historical reply information of the historical question and the question planning direction chain into a preset generation model to obtain a first type of candidate questions;

[0115] The recommendation module 804 is used to determine a recommended question based on the first type of candidate questions, and send the recommended question to the user.

[0116] Optionally, the device further includes: a judgment module 805, a first-round generation module 806;

[0117] The judgment module 805 is specifically used to obtain the user's historical conversation record when monitoring the user's online consultation; determine whether the previous conversation with the user is completed according to the historical conversation record, the conversation time of the historical conversation record and the current time;

[0118] The first round generation module 806 is specifically used to send the pre-generated first round dialogue acceptance information to the user to establish a new dialogue with the user when it is determined that the previous dialogue with the user has been completed;

[0119] The acquisition module 801 is specifically used to acquire the user's historical questions from the historical conversation record through the acquisition module when it is determined that the previous conversation with the user is not completed based on the historical conversation record, the conversation time of the historical conversation record and the current time.

[0120] Optionally, the first round of dialogue acceptance information includes: opening words;

[0121] The first-round generation module 806 is specifically used to obtain user characteristics of the user, which include: user attribute characteristics, user behavior characteristics; according to the user characteristics, select a target opening sentence template from various preset opening sentence templates; input the user characteristics and the target opening sentence template into a preset opening sentence generation model to obtain the opening sentence.

[0122] Optionally, the first-round conversation continuation information includes: first-round recommended questions;

[0123] The first-round generation module 806 is specifically configured to obtain user characteristics of the user, where the user characteristics include: user attribute characteristics and user behavior characteristics; determine a basic recommended question from a preset relationship graph of user characteristics and user questions according to the user characteristics; and determine the product involved in the historical conversation record of the user as the target product; query a question that matches the basic recommended question and the target product from a preset question library as the first-round recommended question.

[0124] Optionally, the first-round generation module 806 is specifically configured to query a question that matches the basic recommended question and the target product from a preset question library as each alternative first-round recommended question; select a first-round recommended question from the alternative first-round recommended questions according to the real-time click exposure volume of the user corresponding to each alternative first-round recommended question.

[0125] Optionally, the recommendation module 804 is specifically configured to query each question from a preset question library according to the historical conversation record of the user as a recall question; select a second type of alternative question from the recall questions according to the user characteristics of the user; select at least some alternative questions from the first type of alternative questions and the second type of alternative questions as recommended questions.

[0126] Optionally, the recommendation module 804 is specifically configured to determine a process node that matches the historical conversation record of the user and the user characteristics of the user from a preset product service interaction process chain as the target process node; different process nodes in the product service interaction process chain are used to represent different interaction stages included in the process of interacting with the user; obtain a third type of alternative question according to the target process node and a preset mapping graph between process nodes and questions; select at least some alternative questions from the first type of alternative questions and the third type of alternative questions as recommended questions.

[0127] Optionally, the device further includes: a reply module 807;

[0128] The reply module 807 is specifically configured to obtain the target question sent by the user according to the recommended question and the historical conversation record of the user; obtain each reference reply information from a preset question library of different types according to the historical conversation record and the target question; input the target question and the reference reply information into a preset reply model to generate a reply information through the reply model and reply to the user.

[0129] Based on the same concept as the above method, this specification also provides an electronic device, including: a processor; a memory for storing executable instructions that can be executed by the processor; wherein, the processor realizes the steps of the method as described in any of the above embodiments by running the executable instructions.

[0130] Based on the same concept as the above method, this specification also provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method as described in any of the above embodiments are realized.

[0131] Based on the same concept as the above method, this specification also provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method as described in any of the above embodiments are realized.

Claims

1. A question guiding method for intelligent customer service, comprising: Get the user's historical questions; Determine the node corresponding to the historical problem from the preset user problem transfer graph as the target problem node, and determine at least one path including the target problem node as the problem planning direction chain; Input the historical questions, the historical response information of the historical questions and the problem planning direction chain into a preset generation model to obtain a first type of candidate questions; Determine recommended questions based on the first category of candidate questions, and send the recommended questions to the user.

2. The method according to claim 1, before obtaining the user's historical questions, the method further comprises: When monitoring a user's online consultation, obtain the user's historical conversation records; When it is determined that the last conversation with the user has ended according to the historical conversation record, the conversation time of the historical conversation record and the current time, the pre-generated first-round conversation acceptance information is sent to the user to restart the conversation with the user; Get the user's historical questions, including: When it is determined that the last conversation with the user is not completed according to the historical conversation record, the conversation time of the historical conversation record and the current time, the historical questions of the user are obtained from the historical conversation record.

3. The method according to claim 2, wherein the first round of dialogue acceptance information comprises: Opening words; Generate the first round of dialogue acceptance information, including: Acquire user characteristics of the user, wherein the user characteristics include: user attribute characteristics and user behavior characteristics; According to the user characteristics, a target opening phrase template is selected from various preset opening phrase templates; The user characteristics and the target opening phrase template are input into a preset opening phrase generation model to obtain an opening phrase.

4. The method according to claim 2, wherein the first round of dialogue acceptance information comprises: First round of recommendation questions; Generate the first round of dialogue acceptance information, including: Acquire user characteristics of the user, wherein the user characteristics include: user attribute characteristics and user behavior characteristics; According to the user characteristics, determining basic recommendation questions from a preset user characteristics and user question relationship diagram; and determining products involved in the user's historical conversation records as target products; Questions matching the basic recommendation questions and the target product are obtained from a preset question library as the first round of recommendation questions.

5. The method according to claim 4, querying from a preset question library to obtain questions matching the basic recommendation questions and the target product as the first round of recommendation questions, specifically comprising: Querying a preset question library to obtain questions that match the basic recommendation questions and the target product as candidate first-round recommendation questions; A first-round recommended question is selected from the candidate first-round recommended questions according to the user real-time click exposure corresponding to each candidate first-round recommended question.

6. The method of claim 1, further comprising: According to the historical conversation records of the user, various questions are obtained from a preset question library as recall questions; Selecting a second type of candidate questions from the recalled questions according to the user characteristics of the user; At least some of the candidate questions are selected from the first category of candidate questions and the second category of candidate questions as recommended questions.

7. The method of claim 1, further comprising: Determine, from the preset product business interaction process chain, a process node that matches the historical conversation record of the user and the user characteristics of the user as the target process node; Different process nodes in the product-business interaction process chain are used to represent different interaction stages included in the process of interacting with the user; According to the target process node and the mapping diagram between the preset process node and the question, a third type of candidate question is obtained; At least some of the candidate questions are selected from the first category of candidate questions and the third category of candidate questions as recommended questions.

8. The method of claim 1, further comprising: Obtaining a target question sent by the user according to the recommended question and a historical conversation record of the user; According to the historical conversation records and the target question, obtaining reference response information from preset question libraries of different types; The target question and each reference reply information are input into a preset reply model, so as to generate reply information through the reply model and reply to the user.

9. A question guiding device for intelligent customer service, comprising: The acquisition module is used to obtain the user's historical questions; A determination module, used to determine the node corresponding to the historical problem from a preset user problem transfer graph as a target problem node, and determine at least one path including the target problem node as a problem planning direction chain; A question generation module, used for inputting the historical question, the historical reply information of the historical question and the question planning direction chain into a preset generation model to obtain a first type of candidate question; A recommendation module is used to determine a recommended question based on the first category of candidate questions, and send the recommended question to the user.

10. The device according to claim 9, wherein the device further comprises: Judgment module, first round generation module; The judgment module is specifically used to obtain the user's historical conversation record when monitoring the user's online consultation; Determining whether the last conversation with the user is completed according to the historical conversation record, the conversation time of the historical conversation record and the current time; The first round generation module is specifically used to send the pre-generated first round dialogue acceptance information to the user to establish a new dialogue with the user when it is determined that the previous dialogue with the user has been completed; The acquisition module is specifically used to acquire the user's historical questions from the historical conversation record through the acquisition module when it is determined that the previous conversation with the user is not completed based on the historical conversation record, the conversation time of the historical conversation record and the current time.

11. The device according to claim 10, wherein the first-round dialogue acceptance information comprises: First round of recommendation questions; The first-round generation module is specifically configured to obtain user features of a user, where the user features include: user attribute features and user behavior features; determine a basic recommended question from a preset relationship graph of user features and user questions according to the user features; and determine a product involved in the historical conversation record of the user as a target product; query a question that matches the basic recommended question and the target product from a preset question library as the first-round recommended question.

12. An electronic device, comprising: Processor; A memory for storing executable instructions of the processor; wherein, the processor realizes the steps of the method according to any one of claims 1-8 by running the executable instructions.

13. A computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method according to any one of claims 1-8 are realized.

14. A computer program product, including computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the method according to any one of claims 1-8 are realized.

Images