Insurance operation process full-link analysis method and device, equipment and medium

Abstract

The invention relates to the field of intelligent decision, and discloses an insurance operation process full-link analysis method and device, electronic equipment and a storage medium, and the method comprises the steps: defining a standard operation process of an agent; training a pre-trained hidden Markov model by using the labeled text to obtain a trained hidden Markov model; training the pre-trained long-short-term memory model through the hidden state to obtain a trained long-short-term memory model; inputting the small-stage distribution into a pre-trained language representation model so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model; analyzing a seat operation process corresponding to the current interaction record through the trained hidden Markov model, the trained long and short term memory model and the trained language representation model; and updating the question and answer knowledge base through the seat operation process to obtain an updated knowledge base. According to the invention, the insurance operation process of the agent to the user can be finely managed.

Description

Technical Field

[0001] The present invention relates to the fields of intelligent decision-making and financial technology, and in particular to a full-link analysis method, device, electronic device and computer-readable storage medium for an insurance operation process. Background Art

[0002] As a highly competitive field, the insurance industry has significant complexity and professional characteristics in its products, which requires agents to have certain professional knowledge and skills. At the same time, the purchase decision of insurance products usually requires users to think carefully, so agents need to communicate and exchange with users in depth, understand their needs and concerns, and provide targeted products and services.

[0003] In order to improve the efficiency and effectiveness of agent operations, insurance companies need to conduct refined management of operating processes and analysis of user demands. Refined management can better organize and manage operating processes and improve work efficiency. It can also better analyze and optimize operating processes and improve agent operation results. Analysis of user demands can help agents better understand user needs and expectations, provide products and services that better meet user needs, and improve user satisfaction and purchase intention. At the same time, through the analysis of user demands, operators can also better understand market trends and competition, and provide support for the company to formulate more effective application strategies and plans. Therefore, refined management and analysis of user demands are important means to improve the efficiency and effectiveness of agent operations in the insurance field, and are also one of the key factors for insurance companies to enhance their competitiveness.

[0004] In the prior art, there is a lack of an effective method to finely manage the insurance operation process of agents for users. Summary of the invention

[0005] The present invention provides a method, device, electronic device and computer-readable storage medium for analyzing the entire link of an insurance operation process, the main purpose of which is to finely manage the insurance operation process of agents for users.

[0006] To achieve the above-mentioned purpose, the present invention provides a full-link analysis method for insurance operation process, comprising:

[0007] Collecting historical interaction records between agents and historical users, converting the historical interaction records into historical interaction texts, and performing text preprocessing on the historical interaction texts to obtain preprocessed texts;

[0008] Defining a standard operating procedure for the agent, annotating the preprocessed text using the standard operating procedure to obtain annotated text, and constructing a question-and-answer knowledge base between the agent and the historical user using the preprocessed text;

[0009] Constructing a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user;

[0010] Using the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text;

[0011] Training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small-stage distribution of the trained long short-term memory model with respect to the hidden state;

[0012] Inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model;

[0013] Obtaining a current interaction record between the agent and the current user, and analyzing the agent operation process corresponding to the current interaction record based on the question-answer knowledge base by using the trained hidden Markov model, the trained long short-term memory model, and the trained language representation model;

[0014] The question-and-answer knowledge base is updated through the agent operation process to obtain an updated knowledge base, and the agent operation process and the updated knowledge base are used to determine the full-link analysis result of the operation process between the agent and the current user.

[0015] In order to solve the above problems, the present invention also provides a device for analyzing the entire chain of insurance operation processes, the device comprising:

[0016] A text preprocessing module is used to collect historical interaction records between agents and historical users, convert the historical interaction records into historical interaction texts, and perform text preprocessing on the historical interaction texts to obtain preprocessed texts;

[0017] A knowledge base construction module, used to define a standard operating procedure for the agent, annotate the preprocessed text using the standard operating procedure to obtain annotated text, and construct a question-and-answer knowledge base between the agent and the historical user using the preprocessed text;

[0018] A model building module, used to build a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user;

[0019] A state acquisition module, used for training the pre-trained hidden Markov model using the annotated text to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text;

[0020] A stage acquisition module, used for training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small stage distribution of the trained long short-term memory model with respect to the hidden state;

[0021] A language training module, used for inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model;

[0022] A process analysis module, used to obtain the current interaction record between the agent and the current user, and analyze the agent operation process corresponding to the current interaction record based on the question-answer knowledge base through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model;

[0023] The full-link analysis module is used to update the question-and-answer knowledge base through the agent operation process to obtain an updated knowledge base, and use the agent operation process and the updated knowledge base to determine the full-link analysis result of the operation process between the agent and the current user.

[0024] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0025] at least one processor; and,

[0026] a memory communicatively connected to the at least one processor; wherein,

[0027] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to implement the above-mentioned full-link analysis method of the insurance operation process.

[0028] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned insurance operation process full-link analysis method.

[0029] It can be seen that the embodiment of the present invention defines the standard operating procedures of the agent to divide the agent operating procedures into macroscopic stages and specific small stages from the dual dimensions of online and offline, and deeply analyzes the user's demands for each small stage. Furthermore, the embodiment of the present invention uses the pre-processed text to build a question-and-answer knowledge base between the agent and the historical user to provide users with seamless and efficient problem solving. This strategy strengthens the control over the agent's operating procedures and improves the user experience. The embodiment of the present invention uses the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model for efficiently identifying the large stages of the operating procedures and providing a macro perspective. The embodiment of the present invention trains the pre-trained long short-term memory model through the hidden state to use LSTM to go into details. , accurately capture the small stage changes in the operation process, and complement each other with the aforementioned hidden Markov model to jointly build a more complete and accurate agent operation process stage identification system. The embodiment of the present invention trains the pre-trained language representation model through the small stage distribution to obtain a trained language representation model, which is used to utilize the bidirectional encoding mechanism of BERT and the sequence modeling ability of CRF, so that the combined model can effectively process long texts and complex sentence structures. Even in the face of user demand expressions that have never appeared, it can still be recognized based on the existing learning model and has a strong generalization ability. The embodiment of the present invention updates the question-and-answer knowledge base through the agent operation process to regularly review the content of the knowledge base to ensure the accuracy and timeliness of the information. At the same time, a feedback mechanism is established to collect feedback from users and agent teams on the use of the knowledge base and continuously optimize the content. Therefore, the full-link analysis method, device, electronic device and computer-readable storage medium of the insurance operation process proposed in the embodiment of the present invention can finely manage the insurance operation process of the agent to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of an application environment of a full-link analysis method for an insurance operation process provided by an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of a process flow of a full-link analysis method for an insurance operation process provided by an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of a module of a device for analyzing the entire link of an insurance operation process provided by an embodiment of the present invention;

[0033] Figure 4 A schematic diagram of the internal structure of an electronic device for implementing a full-link analysis method for an insurance operation process provided by an embodiment of the present invention;

[0034] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] The embodiment of the present invention provides a full-link analysis method for insurance operation process, which can be applied in Figure 1 In an application environment, wherein the client communicates with the server through a network, the server can collect historical interaction records between the agent and historical users through the client, convert the historical interaction records into historical interaction texts, perform text preprocessing on the historical interaction texts to obtain preprocessed texts; define the standard operating procedures of the agent, annotate the preprocessed texts with the standard operating procedures to obtain annotated texts, and use the preprocessed texts to build a question-and-answer knowledge base between the agent and the historical users; build a pretrained hidden Markov model, a pretrained long-short term memory model, and a pretrained language representation model between the agent and the historical users; train the pretrained hidden Markov model with the annotated text to obtain a trained hidden Markov model, and obtain the hidden state of the trained hidden Markov model with respect to the annotated text; The pre-trained long short-term memory model is trained through the hidden state to obtain a trained long short-term memory model, and the small-stage distribution of the trained long short-term memory model with respect to the hidden state is obtained; the small-stage distribution is input into the pre-trained language representation model to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model; the current interaction record between the agent and the current user is obtained, and based on the question-and-answer knowledge base, the agent operation process corresponding to the current interaction record is analyzed through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model; the question-and-answer knowledge base is updated through the agent operation process to obtain an updated knowledge base, and the full-link analysis result of the operation process between the agent and the current user is determined by using the agent operation process and the updated knowledge base.

[0037] Reference Figure 2 FIG. 1 is a flow chart of a method for analyzing the entire insurance operation process according to an embodiment of the present invention. In an embodiment of the present invention, the method for analyzing the entire insurance operation process according to an embodiment of the present invention comprises the following steps S1-S5:

[0038] S1. Collect historical interaction records between agents and historical users, convert the historical interaction records into historical interaction texts, and perform text preprocessing on the historical interaction texts to obtain preprocessed texts.

[0039] In an embodiment of the present invention, the historical interaction records refer to historical call records between humans and AI robots, online interaction records, etc. (voice, text data) in the insurance industry.

[0040] Optionally, the process of converting the historical interaction records into historical interaction texts may use ASR (automatic speech recognition) to convert the insurance voice call records into texts, while the historical interaction records that are originally texts do not need to be converted.

[0041] Optionally, the process of performing text preprocessing on the historical interactive text includes removing stop words and special characters, correcting spelling errors, restoring stems, etc.

[0042] S2. Define a standard operating procedure for the agent, use the standard operating procedure to annotate the preprocessed text to obtain annotated text, and use the preprocessed text to construct a question-and-answer knowledge base between the agent and the historical user.

[0043] The embodiment of the present invention defines the standard operating procedures of the agents, so as to divide the agent operating procedures into macroscopic stages and specific sub-stages from the dual dimensions of online and offline, and deeply analyze the user's demands for each sub-stage.

[0044] The standard operating procedure refers to the response that the agent needs to make based on the user's inquiry or text after the agent has a conversation with the user.

[0045] In one embodiment of the present invention, the definition of the standard operating procedures of the agent includes: dividing the operating procedure stages of the agent; wherein the operating procedure stages include an opening remarks stage, a demand exploration stage, a product introduction stage, an objection handling stage, a result facilitation stage, and a closing remarks stage; subdividing the operating procedure stages to obtain subdivided procedure stages; and using the operating procedure stages and the subdivided procedure stages as the standard operating procedures of the agent.

[0046] Exemplarily, the dividing of the agent's work process stages and the subdividing of the work process stages to obtain the subdivided process stages include: defining a standard agent work process stage and subdivided small stages, such as opening remarks (agent self-introduction, confirming user identity), demand exploration (asking users about their specific needs and questions), product introduction (product recommendation, detailed introduction to the features and advantages of products or services), objection handling (answering users' concerns, membership rules consultation, claims process consultation, discount consultation), transaction facilitation (making purchase suggestions, quotations, and attempting to close deals), closing remarks (thanking users for their time, providing follow-up contact information, and making appointments for services), etc.

[0047] Optionally, the process of marking the pre-processed text using the standard operating procedure refers to marking the texts in the pre-processed text that belong to the same major stage or minor stage based on the standard operating procedure. For example, the pre-processed text between the agent and the user includes 1 to 5, of which 1 to 2 belong to the major stage of opening remarks, 1 belongs to the minor stage of agent self-introduction, and 2 belongs to the minor stage of confirming the user's identity.

[0048] Furthermore, the embodiment of the present invention constructs a question-and-answer knowledge base between the agent and the historical user by utilizing the preprocessed text, so as to provide users with seamless and efficient question answering. This strategy strengthens the control over the agent's operating process and improves the user experience.

[0049] Among them, the question and answer knowledge base is mainly composed of keywords in the preprocessed text, user intentions and agent responses corresponding to user intentions (agent responses refer to the insurance operation processes made by agents, such as sending product information, arranging follow-up words, etc.).

[0050] S3. Constructing a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user.

[0051] In an embodiment of the present invention, the pre-trained hidden Markov model refers to a model composed of an untrained hidden Markov model, the pre-trained long short-term memory model refers to a model composed of an untrained multi-layer stacked LSTM network, and the pre-trained language representation model is composed of an untrained BERT model, a CRF (conditional random field) layer and an intent recognition module, wherein the CRF (conditional random field) layer and the intent recognition module serve as two output ends of the BERT model, and the intent recognition module is composed of a fully connected layer and a softmax layer.

[0052] S4. Use the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, and obtain the hidden state of the trained hidden Markov model with respect to the annotated text.

[0053] The embodiment of the present invention uses the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, which is used to efficiently identify large stages of the work process and provide a macro perspective.

[0054] In one embodiment of the present invention, the method of using the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model includes: obtaining the real large stage distribution in the annotated text; extracting keywords and phrases in the annotated text; generating observation data of the annotated text using the keywords and phrases; calculating the state transition probability between each real large stage distribution in the real large stage distribution in the pre-trained hidden Markov model; calculating the observation probability of the observation data under the real large stage distribution in the pre-trained hidden Markov model; analyzing the virtual large stage distribution corresponding to the observation data based on the state transition probability and the observation probability; calculating the loss function value between the real large stage distribution and the virtual large stage distribution; and training the parameters to be trained of the pre-trained hidden Markov model based on the loss function value to obtain a trained hidden Markov model.

[0055] The observation data refers to a sequence consisting of the keywords and the phrases.

[0056] Optionally, the process of analyzing the virtual large stage distribution corresponding to the observed data according to the state transition probability and the observation probability refers to: obtaining the state transition probability P1(1, 2) between 1 real large stage distribution and 2 real large stage distribution, obtaining the observation probability P2(1, 3) between 1 real large stage distribution and 3 observed data, and calculating P3(2, 3) according to P1(1, 2) and P2(1, 3), which is generally calculated by the extended formula Bayesian probability, prior probability and posterior probability of the Markov chain.

[0057] Optionally, the process of training the parameters to be trained of the pre-trained hidden Markov model based on the loss function value can be implemented by a gradient descent algorithm.

[0058] Furthermore, in an embodiment of the present invention, the hidden state refers to a virtual large-stage distribution corresponding to the annotated text output by the trained hidden Markov model.

[0059] S5. Train the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtain the small-stage distribution of the trained long short-term memory model with respect to the hidden state.

[0060] The embodiment of the present invention trains the pre-trained long short-term memory model through the hidden state, so as to use LSTM to go deep into the details and accurately capture the small stage changes in the operation process. The two complement each other with the aforementioned hidden Markov model to jointly construct a more complete and accurate agent operation process stage identification system.

[0061] In one embodiment of the present invention, the pre-trained long short-term memory model is trained through the hidden state to obtain a trained long short-term memory model, including: obtaining an input layer, a multi-layer long short-term memory layer and a fully connected layer in the pre-trained long short-term memory model; obtaining an input gate, an output gate and a forget gate of the multi-layer long short-term memory layer; after receiving the hidden state through the input end, extracting an initial feature vector of the hidden state using the input gate, the output gate and the forget gate; identifying a long-term dependency relationship of the initial feature vector using the multi-layer long short-term memory layer; converting the initial feature vector into a final feature vector using the multi-layer long short-term memory layer according to the long-term dependency relationship; obtaining a real small stage distribution in the annotated text; calculating a probability distribution of the final feature vector with respect to the real small stage distribution using the fully connected layer; identifying a virtual small stage distribution of the hidden state through the probability distribution; calculating a loss value between the virtual small stage distribution and the real small stage distribution; optimizing the parameters to be optimized in the pre-trained long short-term memory model according to the loss value to obtain a trained long short-term memory model.

[0062] Among them, the long-term dependency relationship refers to the implicit relationship recognized by the stacked multi-layer LSTM, which is represented by the final feature vector. For example, in the scenario where the LSTM of the first layer is stacked with the LSTM of the second layer, the initial feature vector is converted into the final feature vector by the connection relationship between the LSTM of the first layer and the LSTM of the second layer. During the conversion process, the long-term dependency relationship of the identified initial feature vector is blended into the vector to obtain the final feature vector. Therefore, the final feature vector is evolved from the long-term dependency relationship and represents the long-term dependency relationship.

[0063] Furthermore, in an embodiment of the present invention, the small-stage distribution refers to a virtual small-stage distribution corresponding to the hidden state output by the trained long short-term memory model.

[0064] S6. Input the small-stage distribution into the pre-trained language representation model to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model.

[0065] The embodiment of the present invention obtains a trained language representation model by training the pre-trained language representation model through the small-stage distribution, so as to utilize the bidirectional encoding mechanism of BERT and the sequence modeling capability of CRF, so that the combined model can effectively process long texts and complex sentence structures, and even when faced with user demand expressions that have never appeared before, it can still be recognized based on the existing learning model, and has strong generalization ability.

[0066] In one embodiment of the present invention, the pre-trained language representation model is trained through the small-stage distribution to obtain a trained language representation model, including: obtaining a BERT module, an entity recognition module and an intent recognition module in the pre-trained language representation model; using the BERT module to output a semantic vector of the small-stage distribution; using the entity recognition module to identify key entities of the semantic vector; using the intent recognition module to identify user intent corresponding to the semantic vector; according to the key entities, optimizing the first network parameters between the BERT module and the entity recognition module using a preset multi-task learning framework to obtain optimized first network parameters; according to the user intent, optimizing the second network parameters between the BERT module and the intent recognition module using the multi-task learning framework to obtain optimized second network parameters; determining the trained language representation model corresponding to the pre-trained language representation model through the optimized first network parameters and the optimized second network parameters.

[0067] Among them, the first network parameters and the second network parameters both include weights and biases, the first network parameters between the BERT module and the entity recognition module are equivalent to the network parameters that only include the BERT module and the entity recognition module, and the second network parameters between the BERT module and the intent recognition module are equivalent to the network parameters that only include the BERT module and the intent recognition module.

[0068] Optionally, the process of determining the trained language representation model corresponding to the pre-trained language representation model through the optimized first network parameters and the optimized second network parameters refers to taking the language representation model including the optimized first network parameters and the optimized second network parameters as the trained language representation model.

[0069] S7. Obtain the current interaction record between the agent and the current user, and based on the question-answer knowledge base, analyze the agent operation process corresponding to the current interaction record through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model.

[0070] In one embodiment of the present invention, based on the question and answer knowledge base, the agent operation process corresponding to the current interaction record is analyzed through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model, including: identifying the first stage distribution corresponding to the current interaction record through the trained hidden Markov model; outputting the second stage distribution corresponding to the first stage distribution through the trained long short-term memory model; extracting the text key entities and text intent of the second stage distribution using the trained language representation model; querying the question and answer record corresponding to the text key entity in the question and answer knowledge base; and querying the agent operation process corresponding to the text intent in the question and answer record.

[0071] Among them, the text key entity refers to the user text keywords, such as insurance type, insurance amount, terms, etc., the text intention refers to the intention of the user text, such as consultation, quotation, complaint, purchase, etc., and the question and answer record refers to the conversation record related to the text key entity (insurance type, insurance amount, terms, etc.).

[0072] S8. Update the question-and-answer knowledge base through the agent operation process to obtain an updated knowledge base, and use the agent operation process and the updated knowledge base to determine a full-link analysis result of the operation process between the agent and the current user.

[0073] The embodiment of the present invention updates the question and answer knowledge base through the agent operation process to regularly review the content of the knowledge base to ensure the accuracy and timeliness of the information, and at the same time establishes a feedback mechanism to collect feedback from users and the agent team on the use of the knowledge base and continuously optimize the content.

[0074] Among them, the agent operation process refers to the agent response, and the agent response refers to the insurance operation process made by the agent, such as sending product information, arranging subsequent follow-up words, etc.

[0075] In one embodiment of the present invention, the updating of the question and answer knowledge base through the agent operation process to obtain an updated knowledge base includes: obtaining feedback information of the current user on the agent operation process; determining whether the feedback information is negative information; when the feedback information is negative information, querying the required operation process of the current user; correcting the agent operation process to the required operation process to obtain a corrected operation process; and using the corrected operation process to complete the update of the question and answer knowledge base to obtain an updated knowledge base.

[0076] Among them, the feedback information refers to the user's satisfaction survey on the agent's operating process collected in the questionnaire initiated to the user after the question and answer session is completed, the negative information refers to the dissatisfaction information, and the required operating process refers to the agent's operating process required by the current user determined by human experts.

[0077] It can be seen that the embodiment of the present invention defines the standard operating procedures of the agent to divide the agent operating procedures into macroscopic stages and specific small stages from the dual dimensions of online and offline, and deeply analyzes the user's demands for each small stage. Furthermore, the embodiment of the present invention uses the pre-processed text to build a question-and-answer knowledge base between the agent and the historical user to provide users with seamless and efficient problem solving. This strategy strengthens the control over the agent's operating procedures and improves the user experience. The embodiment of the present invention uses the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model for efficiently identifying the large stages of the operating procedures and providing a macro perspective. The embodiment of the present invention trains the pre-trained long short-term memory model through the hidden state to use LSTM to go into details. , accurately capture the small stage changes in the operation process, and complement each other with the aforementioned hidden Markov model to jointly build a more complete and accurate agent operation process stage identification system. The embodiment of the present invention trains the pre-trained language representation model through the small stage distribution to obtain a trained language representation model, which is used to utilize the bidirectional encoding mechanism of BERT and the sequence modeling ability of CRF, so that the combined model can effectively process long texts and complex sentence structures. Even in the face of user demand expressions that have never appeared, it can still be recognized based on the existing learning model and has a strong generalization ability. The embodiment of the present invention updates the question-and-answer knowledge base through the agent operation process to regularly review the content of the knowledge base to ensure the accuracy and timeliness of the information. At the same time, a feedback mechanism is established to collect feedback from users and agent teams on the use of the knowledge base and continuously optimize the content. Therefore, the full-link analysis method, device, electronic device and computer-readable storage medium of the insurance operation process proposed in the embodiment of the present invention can finely manage the insurance operation process of the agent to the user.

[0078] like Figure 3 The figure shows a functional module diagram of the full-link analysis device for the insurance operation process of the present invention.

[0079] The insurance operation process full-link analysis device 100 of the present invention can be installed in an electronic device. According to the functions to be implemented, the insurance operation process full-link analysis device can include a text preprocessing module 101, a knowledge base construction module 102, a model construction module 103, a state acquisition module 104, a stage acquisition module 105, a language training module 106, a process analysis module 107 and a full-link analysis module 108. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, which are stored in the memory of the electronic device.

[0080] In this embodiment, the functions of each module / unit are as follows:

[0081] The text preprocessing module 101 is used to collect historical interaction records between the seat personnel and historical users, convert the historical interaction records into historical interaction texts, and perform text preprocessing on the historical interaction texts to obtain preprocessed texts;

[0082] The knowledge base construction module 102 is used to define the standard operating procedures of the seat staff, annotate the pre-processed text using the standard operating procedures to obtain annotated text, and use the pre-processed text to construct a question-and-answer knowledge base between the seat staff and the historical user;

[0083] The model building module 103 is used to build a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user;

[0084] The state acquisition module 104 is used to train the pre-trained hidden Markov model using the annotated text to obtain a trained hidden Markov model, and obtain the hidden state of the trained hidden Markov model with respect to the annotated text;

[0085] The stage acquisition module 105 is used to train the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtain the small stage distribution of the trained long short-term memory model with respect to the hidden state;

[0086] The language training module 106 is used to input the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model;

[0087] The process analysis module 107 is used to obtain the current interaction record between the agent and the current user, and analyze the agent operation process corresponding to the current interaction record based on the question-answer knowledge base through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model;

[0088] The full-link analysis module 108 is used to update the question-and-answer knowledge base through the agent operation process to obtain an updated knowledge base, and use the agent operation process and the updated knowledge base to determine the full-link analysis result of the operation process between the agent and the current user.

[0089] Optionally, the knowledge base construction module 102 defines the standard operating procedures of the agents, including:

[0090] Divide the work process stages of the agent; wherein the work process stages include the opening stage, the demand exploration stage, the product introduction stage, the objection handling stage, the result facilitation stage and the closing stage;

[0091] Subdividing the operation process phases into subdivided process phases;

[0092] The operation process stage and the subdivided process stage are used as the standard operation process of the agent.

[0093] When the model building module 103 trains the hidden Markov model, it includes:

[0094] Obtaining the true large-scale phase distribution in the annotated text;

[0095] Extracting keywords and phrases from the annotated text;

[0096] generating observation data of the annotated text using the keywords and the phrases;

[0097] calculating, in the pre-trained hidden Markov model, a state transition probability between each of the real large stage distributions in the real large stage distribution;

[0098] Calculating the observation probability of the observation data under the true large-stage distribution in the pre-trained hidden Markov model;

[0099] Analyzing the virtual large-stage distribution corresponding to the observed data according to the state transition probability and the observation probability;

[0100] Calculating a loss function value between the real large stage distribution and the virtual large stage distribution;

[0101] Based on the loss function value, the parameters to be trained of the pre-trained hidden Markov model are trained to obtain a trained hidden Markov model.

[0102] Accordingly, when the model building module 103 trains the long short-term memory model, it includes:

[0103] Obtaining an input layer, a multi-layer long short-term memory layer, and a fully connected layer in the pre-trained long short-term memory model;

[0104] Obtaining input gates, output gates, and forget gates of the multi-layer long short-term memory layer;

[0105] After receiving the hidden state through the input end, extracting an initial feature vector of the hidden state using the input gate, the output gate and the forget gate;

[0106] Using the multi-layer long short-term memory layer to identify the long-term dependency of the initial feature vector;

[0107] According to the long-term dependency, the initial feature vector is converted into a final feature vector using the multi-layer long short-term memory layer;

[0108] Obtain the true small-stage distribution in the annotated text;

[0109] Calculating the probability distribution of the final feature vector with respect to the true small stage distribution using the fully connected layer;

[0110] Identify the virtual small stage distribution of the hidden state through the probability distribution;

[0111] Calculating a loss value between the virtual small stage distribution and the real small stage distribution;

[0112] According to the loss value, the parameters to be optimized in the pre-trained long short-term memory model are optimized to obtain a trained long short-term memory model.

[0113] Accordingly, when the model building module 103 trains the language representation model, it includes:

[0114] Obtaining the BERT module, entity recognition module, and intent recognition module in the pre-trained language representation model;

[0115] Outputting the semantic vector of the small-stage distribution using the BERT module;

[0116] Identifying key entities of the semantic vector using the entity recognition module;

[0117] Using the intention recognition module to identify the user intention corresponding to the semantic vector;

[0118] According to the key entity, optimizing the first network parameters between the BERT module and the entity recognition module using a preset multi-task learning framework to obtain optimized first network parameters;

[0119] According to the user intent, optimizing the second network parameters between the BERT module and the intent recognition module using the multi-task learning framework to obtain optimized second network parameters;

[0120] The trained language representation model corresponding to the pre-trained language representation model is determined by using the optimized first network parameters and the optimized second network parameters.

[0121] Reliably, the process analysis module 107 is specifically used for:

[0122] Identify the first stage distribution corresponding to the current interaction record by using the trained hidden Markov model;

[0123] Outputting the second stage distribution corresponding to the first stage distribution through the trained long short-term memory model;

[0124] Extracting the text key entities and text intent distributed in the second stage using the trained language representation model;

[0125] Querying the question and answer records corresponding to the key text entities in the question and answer knowledge base;

[0126] The agent operation flow corresponding to the text intent is searched in the question and answer record.

[0127] The full-link analysis module 108 is specifically used for:

[0128] Obtaining feedback information from the current user on the agent's operating process;

[0129] Determining whether the feedback information is negative information;

[0130] When the feedback information is negative information, querying the required operation flow of the current user;

[0131] Modifying the seat operation flow into the required operation flow to obtain a modified operation flow;

[0132] The question-and-answer knowledge base is updated using the correction process to obtain an updated knowledge base.

[0133] like Figure 4 As shown, it is a structural schematic diagram of an electronic device 1 for implementing a full-link analysis method of an insurance operation process according to the present invention.

[0134] The electronic device 1 may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a full-link analysis program for an insurance operation process.

[0135] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (ControlUnit) of the electronic device 1, and uses various interfaces and lines to connect the various components of the entire electronic device 1, and executes or executes programs or modules stored in the memory 11 (for example, executing the full-link analysis program of the insurance operation process, etc.), and calls the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0136] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Furthermore, the memory 11 can also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed on the electronic device 1, such as the code of the full-link analysis program of the insurance operation process, but also can be used to temporarily store data that has been output or is to be output.

[0137] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0138] The communication interface 13 is used for communication between the above-mentioned electronic device 1 and other devices, including a network interface and an employee interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices 1. The employee interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the employee interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual employee interface.

[0139] Figure 4 Only the electronic device 1 having components is shown, and those skilled in the art can understand that Figure 4 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0140] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0141] It should be understood that the embodiment is for illustrative purposes only and the scope of the patent invention is not limited to this structure.

[0142] The insurance operation process full-link analysis program stored in the memory 11 of the electronic device 1 is a combination of multiple computer programs. When running in the processor 10, it can achieve:

[0143] Collecting historical interaction records between agents and historical users, converting the historical interaction records into historical interaction texts, and performing text preprocessing on the historical interaction texts to obtain preprocessed texts;

[0144] Defining a standard operating procedure for the agent, annotating the preprocessed text using the standard operating procedure to obtain annotated text, and constructing a question-and-answer knowledge base between the agent and the historical user using the preprocessed text;

[0145] Constructing a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user;

[0146] Using the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text;

[0147] Training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small-stage distribution of the trained long short-term memory model with respect to the hidden state;

[0148] Inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model;

[0149] Obtaining a current interaction record between the agent and the current user, and analyzing the agent operation process corresponding to the current interaction record based on the question-answer knowledge base by using the trained hidden Markov model, the trained long short-term memory model, and the trained language representation model;

[0150] The question-and-answer knowledge base is updated through the agent operation process to obtain an updated knowledge base, and the agent operation process and the updated knowledge base are used to determine the full-link analysis result of the operation process between the agent and the current user.

[0151] Specifically, the specific implementation method of the processor 10 for the above computer program can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0152] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0153] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of the electronic device 1, the computer program can implement:

[0154] Collecting historical interaction records between agents and historical users, converting the historical interaction records into historical interaction texts, and performing text preprocessing on the historical interaction texts to obtain preprocessed texts;

[0155] Defining a standard operating procedure for the agent, annotating the preprocessed text using the standard operating procedure to obtain annotated text, and constructing a question-and-answer knowledge base between the agent and the historical user using the preprocessed text;

[0156] Constructing a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user;

[0157] Using the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text;

[0158] Training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small-stage distribution of the trained long short-term memory model with respect to the hidden state;

[0159] Inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model;

[0160] Obtaining a current interaction record between the agent and the current user, and analyzing the agent operation process corresponding to the current interaction record based on the question-answer knowledge base by using the trained hidden Markov model, the trained long short-term memory model, and the trained language representation model;

[0161] The question-and-answer knowledge base is updated through the agent operation process to obtain an updated knowledge base, and the agent operation process and the updated knowledge base are used to determine the full-link analysis result of the operation process between the agent and the current user.

[0162] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0163] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0165] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0166] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0167] The embodiments of the present invention can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0168] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

[0170] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0171] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit it. If software tools or components other than those of the company appear in the application embodiments, they are only used for illustrative purposes and do not represent actual use. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A full-link analysis method for insurance operation process, characterized in that: The method comprises: Collecting historical interaction records between agents and historical users, converting the historical interaction records into historical interaction texts, and performing text preprocessing on the historical interaction texts to obtain preprocessed texts; Defining a standard operating procedure for the agent, annotating the preprocessed text using the standard operating procedure to obtain annotated text, and constructing a question-and-answer knowledge base between the agent and the historical user using the preprocessed text; Constructing a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user; Using the annotated text to train the pre-trained hidden Markov model to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text; Training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small-stage distribution of the trained long short-term memory model with respect to the hidden state; Inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model; Obtaining a current interaction record between the agent and the current user, and analyzing the agent operation process corresponding to the current interaction record based on the question-answer knowledge base by using the trained hidden Markov model, the trained long short-term memory model, and the trained language representation model; The question-and-answer knowledge base is updated through the agent operation process to obtain an updated knowledge base, and the agent operation process and the updated knowledge base are used to determine the full-link analysis result of the operation process between the agent and the current user.

2. The insurance operation process full-link analysis method according to claim 1, characterized in that: The standard operating procedures for defining the agents include: Divide the work process stages of the agent; wherein the work process stages include the opening stage, the demand exploration stage, the product introduction stage, the objection handling stage, the result facilitation stage and the closing stage; Subdividing the operation process phases into subdivided process phases; The operation process stage and the subdivided process stage are used as the standard operation process of the agent.

3. The insurance operation process full-link analysis method according to claim 1, characterized in that: The method of training the pre-trained hidden Markov model using the annotated text to obtain a trained hidden Markov model includes: Obtaining the true large-scale phase distribution in the annotated text; Extracting keywords and phrases from the annotated text; generating observation data of the annotated text using the keywords and the phrases; calculating, in the pre-trained hidden Markov model, a state transition probability between each of the real large stage distributions in the real large stage distribution; Calculating the observation probability of the observation data under the true large-stage distribution in the pre-trained hidden Markov model; Analyzing the virtual large-stage distribution corresponding to the observed data according to the state transition probability and the observation probability; Calculating a loss function value between the real large stage distribution and the virtual large stage distribution; Based on the loss function value, the parameters to be trained of the pre-trained hidden Markov model are trained to obtain a trained hidden Markov model.

4. The full-link analysis method for insurance operation process according to claim 1, characterized in that: The pre-trained long short-term memory model is trained by the hidden state to obtain a trained long short-term memory model, including: Obtaining an input layer, a multi-layer long short-term memory layer, and a fully connected layer in the pre-trained long short-term memory model; Obtaining input gates, output gates, and forget gates of the multi-layer long short-term memory layer; After receiving the hidden state through the input end, extracting an initial feature vector of the hidden state using the input gate, the output gate and the forget gate; Using the multi-layer long short-term memory layer to identify the long-term dependency of the initial feature vector; According to the long-term dependency, the initial feature vector is converted into a final feature vector using the multi-layer long short-term memory layer; Obtain the true small-stage distribution in the annotated text; Calculating the probability distribution of the final feature vector with respect to the true small stage distribution using the fully connected layer; Identify the virtual small stage distribution of the hidden state through the probability distribution; Calculating a loss value between the virtual small stage distribution and the real small stage distribution; According to the loss value, the parameters to be optimized in the pre-trained long short-term memory model are optimized to obtain a trained long short-term memory model.

5. The full-link analysis method for insurance operation process according to claim 1, characterized in that: The pre-trained language representation model is trained by the small-stage distribution to obtain a trained language representation model, including: Obtaining the BERT module, entity recognition module, and intent recognition module in the pre-trained language representation model; Outputting the semantic vector of the small-stage distribution using the BERT module; Identifying key entities of the semantic vector using the entity recognition module; Using the intention recognition module to identify the user intention corresponding to the semantic vector; According to the key entity, optimizing the first network parameters between the BERT module and the entity recognition module using a preset multi-task learning framework to obtain optimized first network parameters; According to the user intent, optimizing the second network parameters between the BERT module and the intent recognition module using the multi-task learning framework to obtain optimized second network parameters; The trained language representation model corresponding to the pre-trained language representation model is determined by using the optimized first network parameters and the optimized second network parameters.

6. The insurance operation process full-link analysis method according to claim 1, characterized in that: The analyzing the agent operation process corresponding to the current interaction record based on the question-answer knowledge base through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model includes: Identify the first stage distribution corresponding to the current interaction record by using the trained hidden Markov model; Outputting the second stage distribution corresponding to the first stage distribution through the trained long short-term memory model; Extracting the text key entities and text intent distributed in the second stage using the trained language representation model; Querying the question and answer records corresponding to the key text entities in the question and answer knowledge base; The agent operation flow corresponding to the text intent is searched in the question and answer record.

7. The full-link analysis method for insurance operation process according to claim 1, characterized in that: The updating of the question-and-answer knowledge base through the agent operation process to obtain an updated knowledge base includes: Obtaining feedback information from the current user on the agent's work process; Determining whether the feedback information is negative information; When the feedback information is negative information, querying the required operation flow of the current user; Modifying the seat operation flow into the required operation flow to obtain a modified operation flow; The question-and-answer knowledge base is updated using the correction process to obtain an updated knowledge base.

8. A device for analyzing the entire chain of insurance operation processes, characterized in that: The device comprises: A text preprocessing module is used to collect historical interaction records between agents and historical users, convert the historical interaction records into historical interaction texts, and perform text preprocessing on the historical interaction texts to obtain preprocessed texts; A knowledge base construction module, used to define a standard operating procedure for the agent, annotate the preprocessed text using the standard operating procedure to obtain annotated text, and construct a question-and-answer knowledge base between the agent and the historical user using the preprocessed text; A model building module, used to build a pre-trained hidden Markov model, a pre-trained long short-term memory model and a pre-trained language representation model between the agent and the historical user; A state acquisition module, used for training the pre-trained hidden Markov model using the annotated text to obtain a trained hidden Markov model, and obtaining a hidden state of the trained hidden Markov model with respect to the annotated text; A stage acquisition module, used for training the pre-trained long short-term memory model through the hidden state to obtain a trained long short-term memory model, and obtaining a small stage distribution of the trained long short-term memory model with respect to the hidden state; A language training module, used for inputting the small-stage distribution into the pre-trained language representation model, so as to train the pre-trained language representation model through the small-stage distribution to obtain a trained language representation model; A process analysis module, used to obtain the current interaction record between the agent and the current user, and analyze the agent operation process corresponding to the current interaction record based on the question-answer knowledge base through the trained hidden Markov model, the trained long short-term memory model and the trained language representation model; The full-link analysis module is used to update the question-and-answer knowledge base through the agent operation process to obtain an updated knowledge base, and use the agent operation process and the updated knowledge base to determine the full-link analysis result of the operation process between the agent and the current user.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the full-link analysis method of the insurance operation process as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the full-link analysis method of the insurance operation process as described in any one of claims 1 to 7.

Images