Object data processing method and device based on user objection, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence and the field of financial science and technology, and discloses an objection data processing method and device based on user objection, equipment and a medium, and the method comprises the steps: obtaining recording data of a target user, and converting the recording data into text data; screening and cleaning the text data through a preset attribute field to obtain preprocessed data; performing filling processing on the preprocessed data by utilizing a preset prompt word model to obtain a prompt word template; performing conversion processing on the prompt word template by utilizing a preset large model to obtain user objection data; performing multi-level grouping on the user objection data according to a preset category to obtain grouped data, and performing statistics on frequency distribution of different categories of user objection data according to the grouped data; a specific data mining technology is applied to the insurance industry and used for mining records between a user and staff such as a supervisor to reply some insurance problems of the user, so that the working efficiency of the insurance supervisor is improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, device, equipment and medium for processing objection data based on user objections. Background Art

[0002] Insurance companies are financial institutions that specialize in risk management and economic security. They collect premiums from many customers, provide risk protection for them, transfer individual risks to group risks, and provide financial compensation to customers when they encounter risks to help them overcome difficulties. The existence of insurance companies can not only help individuals and families deal with unexpected risks and maintain economic security, but also stabilize economic and social development, promote production and consumption, and improve the level of social welfare.

[0003] The renewal business of an insurance company refers to the business of renewing an already effective policy after the expiration of the policy and collecting renewal premiums. It is an important part of the insurance company's continued operation and is crucial to the insurance company's profitability and sustainable development. The renewal business of an insurance company is generally followed up by supervisors and other staff members, who generally make outbound calls to users to provide service reminders before the policy expires.

[0004] In actual operations, supervisors and other staff members often check the user's policy information, historical contact records and recordings on their systems before contacting the user, prepare questions that the user may ask in advance, and prepare relevant response scripts in advance; by increasing the user's recognition of their own professional abilities, the user's renewal payment rate can be increased.

[0005] However, with the increasing number of existing policy users, the scale of users that supervisors and other staff members need to manage each year is also gradually increasing, and the policy terms, economic situation, and family situation of each user may be completely different. The traditional method of manually organizing user information in advance is time-consuming, labor-intensive, and inefficient. Secondly, since user information can only reflect the user's past situation, questions that users have not reported still need to be answered on-site by supervisors and other staff members based on their experience, which places high demands on the work ability of supervisors and other staff members. Summary of the invention

[0006] The present invention provides a method, device, computer equipment and medium for processing objection data based on user objection, so as to solve the technical problems of low work efficiency and insufficient accuracy of insurance supervisors.

[0007] In a first aspect, a method for processing objection data based on user objection is provided, comprising:

[0008] Acquire the target user's audio recording data, and convert the audio recording data into text data;

[0009] Screening and cleaning the text data through preset attribute fields to obtain preprocessed data;

[0010] Using a preset prompt model to fill in the preprocessed data to obtain a prompt template;

[0011] The prompt template is transformed by using a preset large model to obtain user objection data;

[0012] The user objection data is grouped in multiple levels according to preset categories to obtain grouped data, and frequency distribution of user objection data of different categories is counted based on the grouped data.

[0013] In a second aspect, a device for processing objection data based on user objection is provided, comprising:

[0014] The acquisition module is used to obtain the recording data of the target user;

[0015] A cleaning module, used to filter and clean the text data through preset attribute fields to obtain pre-processed data;

[0016] A filling module, used to fill the pre-processed data with a preset prompt model to obtain a prompt template;

[0017] A conversion module, used to convert the prompt template using a preset large model to obtain user objection data;

[0018] The analysis module is used to group the user objection data into multiple levels according to preset categories to obtain group data, and to calculate the frequency distribution of user objection data of different categories based on the group data.

[0019] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for processing objection data based on user objection when executing the computer program.

[0020] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the objection data processing method based on user objection is implemented.

[0021] The above-mentioned objection data processing method, device, computer equipment and storage medium based on user objection can be applied to objection data processing engines such as supervisors or assistants in various application scenarios. The objection data processing engine is usually implemented through a server, which can obtain the recording data in the database in real time. In the embodiment of the present invention, the objection data processing method is applied to the insurance industry to mine the recordings between users and supervisors and other staff to answer some insurance questions of users, so as to improve the work efficiency and user experience of supervisors and other staff. Even if the number of existing policy users increases, supervisors and other staff can also handle it effectively. By mining the historical recordings of customer service and customers, high-frequency objection questions and dissuasion strategies and response dialogues of customers can be generated, thereby improving the efficiency and accuracy of objection data processing and experience summary. On the other hand, the completeness and availability of the dialogues can be improved, further improving the supervisory operation efficiency and success rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0023] Figure 1 It is a schematic diagram of an application environment of a method for processing objection data based on user objection in one embodiment of the present invention;

[0024] Figure 2 is a flow chart of a method for processing objection data based on user objection in one embodiment of the present invention;

[0025] Figure 3 is an example diagram of a prompt template of a method for processing objection data based on user objection in one embodiment of the present invention;

[0026] Figure 4 is an example diagram of frequency distribution of user objection data according to an objection data processing method based on user objection in one embodiment of the present invention;

[0027] Figure 5 It is an example diagram of personalized user objection data of an objection data processing method based on user objection in one embodiment of the present invention;

[0028] Figure 6 is a structural schematic diagram of an objection data processing device based on user objection in one embodiment of the present invention;

[0029] Figure 7 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;

[0030] Figure 8 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0031] 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.

[0032] The objection data processing method based on user objection provided by the embodiment of the present invention can be applied in the following aspects: Figure 1 In the application environment, the user terminal communicates with the server terminal through the network. The server terminal can receive user questions through the user terminal, and perform question recall processing from a preset standard question library according to the user questions to obtain multiple candidate questions; sort the multiple candidate questions according to the semantic matching degree between the user questions and the candidate questions, and obtain the first sorting result of the candidate questions; respectively judge the entity alignment of the user questions and each candidate question. In the present invention, for complex insurance entities such as insurance business, the entity alignment can be used to optimize the question-answering engine. First, the semantic matching degree is used for rough sorting, and then the sorting is adjusted by the entity alignment method. The top matching questions are selected to respond to the answers obtained, which can effectively avoid the generalization ability defects of the model, greatly and efficiently improve the effect of entity matching, and improve the effect of the question-answering engine. Among them, the user terminal can be but not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server terminal can be implemented by an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0033] See also Figure 2 As shown, Figure 2 A flowchart of a method for processing objection data based on user objection provided in an embodiment of the present invention includes the following steps:

[0034] S1. Obtain recording data of a target user, and convert the recording data into text data.

[0035] The objection data processing method based on user objection provided by the present invention can be applied to objection data processing engines such as supervisors or assistants in various application scenarios. The objection data processing engine is usually implemented through a server, which can obtain recorded data in a database in real time. In an embodiment of the present invention, specific data mining technology is applied to the insurance industry to mine recordings between users and supervisors and other staff to respond to some insurance questions of users, so as to improve the work efficiency and work accuracy of insurance supervisors.

[0036] In an embodiment of the present invention, the recorded data refers to the recorded calls between supervisors and other staff and target users in previous work, which are collected by users and stored in a database; converting the recorded data into text data can greatly improve the efficiency of information acquisition and processing, because it can convert audio information into readable, searchable, and analyzable text, which facilitates the subsequent steps to quickly obtain information, perform text analysis, and establish a database, thereby greatly improving the efficiency and value of information processing.

[0037] In the embodiment of the present invention, the converting the recorded data into text data includes:

[0038] Convert the recorded data into feature vectors using preset recording parameters;

[0039] Mapping the feature vector using preset phonemes to obtain a phoneme sequence;

[0040] The characteristic vector is converted using a preset acoustic model to obtain a characteristic signal;

[0041] The characteristic signal and the phoneme sequence are subjected to data fusion to obtain text data.

[0042] In an embodiment of the present invention, the preset recording parameters are the pitch, timbre and rhythm of the recording itself; the acoustic model adopts HMM (Hidden Markov Model); the feature vector extraction adopts MFCC (Mel-Frequency Cepstral Coefficients); the phoneme refers to the smallest meaningful unit in speech, such as "a", "b", "p", and a phoneme corresponds to a specific pronunciation, for example, the phoneme "a" corresponds to the pronunciation of "ah"; the acoustic model output signal and the phoneme sequence are combined using the Viterbi algorithm to find the best phoneme sequence and text sequence.

[0043] In detail, the recording data is converted into a feature vector based on the pitch, timbre and rhythm of the recording itself; the feature vector is input into the HMM acoustic model to obtain the acoustic model output signal; the feature vector is mapped to the most meaningful unit in the speech, where one phoneme corresponds to a specific pronunciation, to obtain a phoneme sequence, and the acoustic model output signal and the phoneme sequence are combined using the Viterbi algorithm to obtain text data; this method has a high conversion efficiency, is highly robust to noise and channel changes, and the conversion result is relatively accurate.

[0044] In the embodiment of the present invention, converting the recorded data into text data can improve work efficiency and facilitate information retrieval and analysis.

[0045] S2. Screening and cleaning the text data through preset attribute fields to obtain preprocessed data.

[0046] In an embodiment of the present invention, the preset attribute field includes the total number of words in the text file and the content of the text data; this embodiment sets an upper limit and a lower limit of the total number of words in the text file. When some supervisors and other staff communicate with users, the communication content may contain content that is not required by this embodiment. When there is too much unnecessary content, it will exceed the set upper limit of the total number of words in the text file and will be screened out in this step; similarly, when some supervisors and other staff communicate with users, there may be less communication content. In this case, the text content is mostly inquiries made by supervisors and other staff, but the user did not answer the inquiry or the answer content was meaningless, resulting in less text content that did not reach the set lower limit of the total number of words in the text file, and will be screened out in this step; secondly, this step sets keywords related to "product, premium, and payment times". If the relevant keywords do not appear in the text data, they will be screened out in this step; specific keywords can be customized according to actual use and needs.

[0047] In the embodiment of the present invention, the screening and cleaning of the text data by using the preset attribute fields includes:

[0048] Performing format conversion on the text data according to a preset format to obtain converted data;

[0049] Get a multidimensional table pre-built based on preset attribute fields;

[0050] The transformed data is screened and cleaned using multidimensional tables to obtain preprocessed data.

[0051] In the embodiment of the present invention, the preset format refers to sentences and paragraphs. If the text data of sentences and paragraphs are not satisfied, they will be filtered, which can improve the readability, analyzability, storability, portability and interactivity of the data, so as to better utilize the data, enhance the data value, filter out useless information and improve the data quality. Multidimensional table, also known as cross table or multidimensional pivot table, is a table form that classifies and displays data in multiple dimensions. It can show the association and cross relationship between data in different dimensions. In the embodiment of the present invention, by building tables for different attributes of the same file, and then setting different restriction conditions for different attributes, finally taking the intersection of each condition, multiple dimensions of screening are realized. Multidimensional table can be used to conveniently screen and clean the conversion data. By classifying and displaying the conversion data according to different dimensions, such as channel, time, region, etc., data outliers, duplicate data and incomplete data can be intuitively identified, and screening and cleaning can be performed according to actual conditions, so as to obtain more accurate and reliable conversion data, and provide more accurate basis for subsequent data analysis and decision-making.

[0052] In detail, the specific content of the text data formatting is to divide the text into sentences and paragraphs, and remove irrelevant characters to facilitate subsequent processing; multi-dimensional table screening is a data processing method, which can build tables for different attributes of the same file, and then set different restrictions for different attributes, and finally take the intersection of each condition to achieve multi-dimensional screening. In this embodiment, a multi-dimensional table is used to match the keywords in the conversion data, screen the keyword sentences required by the present invention, and filter the sentences that do not contain keywords; the method of screening and cleaning text data using a multi-dimensional table adopted in this step is a simple, efficient and intuitive text processing method, which can effectively improve the efficiency of data sorting and analysis.

[0053] S3, using a preset prompt model to fill the pre-processed data to obtain a prompt template;

[0054] In the embodiment of the present invention, please refer to Figure 3 As shown, the preset prompt model first manually listens to the recorded data to understand the conversation content between supervisors and other staff and users, determines the detailed tasks of the large model to mine the recorded data, and establishes the required prompts. The embodiment of the present invention adopts four modules of "role definition-goal description-task introduction-output specification" to design prompts; the preset prompt model is used to fill in the pre-processed data, and missing values ​​or blank fields in the data can be filled according to the preset constraints and output specifications. For example, missing dates, names or product categories can be inferred based on existing data, thereby improving the integrity and consistency of the data and providing a more complete template for subsequent data analysis and application.

[0055] In the embodiment of the present invention, the pre-processing data is processed by using a preset prompt model to obtain a prompt template, including:

[0056] Establishing a basic template structure through the preset prompt model;

[0057] Use the preset tasks to add the basic structure of the template to obtain the preset template;

[0058] Add the preset template using the preset output specification to obtain the standard template;

[0059] Use preset constraints to constrain the standard template to obtain a finished product template;

[0060] The finished template is filled with the preprocessed data to obtain a prompt template.

[0061] In an embodiment of the present invention, the basic structure of the template adopts a four-module structure of "role definition-goal description-task introduction-output specifications", the preset tasks include word limit, style requirements, keyword requirements, content range, etc., the preset output specifications output content in a tabular form, and the preset constraints can set word limits, keywords, format requirements and tone requirements according to actual needs.

[0062] In detail, based on the basic structure of the template, tasks such as word limit, style requirements, keyword requirements, content scope, etc. are added to obtain a preset template; output specifications in tabular form are added to the preset template to obtain a standard template; according to actual needs, constraints such as word limit, keywords, format requirements and tone requirements are added to the standard template, and the finished template is filled with preprocessed data to obtain a prompt template; the prompt template set by this scheme has high versatility, can add restrictions according to actual needs, and can be applied to various scenarios; at the same time, the prompt template set by this scheme has high accuracy, can exert more powerful capabilities, and generate higher quality content.

[0063] In an embodiment of the present invention, supervisors and other staff members can constrain the prompt template according to actual needs, thereby improving the practicality of the prompt template. After the prompt template is established, it is convenient for supervisors and other staff members to perform subsequent operations, and supervisors and other staff members do not need to repeat the operations again, which can effectively improve the work efficiency of supervisors and other staff members. After the prompt template is established, the constraints of the prompt template can be changed according to actual needs, which facilitates the updating of the prompt template and further improves the practicality of the prompt template.

[0064] S4. Using a preset large model to transform the prompt template to obtain user objection data.

[0065] In the embodiment of the present invention, the large model refers to a large language model, and the input of the prompt template into the preset large model refers to the process of using the prompt template in S3 to interact with the large language model. Through the prompt template of S3, the ability of the large model can be stimulated and more expected results can be obtained; the large model is a machine learning model with a large number of parameters, usually based on a neural network, which can build a more accurate voice model by learning massive voice data, improve the accuracy of voice recognition, and recognize more complex voice content, such as accents, background noise, etc. It can also be used for voice synthesis, voice translation, etc., to provide people with a more convenient and intelligent voice interaction experience.

[0066] In an embodiment of the present invention, the preset large model understands the content of the task to be executed this time based on the description of the prompt template, and according to the requirements of the output specification, finds the four items of "customer name, insurance product, objection question, customer service dissuasion strategy and speech" respectively by traversing the recorded text, and outputs them in sequence.

[0067] In the embodiment of the present invention, the method of processing the prompt template using a preset large model to obtain user objection data includes:

[0068] Inputting the prompt template into a preset large model; the preset large model includes an encoder, a word embedding, a decoder, and a table data processing library;

[0069] Encode the prompt template using an encoder to obtain digital data;

[0070] Use word embedding to map digital data to obtain vector data;

[0071] Use the decoder to encode the vector data to obtain a text sequence;

[0072] Use the table data processing library to create table data;

[0073] Use the table data processing library to sequence the text and output the user objection data.

[0074] In the embodiment of the present invention, the preset large model refers to LLM (Large Language Model). The large model provides a more convenient data processing method. It uses the powerful learning ability of the large model to learn more complex speech patterns and recognize more subtle speech features, thereby improving the recognition accuracy. In this embodiment, the prompt template is converted into user objection data. The encoder adopts One-Hot encoding to convert the phonemes or syllables in the speech signal into numerical form, which is convenient for the model to process; its advantages are simple and easy to understand, easy to implement, and more effective for recognizing a limited number of phonemes or syllables. Word embedding is to map each word to a high-dimensional vector, which can map phonemes or syllables to a low-dimensional vector space. It can capture the semantic relationship between phonemes or syllables, such as tone, rhyme, etc., and at the same time reduce the dimension of high-dimensional sparse vectors, improve the model training efficiency, and make the model easier to learn speech features, thereby improving the accuracy of speech recognition. The table data processing library is used to create, operate and format table data. In the field of speech recognition, it can help us organize and manage a large amount of speech data, such as storing speech recognition results, audio features, etc. in table form, which is convenient for data cleaning, screening and analysis.

[0075] In detail, the prompt template is input into LLM, which includes an encoder, word embedding, decoder, and table data processing library. The prompt template is encoded by One-Hot encoder to obtain digital data; the digital data is mapped by word embedding, and each word is mapped to a high-dimensional vector to obtain vector data; the vector data is encoded by decoder to obtain a text sequence; the table data processing library is used to establish table data; the text sequence is sequenced by the table data processing library, and the user objection data is output; the large model is used to output the user objection data to improve efficiency and output quality. At the same time, the "customer service dissuasion strategy and speech" set in this step can be matched according to the user's "objection problem". When the user raises an objection problem, it can directly help supervisors and other staff to provide corresponding strategies and speech, which can effectively improve customer satisfaction, reduce the high work ability requirements of supervisors and other staff, and improve work efficiency and work quality.

[0076] S5. Group the user objection data into multiple levels according to preset categories to obtain grouped data, and calculate the frequency distribution of user objection data of different categories based on the grouped data.

[0077] In an embodiment of the present invention, user objection data is grouped at multiple levels according to preset categories, and different types of objections can be classified and sorted. For example, they are grouped according to product type, problem type, resolution status, etc., to form multi-level structured data, which helps us to understand the distribution of user objections more clearly, quickly locate the core of the problem, and take targeted solutions according to different categories to improve the efficiency of user objection processing. Frequency distribution of user objection data of different categories based on grouped data can help us analyze the main types and quantities of user objections, such as which products or services have more objections, and which problem types appear more frequently, so as to discover potential pain points in user experience, and give priority to high-frequency objections according to the frequency distribution, formulate more effective improvement measures, and improve user satisfaction.

[0078] In the embodiment of the present invention, the step of grouping the user objection data in multiple levels according to preset categories to obtain grouped data, and collecting frequency distribution statistics of different categories of user objection data according to the grouped data includes:

[0079] Grouping different question types in the user objection data to obtain question type grouping data;

[0080] Grouping different specific items of the same question type in the user objection data to obtain specific item grouping data;

[0081] The frequency distribution data of different objection issues are statistically analyzed based on the frequency of occurrence of specific project grouping data and problem type grouping data.

[0082] In the embodiment of the present invention, please refer to Figure 4 As shown, the same question type refers to questions with the same actual meaning but different wordings, such as "too expensive", "can it be cheaper", "can't bear to buy", etc. These sentences do not have common vocabulary, but express the same meaning. These question types are regarded as the same question type in this step and are marked as "insurance is too expensive". The different specific items in the same question type in the user objection data refer to the types of different types of items with the same problem. For example, "insurance A" and "insurance B" both have the problem of "insurance is too expensive", then "insurance A" and "insurance B" are specific item grouping data.

[0083] In detail, taking "insurance is too expensive" as an example, the questions in the user objection data that have the same actual meaning as "insurance is too expensive" but have different wordings are grouped to obtain "insurance is too expensive" grouped data, and different insurance types in the "insurance is too expensive" grouped data are grouped to obtain different insurance type grouped data, and the insurance type grouped data is divided by the "insurance is too expensive" grouped data to obtain the frequency distribution data of "insurance is too expensive" questions raised by each insurance type; supervisors and other staff can make judgments based on the frequency distribution data, which helps supervisors and other staff understand the actual needs of users, improve products and services, formulate effective strategies, and conduct data analysis and insights, ultimately improving customer satisfaction and corporate competitiveness.

[0084] In the embodiment of the present invention, after counting the frequency distribution of different categories of user objection data according to the grouped data, the method further includes:

[0085] Mapping the prompt template into a SQL (Structured Query Language) statement to obtain a keyword mapping;

[0086] Use the prompt template to parse the keyword mapping and obtain the syntax tree;

[0087] Combine the syntax tree and keyword mapping to get the query statement;

[0088] Establish a user objection database using user objection data;

[0089] Use the query statement to query the user objection database and obtain the query output.

[0090] In an embodiment of the present invention, the SQL is a structured query language. SQL can be used to manage and analyze voice data in the field of speech recognition, such as storing and querying speech recognition results, audio features and other data, and can use the powerful functions of SQL to perform data screening, sorting, statistics and other operations, thereby helping us to better understand the voice data. In this embodiment, it is used to implement interaction with the database. The analysis adopts the method of natural language processing. First, the sentence is segmented and the part-of-speech is tagged, and then a syntax tree is constructed according to dependency syntactic analysis.

[0091] In detail, the prompt template is mapped to the SQL statement to obtain a keyword mapping; parsing is performed based on the prompt template, first the statement is segmented and POS tagged, and then a syntax tree is constructed according to dependency syntax analysis to obtain a syntax tree; the syntax tree and the keyword mapping are combined to obtain a query statement for interaction with the database, and a user objection database is established based on user objection data; the user objection database is queried using the query statement to obtain a query output; this solution can obtain detailed information of the customer through basic information and output, and supervisors and other staff can directly obtain relevant information such as the customer's actual needs through this information. On the basis of S5, the detailed information can be accompanied by corresponding strategies and guiding words, which supervisors and other staff can use directly, greatly improving the work efficiency and work quality of supervisors and other staff, while reducing work pressure and improving user satisfaction.

[0092] In the embodiment of the present invention, after counting the frequency distribution of different categories of user objection data according to the grouped data, the method further includes:

[0093] Create Sphinx documentation using preset files;

[0094] The Sphinx document is used to attach the prompt template, so that the Sphinx document and the prompt template are input into the preset large model together, and the personalized user objection data is output.

[0095] In the embodiment of the present invention, please refer to Figure 5 As shown, the Sphinx document is an information carrier that can store text information. The preset file uses the command sphinx-quickstart to initialize a Sphinx project, sets relevant options according to the prompts, and uses the conf.py file to configure the project. In this embodiment, the configuration project is the personality preferences of supervisors and other staff members.

[0096] Specifically, a Sphinx document is created using a preset personality preference file of supervisors and other staff members, the Sphinx document is attached to a prompt template, the Sphinx document and the prompt template are input into a preset large model together, and personalized user objection data is output; this solution facilitates communication and exchange: using a unified personality grouping label can make communication and exchange more convenient, reduce misunderstandings, and improve efficiency; at the same time, being able to adjust the preset file to a group number and other file can avoid the requirements of different supervisors and other staff members from being applied to the tasks of other supervisors and other staff members, avoid confusion in the large model, and improve the robustness of the large model output.

[0097] The scheme implemented by the objection data processing method based on the above-mentioned user objection can be applied to objection data processing engines such as supervisors or assistants in various application scenarios. The objection data processing engine is usually implemented through a server, which can obtain the recording data in the database in real time. In the embodiment of the present invention, the objection data processing method based on user objection is applied to the insurance industry to mine the recordings between users and supervisors and other staff to answer some insurance questions of users, so as to improve the work efficiency and user experience of supervisors and other staff. Even if the number of existing policy users increases, insurance supervisors can also handle it effectively. By mining the historical recordings of customer service and customers, high-frequency objection questions and golden words of supervision are generated, so as to improve the efficiency and accuracy of objection data processing and experience summary. On the other hand, one-stop completion of key elements in the speech is achieved, the completeness and availability of the speech is improved, and the supervisory operation efficiency and success rate are further improved.

[0098] In one embodiment, a device for processing objection data based on user objection is provided, and the device for processing objection data based on user objection corresponds one-to-one to the method for processing objection data based on user objection in the above embodiment. Figure 6 As shown, the objection data processing device based on user objection includes an acquisition module 101, a cleaning module 102, a filling module 103, a conversion module 104 and an analysis module 105. The functional modules are described in detail as follows:

[0099] The acquisition module 101 is used to acquire the recording data of the target user;

[0100] A cleaning module 102, used to filter and clean the text data through preset attribute fields to obtain pre-processed data;

[0101] A filling module 103 is used to fill the pre-processed data using a preset prompt model to obtain a prompt template;

[0102] The conversion module 104 is used to convert the prompt template using a preset large model to obtain user objection data;

[0103] The analysis module 105 is used to group the user objection data into multiple levels according to preset categories to obtain group data, and to calculate the frequency distribution of user objection data of different categories based on the group data.

[0104] In one embodiment, the analysis module 105 is used to:

[0105] Grouping different question types in the user objection data to obtain question type grouping data;

[0106] Grouping different specific items of the same question type in the user objection data to obtain specific item grouping data;

[0107] The frequency distribution data of different objection issues are statistically analyzed based on the frequency of occurrence of specific project grouping data and problem type grouping data.

[0108] In one embodiment, the analysis module 105 is further configured to:

[0109] Mapping the prompt template to an SQL statement to obtain a keyword mapping;

[0110] Use the prompt template to parse the keyword mapping and obtain the syntax tree;

[0111] Combine the syntax tree and keyword mapping to get the query statement;

[0112] Establish a user objection database using user objection data;

[0113] Use the query statement to query the user objection database and obtain the query output.

[0114] In one embodiment, the cleaning module 101 is used to:

[0115] Convert the recorded data into feature vectors using preset recording parameters;

[0116] Mapping the feature vector using preset phonemes to obtain a phoneme sequence;

[0117] The characteristic vector is converted using a preset acoustic model to obtain a characteristic signal;

[0118] The characteristic signal and the phoneme sequence are subjected to data fusion to obtain text data.

[0119] In one embodiment, the cleaning module 102 is used to:

[0120] Performing format conversion on the text data according to a preset format to obtain converted data;

[0121] Get a multidimensional table pre-built based on preset attribute fields;

[0122] The transformed data is screened and cleaned using multidimensional tables to obtain preprocessed data.

[0123] In one embodiment, the filling module 103 is used to:

[0124] Establishing a basic template structure through the preset prompt model;

[0125] Use the preset tasks to add the basic structure of the template to obtain the preset template;

[0126] Add the preset template using the preset output specification to obtain the standard template;

[0127] Use preset constraints to constrain the standard template to obtain a finished product template;

[0128] The finished template is filled with the preprocessed data to obtain a prompt template.

[0129] In one embodiment, the filling module 104 is used to:

[0130] Encode the prompt template using an encoder to obtain digital data;

[0131] Use word embedding to map digital data to obtain vector data;

[0132] Use the decoder to encode the vector data to obtain a text sequence;

[0133] Use the table data processing library to create table data;

[0134] Use the table data processing library to sequence the text and output the user objection data.

[0135] The present invention provides an objection data processing device based on user objection, which first obtains the preliminary ranking results of candidate questions through semantic matching, and then proposes a solution for optimizing the question-answering engine based on entity alignment. The ranking results of the candidate questions are sorted again through the entity alignment, so that more matching candidate questions are selected, which can effectively avoid the generalization ability defects of the model, greatly and efficiently improve the effect of entity matching, and improve the effect of the question-answering engine.

[0136] For the specific definition of the objection data processing device based on user objection, please refer to the definition of the objection data processing method based on user objection in the above text, which will not be repeated here. Each module in the above-mentioned objection data processing device based on user objection can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0137] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external user terminal via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a service-side method for processing objection data based on user objections.

[0138] In one embodiment, a computer device is provided. The computer device may be a user terminal, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, the steps of the above-mentioned objection data processing method based on user objection are implemented.

[0139] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0140] Receive user questions, and perform question recall processing from a preset standard question library according to the user questions to obtain multiple candidate questions;

[0141] Sorting the plurality of candidate questions according to the semantic matching degree between the user question and the candidate question to obtain a first sorting result of the candidate questions;

[0142] Determine entity alignment between the user question and each of the candidate questions respectively;

[0143] According to the entity alignment, the first sorting result is adjusted to obtain a second sorting result of the candidate questions, in which the candidate questions without entity alignment are sorted behind the candidate questions with entity alignment, and the sorting of the candidate questions with entity alignment according to the first sorting result remains unchanged;

[0144] In the sequence of the second sorting results, a preset number of candidate questions are sequentially intercepted from the first ones as matching questions;

[0145] A matching answer corresponding to the matching question is selected to respond to the user question.

[0146] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0147] Receive user questions, and perform question recall processing from a preset standard question library according to the user questions to obtain multiple candidate questions;

[0148] Sorting the plurality of candidate questions according to the semantic matching degree between the user question and the candidate question to obtain a first sorting result of the candidate questions;

[0149] Determine entity alignment between the user question and each of the candidate questions respectively;

[0150] According to the entity alignment, the first sorting result is adjusted to obtain a second sorting result of the candidate questions, in which the candidate questions without entity alignment are sorted behind the candidate questions with entity alignment, and the sorting of the candidate questions with entity alignment according to the first sorting result remains unchanged;

[0151] In the sequence of the second sorting results, a preset number of candidate questions are sequentially intercepted from the first ones as matching questions;

[0152] A matching answer corresponding to the matching question is selected to respond to the user question.

[0153] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the user side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0154] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0155] 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.

[0156] It should be noted that if software tools or components other than those of our company appear in the embodiments of the present application, they are only used for illustrative purposes and do not represent actual use.

[0157] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features 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 method for processing objection data based on user objection, characterized in that: include: Acquire the target user's audio recording data, and convert the audio recording data into text data; Screening and cleaning the text data through preset attribute fields to obtain preprocessed data; Using a preset prompt model to fill in the preprocessed data to obtain a prompt template; The prompt template is transformed by using a preset large model to obtain user objection data; The user objection data is grouped in multiple levels according to preset categories to obtain grouped data, and frequency distribution of user objection data of different categories is counted based on the grouped data.

2. The objection data processing method based on user objection according to claim 1, characterized in that: The step of converting the recorded data into text data comprises: Convert the recorded data into feature vectors using preset recording parameters; Mapping the feature vector using preset phonemes to obtain a phoneme sequence; The characteristic vector is converted using a preset acoustic model to obtain a characteristic signal; The characteristic signal and the phoneme sequence are subjected to data fusion to obtain text data.

3. The objection data processing method based on user objection according to claim 1, characterized in that: The screening and cleaning of the text data by using preset attribute fields includes: Performing format conversion on the text data according to a preset format to obtain converted data; Get a multidimensional table pre-built based on preset attribute fields; The transformed data is screened and cleaned using multidimensional tables to obtain preprocessed data.

4. The objection data processing method based on user objection according to claim 1, characterized in that: The pre-processing data is processed by using a preset prompt model to obtain a prompt template, including: Establishing a basic template structure through the preset prompt model; Use the preset tasks to add the basic structure of the template to obtain the preset template; Add the preset template using the preset output specification to obtain the standard template; Use preset constraints to constrain the standard template to obtain a finished product template; The finished template is filled with the preprocessed data to obtain a prompt template.

5. The objection data processing method based on user objection according to claim 1, characterized in that: The method of processing the prompt template using the preset large model to obtain user objection data includes: Inputting the prompt template into a preset large model; the preset large model includes an encoder, a word embedding, a decoder, and a table data processing library; Encode the prompt template using an encoder to obtain digital data; Use word embedding to map digital data to obtain vector data; Use the decoder to encode the vector data to obtain a text sequence; Use the table data processing library to create table data; Use the table data processing library to sequence the text and output the user objection data.

6. The objection data processing method based on user objection according to claim 1, characterized in that: The step of grouping the user objection data in multiple levels according to preset categories to obtain grouped data, and counting the frequency distribution of user objection data of different categories according to the grouped data, includes: Grouping different question types in the user objection data to obtain question type grouping data; Grouping different specific items of the same question type in the user objection data to obtain specific item grouping data; The frequency distribution data of different objection issues are statistically analyzed based on the frequency of occurrence of specific project grouping data and problem type grouping data.

7. The objection data processing method based on user objection according to claim 1, characterized in that: After counting the frequency distribution of different categories of user objection data according to the grouped data, the method further includes: Mapping the prompt template to an SQL statement to obtain a keyword mapping; Use the prompt template to parse the keyword mapping and obtain the syntax tree; Combine the syntax tree and keyword mapping to get the query statement; Establish a user objection database using user objection data; Use the query statement to query the user objection database and obtain the query output.

8. A device for processing objection data based on user objection, characterized in that: include: The acquisition module is used to obtain the recording data of the target user; A cleaning module, used to filter and clean the text data through preset attribute fields to obtain pre-processed data; A filling module, used to fill the pre-processed data with a preset prompt model to obtain a prompt template; A conversion module, used to convert the prompt template using a preset large model to obtain user objection data; The analysis module is used to group the user objection data into multiple levels according to preset categories to obtain group data, and to calculate the frequency distribution of user objection data of different categories based on the group data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the objection data processing method based on user objection according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the objection data processing method based on user objection according to any one of claims 1 to 7 is implemented.

Images