Intelligent customer service application construction method and application method based on workflow

Abstract

The invention provides a workflow-based intelligent customer service application construction method and a workflow-based intelligent customer service application use method. The intelligent customer service application is constructed by adopting a low-code platform, and the construction method comprises the following steps: constructing a knowledge base according to an application scene; inputting historical interaction data, deeply understanding a problem input by a user by using a neural network model, and forming a comprehensive feature vector of the problem; compiling knowledge fragments of the knowledge base by using a deep learning model to form knowledge vectors of the knowledge fragments; by learning a matching calling process of the comprehensive feature vector and the knowledge vector in the historical interaction data, forming an answering strategy in a question receiving mode; daily use habits of the user are learned, and active help triggering conditions in a behavior monitoring mode are formed. According to the invention, the technical requirements of intelligent customer service arrangement are reduced, and the individuation of intelligent customer service arrangement is improved; problems are answered for users in an active mode and a passive mode, and the use efficiency and satisfaction degree of intelligent customer service are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent agent orchestration, and specifically relates to a method for constructing and using an intelligent customer service application based on a workflow. Background Art

[0002] Intelligent agents, such as intelligent customer service (AI customer service), are designed to solve pain points in workflows, especially to replace human customer service for 24-hour duty, thereby improving work efficiency and quality. Compared with human customer service, AI customer service can answer customer questions in a faster, more efficient and non-subjective manner, reducing labor and enhancing work efficiency and customer satisfaction. However, current AI customer service still has problems such as limited knowledge base, understanding deviation, lack of humanized communication, insufficient emotional understanding, and limited ability to handle complex problems.

[0003] In addition, in intelligent agent orchestration, traditional technologies usually require developers to have a certain level of code orchestration skills to integrate intelligent agents into existing workflows by writing code. For users who do not understand code, it is often impossible to develop intelligent agents suitable for their own workflows according to personalized needs, which hinders the wide application of intelligent agents.

[0004] In summary, the following problems at least exist in the prior art:

[0005] 1. The construction of intelligent customer service requires a large amount of code orchestration by professionals, with high construction difficulty, and the constructed intelligent agents are difficult to meet personalized needs;

[0006] 2. The ability of intelligent customer service to handle complex problems is limited, and the usage efficiency is low. Summary of the Invention

[0007] The purpose of the present invention is to overcome the defects of the prior art, and provides a method for constructing and using an intelligent customer service application based on a workflow, which can solve the problems of difficult construction of intelligent customer service, insufficient personalization, and limited ability to handle complex problems.

[0008] To achieve the above and other purposes, the present invention is realized by including the following technical solutions: As a first aspect, the present invention proposes a method for constructing an intelligent customer service application based on a workflow, the intelligent customer service application is constructed using a low-code platform, and the construction method includes the steps of:

[0009] Construct a knowledge base according to the application scenario;

[0010] Input historical interaction data, and use a neural network model to deeply understand the problem input by the user to form a comprehensive feature vector of the problem;

[0011] Use a deep learning model to compile the knowledge fragments of the knowledge base to form knowledge vectors of the knowledge fragments;

[0012] By learning the matching and calling process between the comprehensive feature vector and the knowledge vector in the historical interaction data, a solution strategy in the question receiving mode is formed.

[0013] Learn the user's daily usage habits to form the active help trigger conditions in the behavior monitoring mode.

[0014] Furthermore, the knowledge base is a real-time updated knowledge base, including the user's internal database and public knowledge information; the knowledge base is divided into multiple first-level knowledge bases according to the field, and each of the first-level knowledge bases is further divided into multiple second-level knowledge bases.

[0015] Furthermore, the question includes text data and / or picture data; a long short-term memory network or a gated recurrent unit is used to process the text data to extract a text feature vector; a convolutional neural network is used to process the image data to extract an image feature vector; the text feature vector and the image feature vector are spliced to form the comprehensive feature vector.

[0016] Furthermore, the solution strategy includes the selection of the solution idea for the question and the selection of the retrieval mode in the knowledge base.

[0017] Furthermore, the retrieval modes include hybrid retrieval, full-text retrieval, and semantic retrieval.

[0018] As a second aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the construction method described in the first aspect is implemented.

[0019] As a third aspect, the present invention provides a method for using an intelligent customer service application based on a workflow, and the intelligent customer service application includes a question receiving mode and a behavior monitoring mode;

[0020] In the question receiving mode, the usage method includes the steps of:

[0021] Input a question in the intelligent customer service application;

[0022] Capture the key information and time dependence relationship in the question to form a comprehensive feature vector;

[0023] Call the knowledge fragments related to the comprehensive feature vector in the knowledge base and integrate them to form an answer;

[0024] In the behavior monitoring mode, the usage method includes the steps of:

[0025] Monitor the user's operation behavior;

[0026] When the intelligent customer service application recognizes that the user operation behavior meets the active assistance trigger condition, it actively asks if help is needed and presents the predicted questions and solutions.

[0027] Further, the step of "invoking the knowledge fragments related to the comprehensive feature vector in the knowledge base and integrating them into an answer" specifically includes the steps of:

[0028] Matching the comprehensive feature vector of the question with the knowledge vectors in the knowledge base to find the knowledge vectors related to the comprehensive feature vector;

[0029] Sorting the knowledge fragments corresponding to the knowledge vectors according to the relevance, and selecting the most relevant several knowledge fragments as candidate answers;

[0030] Extracting key information from the candidate answers, and the intelligent customer service application integrates, supplements and polishes the answers according to its own experience.

[0031] Further, in the question receiving mode, a loop count threshold X for the question is set; it is judged whether the loop count of the question is less than X; if so, the intelligent customer service application re-invokes the knowledge fragments for integrated output; if not, the manual customer service is automatically called to answer the question.

[0032] As a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in any item of the third aspect is implemented.

[0033] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0034] 1. The present invention uses a low-code platform and performs intelligent customer service orchestration based on a workflow, reducing the technical requirements for intelligent customer service orchestration and improving the personalization of intelligent customer service orchestration;

[0035] 2. The present invention uses a neural network model and a deep learning model to understand questions and knowledge fragments respectively, improving the ability to handle complex questions;

[0036] 3. The present invention constructs an intelligent customer service using two modes of question reception and behavior monitoring, and answers questions for users in both active and passive ways, improving the usage efficiency and satisfaction of the intelligent customer service;

[0037] 4. The present invention constructs a detailed and elaborate knowledge base, enabling the intelligent customer service application to cover a wide range of information in a specific field and solving the problem of limited knowledge base existing in the prior art for intelligent customer service. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1It shows a schematic flowchart of a method for constructing an intelligent customer service application based on a workflow according to the present invention.

[0039] Figure 2 It shows a schematic flowchart of a method for using an intelligent customer service application based on a workflow according to the present invention. Specific embodiments

[0040] In the embodiments of the present invention, by providing a method for constructing and using an intelligent customer service application based on a workflow, the problems in the prior art such as difficult construction of intelligent customer service, insufficient personalization, and limited ability to handle complex problems are solved.

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited by the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention. Additionally, it should be noted that for the sake of description, only the parts related to the present invention are shown in the accompanying drawings rather than all of them.

[0042] The general idea of the technical solution provided by the present invention is as follows: A method for constructing an intelligent customer service application based on a workflow, the intelligent customer service application is constructed using a low-code platform, and the construction method includes the steps:

[0043] Construct a knowledge base according to the application scenario;

[0044] Input historical interaction data, use a neural network model to deeply understand the problem input by the user, and form a comprehensive feature vector of the problem;

[0045] Use a deep learning model to compile the knowledge fragments of the knowledge base to form knowledge vectors of the knowledge fragments;

[0046] By learning the matching call process between the comprehensive feature vector and the knowledge vector in the historical interaction data, form an answering strategy in the problem receiving mode;

[0047] Learn the daily usage habits of users to form active help trigger conditions in the behavior monitoring mode.

[0048] Its main concept is as follows: in the mode of workflow, it realizes the logical orchestration of low-code intelligent customer service, reduces the technical requirements for intelligent customer service orchestration, and improves the personalization of intelligent customer service orchestration; it uses neural network models and deep learning models to understand questions and knowledge fragments respectively, improving the ability to handle complex questions; at the same time, the intelligent customer service adopts two modes of question reception and behavior monitoring, and answers questions for users in both active and passive ways, improving the usage efficiency and satisfaction of the intelligent customer service.

[0049] After introducing the basic principle of the present invention, the various non-limiting implementation manners of the present invention will be specifically introduced below in conjunction with the accompanying drawings of the specification.

[0050] Embodiment 1:

[0051] Please refer to the appendix Figure 1 , this embodiment provides a method for constructing an intelligent customer service application based on a workflow. For the application scenario, an intelligent customer service application is preset, and then through the API call of the intelligent customer service application, the preset intelligent customer service application is integrated into the workflow, and a low-code platform is used to create a customized workflow by configuring the intelligent customer service application. The constructed intelligent customer service application can automatically identify common problems of users according to the preset knowledge base and natural language processing technology, and provide accurate and timely responses.

[0052] Specifically, the construction method includes the steps:

[0053] S1. Construct a knowledge base according to the application scenario;

[0054] Among them, the knowledge base is a detailed and exhaustive knowledge base that can cover a wide range of information in a specific field, specifically including user internal databases such as customer information, employee information, training materials, and financial data, as well as public knowledge information such as existing technologies, academic information, and common knowledge. Taking the intelligent customer service of cloud rendering as an example, its knowledge base can first be divided into multiple first-level knowledge bases according to fields such as software installation, software operation, software after-sales, and task processing; then the first-level knowledge bases are further divided. Specifically, software installation can be divided into second-level knowledge bases for different systems, software operation can be divided into second-level knowledge bases for software tutorials, shortcut keys, plug-in usage, common problems, etc., software after-sales can be divided into second-level knowledge bases for problems such as recharge and invoicing, and task processing can be divided into fault handling, resource handling, etc. for second-level knowledge bases.

[0055] S2. Input historical interaction data, and use a neural network model to deeply understand the questions input by users to form a comprehensive feature vector of the questions;

[0056] Among them, when training the intelligent customer service application, the input historical interaction data can be first cleaned to remove duplicate data, invalid data, etc.

[0057] During the interaction process, the questions input by the user can be presented in text form, in picture form, or in the form of pictures combined with text, that is, the questions include text data and / or picture data. When performing algorithmic processing on the questions input by the user in the historical interaction data, neural network models such as long short-term memory network (LSTM) or gated recurrent unit (GRU) can be used to process the text data, capture the key information of the text and the time-dependent relationship, etc., and form a text feature vector. The key information may include the theme of the question (such as "how to install software", "product troubleshooting", etc.), the specific content of the question (such as specific error codes, problem descriptions encountered, etc.), the user's emotion (such as "I'm very anxious", "this is so annoying", etc.), time information (such as "this morning", "yesterday afternoon", etc.), and this information can help understand the urgency or background of the question. Convolutional neural network is used to process the image data, and image feature vectors are extracted through structures such as convolutional layers, pooling layers, and fully connected layers; finally, the text feature vector and the image feature vector are concatenated to form a comprehensive feature vector. For example, if the picture of the question prompts a rendering task error and the error code, and the text describes the scenario where the problem occurs, then both the background where the problem occurs and the error code are recorded.

[0058] S3. Use a deep learning model to compile the knowledge fragments in the knowledge base to form knowledge vectors of the knowledge fragments;

[0059] Among them, the deep learning model can adopt a transformer model, etc. The knowledge vectors can be matched with the comprehensive feature vectors to improve the ability to call the knowledge fragments in the knowledge base according to the questions, and improve the ability and efficiency of screening question-related information.

[0060] S4. By learning the matching and calling process between the comprehensive feature vectors and the knowledge vectors in the historical interaction data, a solution strategy in the question receiving mode is formed;

[0061] Among them, by deeply understanding the questions input by the user, learning the solution ideas, answer frameworks, and retrieval modes in the knowledge base for the questions in the historical interaction data, a solution strategy for automatically retrieving knowledge fragments and integrating answers oriented to the questions is formed, specifically including forming a selection rule for solution ideas for different questions and a selection rule for retrieval modes adopted in the knowledge base.

[0062] When matching the comprehensive feature vector with the knowledge vector, the appropriate retrieval mode can be selected based on the keywords and semantics of the question and the characteristics of the knowledge base. Common retrieval modes include hybrid retrieval, full-text retrieval, and semantic retrieval. Specifically, for a knowledge base with a high degree of structure, such as a TV station application scenario, the question is: Find reports on xx published in the xx column on xx day. It is necessary to consider multiple factors comprehensively, and full-text retrieval can be used. For user questions that require more than just keyword matching, but also understanding of context and intent. For example, if a user asks "how to solve the problem of a computer not starting up", semantic retrieval can be used.

[0063] In addition, the threshold of the number of question cycles can be set to x (the value of x can be different depending on the questioning scenario, such as 2 to 3 times or 4 to 5 times, and the intelligent customer service application can make the specific judgment based on the answering process). The questions do not have to be exactly the same, but can be similar or related questions; when the number of cycles is less than x, the AI ​​agent re-calls the knowledge base knowledge fragments for integrated output; when the number of cycles is greater than or equal to x, manual customer service is automatically called out to answer the question.

[0064] S5. Learn the daily usage habits of users and form the trigger conditions for active help in the behavior monitoring mode. That is, by observing user behavior and habits, identify possible problems in advance and actively provide effective suggestions or solutions.

[0065] Specifically, by learning the daily usage habits of users, we can form active help triggering conditions, such as operation behavior timeout, the number of times the setting tool is not used after selection is much higher than the number of times it is used in a habit, and the time spent in selecting the renderer is long, etc.; we monitor the user's real-time operation behavior, and when the user's operation behavior is abnormal and meets the active help triggering conditions, we will actively ask whether help is needed and make predictions about possible problems. For example, if the average time for a certain step of the user's daily operation is t, when the operation time is greater than t when it is used again, we will actively prompt the user whether the problem needs to be solved, and list several possible problems predicted after learning the user's behavior habits.

[0066] In addition, the constructed intelligent customer service application can also improve its ability to solve problems by itself through learning and experience accumulation. Intelligent customer service applications need to accurately understand customer needs and provide personalized responses, while improving their own understanding and problem-solving abilities through iterative learning. Intelligent customer service applications need to continuously update their knowledge base, enhance their ability to handle special and common problems, and ensure that the generated response strategies are accurate and effective. In addition, intelligent customer service applications should be able to intelligently guide questions, improve communication efficiency with customers, avoid ineffective communication, and thus improve customer satisfaction.

[0067] Embodiment 2:

[0068] This embodiment provides a computer-readable storage medium with a computer program stored thereon. When the computer program is executed by a processor, the method of Embodiment 1 is implemented.

[0069] Embodiment 3:

[0070] As Figure 2 shown, based on the same inventive concept as in the foregoing Embodiment 1, this embodiment provides a method for using an intelligent customer service application based on a workflow. The intelligent customer service application includes a problem receiving mode and a behavior monitoring mode;

[0071] In the problem receiving mode, the method of use includes the steps of:

[0072] Input a problem into the intelligent customer service application;

[0073] Capture the key information and time dependence relationship in the problem to form a comprehensive feature vector;

[0074] Call knowledge fragments related to the comprehensive feature vector in the knowledge base and integrate them into an answer; specifically, it includes the steps of: matching the comprehensive feature vector of the problem with the knowledge vectors in the knowledge base to find the knowledge vectors related to the comprehensive feature vector; sorting the knowledge fragments corresponding to the knowledge vectors according to the relevance, and selecting the most relevant several knowledge fragments as candidate answers; extracting key information from the candidate answers, and the intelligent customer service application learns according to its own knowledge and experience, integrates them into an answer, and polishes, supplements and adjusts the answer to make it more in line with the user's needs.

[0075] Further, in the problem receiving mode, set the loop count threshold X of the problem; determine whether the loop count of the problem is less than X; if so, the intelligent customer service application re-calls the knowledge fragments for integrated output; if not, automatically call a human customer service to answer the problem.

[0076] In the behavior monitoring mode, the method of use includes the steps of:

[0077] Monitor the user's operation behavior;

[0078] When the intelligent customer service application recognizes that the user's operation behavior meets the active help trigger condition, actively ask if help is needed, and propose predicted questions and solutions.

[0079] For example, if the average time for a user to operate a certain step in daily life is t, when the operation time is greater than t when using it again, actively prompt the user whether they need to solve the problem, and list several possible problems predicted by learning the user's behavior habits.

[0080] Embodiment 4:

[0081] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method of Embodiment 3 is implemented.

[0082] In summary, in actual interactions, whether facing pure text questions or complex queries combining text and images, the intelligent customer service application provided by the present invention can extract key features through advanced algorithms, quickly call relevant knowledge, and generate accurate answers that meet the user's needs; it solves the limitations of human customer service in dealing with complex and diverse user queries, improves the accuracy and efficiency of question answering, and at the same time reduces the dependence on humans, providing users with a more intelligent and personalized service experience.

[0083] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the same technology of the present invention, the present invention also intends to include these changes and modifications.

Claims

1. A method for constructing an intelligent customer service application based on workflow, characterized in that: The intelligent customer service application is built using a low-code platform, and the construction method includes the following steps: Build a knowledge base based on application scenarios; Input historical interaction data, use the neural network model to deeply understand the questions input by users, and form a comprehensive feature vector of the questions; Compiling the knowledge fragments of the knowledge base using a deep learning model to form knowledge vectors of the knowledge fragments; Forming a solution strategy in a question receiving mode by learning the matching and calling process of the comprehensive feature vector and the knowledge vector in the historical interaction data; Learn users' daily usage habits and form active help trigger conditions under behavior monitoring mode.

2. The method for constructing a workflow-based intelligent customer service application according to claim 1, characterized in that: The knowledge base is a real-time updated knowledge base, including a user internal database and public knowledge information; the knowledge base is divided into a plurality of first-level knowledge bases according to fields, and each of the first-level knowledge bases is subdivided into a plurality of second-level knowledge bases.

3. The method for constructing a workflow-based intelligent customer service application according to claim 1, characterized in that: The problem includes text data and / or image data; the text data is processed by a long short-term memory network or a gated recurrent unit to extract a text feature vector; the image data is processed by a convolutional neural network to extract an image feature vector; and the text feature vector and the image feature vector are concatenated to form the comprehensive feature vector.

4. The method for constructing a workflow-based intelligent customer service application according to claim 1, characterized in that: The solution strategy includes the selection of a solution idea for the problem and the selection of a search mode in the knowledge base.

5. The method for constructing a workflow-based intelligent customer service application according to claim 4, characterized in that: The retrieval modes include hybrid retrieval, full-text retrieval and semantic retrieval.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the construction method according to any one of claims 1 to 5 is implemented.

7. A method for using a workflow-based intelligent customer service application, characterized in that: The intelligent customer service application includes a question receiving mode and a behavior monitoring mode; In the question receiving mode, the method of use comprises the steps of: Entering a question in the intelligent customer service application; Capture the key information and time dependency in the problem to form a comprehensive feature vector; Retrieving knowledge fragments related to the comprehensive feature vector in the knowledge base and integrating them to form an answer; In the behavior monitoring mode, the method of use includes the steps of: Monitor user operation behavior; When the intelligent customer service application recognizes that the user's operation behavior meets the active help triggering condition, it actively asks whether help is needed and proposes predicted problems and solutions.

8. The method for using a workflow-based intelligent customer service application according to claim 8, characterized in that: The "calling the knowledge fragments related to the comprehensive feature vector in the knowledge base and integrating them to form an answer" specifically includes the following steps: Matching the comprehensive feature vector of the problem with the knowledge vector of the knowledge base to find the knowledge vector related to the comprehensive feature vector; Sort the knowledge fragments corresponding to the knowledge vectors according to their relevance, and select the most relevant knowledge fragments as candidate answers; Key information is extracted from the candidate answers, and the intelligent customer service application integrates, supplements and polishes the answers based on its own experience.

9. The method for using a workflow-based intelligent customer service application according to claim 1, characterized in that: In the question receiving mode, a threshold value X for the number of cycles of the question is set; it is determined whether the number of cycles of the question is less than X; if so, the intelligent customer service application calls the knowledge fragments again for integrated output; if not, a manual customer service is automatically called out to answer the question.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method of use according to any one of claims 8 to 9 is implemented.

Images