Context verification method, device and equipment based on bubbling robot and storage medium
Through the context verification method based on bubble robots, the problem of low data processing efficiency and insufficient simulating users in the intelligent interactive system in the financial and maintenance field is solved, and the system is efficient and accurate verification in complex business scenarios is achieved, ensuring response correctness and consistency.
Patent Information
- Application Number
- CN202411235540.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-04
- Publication Date
- 2024-11-29
AI Technical Summary
In the intelligent interactive system in the field of finance and preservation, data processing efficiency is low, insufficient simulation user questions are required, and it is difficult to fully verify the system response and behavior, resulting in the impact of user experience and business processing accuracy.
Using the context verification method based on the bubble robot, the preset configuration data is exported from the configuration platform, the request parameters are extracted using scripts, and batch requests are initiated to the bubble policy interface, and the data is obtained and compared to ensure that the bubble robot passes verification. Export FAQ corpus from the knowledge base, select simulated user problems, generate context problems, use context-aware algorithms and deep learning models for semantic matching, and verify the system's response and behavior.
It realizes efficient and accurate verification of the performance of the intelligent interactive system in different business scenarios, ensures the correctness and consistency of its response, and improves the stability and user experience of the system.
Smart Images

Figure CN119046437A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of research and development management, and in particular to a context verification method, device, equipment and storage medium based on a bubbling robot. Background Art
[0002] In the financial and security fields, with the digital transformation of business processes, intelligent interactive systems (such as bubble robots) are widely used in various business scenarios to improve user experience and simplify operational processes. These systems usually rely on complex business logic and configuration data, and provide guidance, reminders, information query and other services for security business through interaction with users. However, due to the complexity of financial business, the large and diverse amount of data, and the personalization of user needs, these systems may have inconsistent or unexpected responses in actual operation, affecting user experience and the accuracy of business processing.
[0003] In the existing technology, manual testing and some automated tools are usually relied on to verify the performance of intelligent interactive systems in different business scenarios. However, these methods have several significant shortcomings:
[0004] Low data processing efficiency: The amount of data involved in the CMS platform is huge and complex. Manually processing this data and converting it into a parameter format readable by the interface is often time-consuming and labor-intensive, making it difficult to achieve batch processing.
[0005] Insufficient questions asked by simulated users: In existing technologies, it is difficult to efficiently filter out questions related to specific businesses from the knowledge base for simulation testing, resulting in insufficient test coverage and inability to fully verify the system's responsiveness.
[0006] Difficulty in verifying returned results: Existing methods for verifying returned results often lack systematicity and comprehensiveness, especially in complex security business scenarios, and are unable to effectively identify and correct possible logical errors and data inconsistencies in the system.
[0007] Therefore, in the fields of finance and security, there is an urgent need for a more automated and systematic verification method that can efficiently and accurately process large-scale data, simulate real user behavior, and comprehensively verify the system's response and behavior to ensure the stability and consistency of intelligent interactive systems in complex business scenarios. Summary of the invention
[0008] The main purpose of the present invention is to provide a context verification method, device, equipment and storage medium based on a bubble robot, aiming to solve the technical problems that the prior art lacks simulation tests for business scenarios, and the coverage of the results returned by the verification system is insufficient, making it difficult to ensure that the system's performance in complex business scenarios meets expectations.
[0009] To achieve the above object, the present invention provides a context verification method based on a bubble robot, comprising:
[0010] Export the preset configuration data of the bubble robot from the configuration platform, and use the script to extract the request parameters in the preset configuration data;
[0011] Initiate a batch request to the bubble strategy interface based on the request parameters, and obtain the data returned by the bubble strategy interface;
[0012] Compare the returned data with the preset configuration data, and if the comparison result is consistent, confirm that the bubble robot has passed the verification;
[0013] When the bubble robot passes the verification, the FAQ corpus of the security service is exported from the knowledge base platform, and the FAQ corpus contains standard questions and similar questions of the security service;
[0014] Selecting a simulated user question from the FAQ corpus, initiating a request to the intelligent interaction interface with the simulated user question as a request parameter, and obtaining the returned context question and original question;
[0015] If the context question is equal to the original question and is successfully matched to the standard question in the FAQ corpus, the context verification is confirmed to be successful;
[0016] If the context question is not equal to the original question, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question, it is confirmed that the context verification has passed.
[0017] In one embodiment, selecting simulated user questions from the FAQ corpus includes:
[0018] Analyze common user questions based on the target business scenario, extract keywords of the target business scenario from the common user questions to form a keyword group;
[0019] Using a keyword matching algorithm to traverse and search the FAQ corpus, and screen out a preset number of corpora that have the highest matching degree with at least one keyword in the keyword group;
[0020] A correlation analysis is performed between the preset number of screened corpora and the target business scenario to generate the simulated user questions for simulating actual user questions.
[0021] In one embodiment, after confirming that the bubble robot has passed the verification, the method further includes:
[0022] Collect user context information, including user operation logs, device status, geographic location, time, and user historical behavior data;
[0023] Analyze the context information in real time to identify the user's current business scenario;
[0024] Design multiple dialogue strategy templates based on the business scenario, wherein the dialogue strategy templates include response logic, prompt information display order, and related suggestion push rules;
[0025] Selecting a target template from the dialogue strategy templates according to the user context information, and dynamically adjusting the dialogue strategy in the target template based on context changes;
[0026] A plurality of context event triggering conditions are set to monitor changes in the user context information, and when the context event triggering conditions are met, the bubble robot actively initiates a dialogue based on the target template to provide prompt information or suggestions for the current scenario.
[0027] In one embodiment, a request is initiated to the intelligent interaction interface using the simulated user question as a request parameter to obtain the returned context question and the original question, including:
[0028] Passing the simulated user question to the intelligent interactive interface and integrating user context information, including user operation history, device status and geographic location;
[0029] Generate contextual questions based on the user context information and the simulated user questions using a context-aware algorithm and a deep learning model, wherein the user context information includes user historical behavior data, device status, and geographic location;
[0030] Construct semantic representations for the generated contextual questions and standard questions and similar questions in the FAQ corpus to generate high-dimensional semantic vectors;
[0031] Using a similarity algorithm to calculate the semantic similarity between the high-dimensional semantic vector of the context question and the high-dimensional semantic vector of the standard question or similar question in the FAQ corpus, and generate a similarity score;
[0032] The standard question or similar question with the highest similarity is selected according to the similarity score, and the context question and the corresponding original question and the standard question or similar question with the highest similarity returned by the intelligent interaction interface are obtained.
[0033] In one embodiment, using a context-aware algorithm and a deep learning model to generate contextual questions based on the user context information and the simulated user questions includes:
[0034] Performing word segmentation on the simulated user question and performing part-of-speech tagging to identify key words;
[0035] Extracting keywords from the simulated user question based on the part-of-speech tagging result, and calculating weights for the keywords based on their sequence positions and word frequencies in the simulated user question;
[0036] Adjust the weight of the keyword based on the frequency of occurrence of the keyword in the FAQ corpus;
[0037] Use the pre-trained word vector model to convert each keyword into a keyword vector, apply the adjusted keyword weight to the keyword vector, and generate a weighted keyword vector in a weighted manner;
[0038] Encode the user's historical behavior data into a sequence vector, the device status into a one-hot vector, and the geographic location into a floating-point vector, and combine them to form a user context information vector;
[0039] The weighted keyword vector is combined with the user context information vector to generate a comprehensive semantic vector, and the comprehensive semantic vector is input into a deep learning model to generate a final contextual question.
[0040] In one embodiment, comparing the returned data with the preset configuration data, and if the comparison result is consistent, confirming that the bubble robot has passed the verification, includes:
[0041] Extracting predefined key data fields consistent with a preset configuration data structure and format from the returned data;
[0042] Defining corresponding comparison rules according to the type of the predefined key data field, wherein the comparison rules include exact matching, tolerance matching and inclusion matching;
[0043] If the type of the predefined key data field is a business type, a page ID, or a half-screen display identifier, accurately matching the predefined key data field with the preset configuration data;
[0044] If the type of the predefined key data field is display time, duration or page height, performing tolerance matching on the predefined key data field and the preset configuration data;
[0045] If the type of the predefined key data field is message content or title, performing inclusion matching on the predefined key data field and the preset configuration data;
[0046] According to the comparison rules, the predefined key data fields in the returned data are compared with the preset configuration data item by item. If all the predefined key data fields pass the comparison, it is confirmed that the bubble robot has passed the verification.
[0047] In one embodiment, using a script to extract the request parameters in the preset configuration data includes:
[0048] Set up the script development environment, select and configure the script language, install dependent libraries, and set up the logging mechanism;
[0049] Load and parse the preset configuration data file and convert it into an operational data structure;
[0050] Determine and define the request parameter extraction rules, including page ID, business type, prompt information, display duration, and trigger conditions;
[0051] Based on the request parameter extraction rule, extract the request parameter from the corresponding node or key-value pair of the preset configuration data;
[0052] Format the extracted request parameters, unify the parameter format, store the formatted request parameters in a collection, and save them to memory or output them to a file.
[0053] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a context verification device based on a bubble robot, comprising:
[0054] A configuration data management module exports preset configuration data of the bubble robot from the configuration platform and uses a script to extract request parameters in the preset configuration data;
[0055] A bubble strategy request module initiates a batch request to the bubble strategy interface based on the request parameters, and obtains data returned by the bubble strategy interface;
[0056] A data comparison and verification module compares the returned data with the preset configuration data, and if the comparison result is consistent, confirms that the bubble robot has passed the verification;
[0057] A FAQ corpus management module, when the bubble robot passes the verification, exports the FAQ corpus of the security service from the knowledge base platform, wherein the FAQ corpus contains standard questions and similar questions of the security service;
[0058] A user question simulation and interaction module selects simulated user questions from the FAQ corpus, initiates a request to the intelligent interaction interface using the simulated user questions as request parameters, and obtains returned context questions and original questions;
[0059] A context verification module confirms that the context verification has passed if the context question is equal to the original question and is successfully matched to a standard question in the FAQ corpus;
[0060] The similar question matching module confirms that the context verification has passed if the context question and the original question are not equal, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question.
[0061] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a context verification device based on a bubble robot, wherein the context verification device based on a bubble robot comprises a memory, a processor, and a context verification program based on a bubble robot stored in the memory and executable on the processor, wherein the context verification program based on a bubble robot, when executed by the processor, implements the steps of the context verification method based on a bubble robot as described above.
[0062] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer storage medium, on which a context verification program based on a bubble robot is stored, and when the context verification program based on a bubble robot is executed by a processor, the steps of the context verification method based on a bubble robot as described above are implemented.
[0063] Beneficial effects: The present invention relates to a context verification method based on a bubble robot, which exports preset configuration data related to the bubble robot from a configuration platform and uses a script to extract key request parameters. Based on these parameters, a batch request is initiated to the bubble strategy interface to obtain the returned data. By comparing the returned data with the preset configuration data, it is confirmed whether the bubble robot has passed the verification. After the verification is passed, the FAQ corpus related to the security business is exported from the knowledge base platform, and simulated user questions are selected and used as request parameters to initiate a request to the intelligent interaction interface to obtain and analyze the returned context questions and original questions. If the context question and the original question match successfully, or a similar question with a score exceeding a preset threshold is found in the FAQ corpus, it is confirmed that the context verification has passed. The present invention can efficiently and accurately verify the performance of the intelligent interaction system in different business scenarios, and ensure the correctness and consistency of its response. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:
[0065] Figure 1 It is a flow chart of an embodiment of a context verification method based on a bubble robot of the present invention;
[0066] Figure 2 It is a functional module diagram of a preferred embodiment of a context verification device based on a bubble robot of the present invention;
[0067] Figure 3 It is a structural schematic diagram of the device hardware operating environment involved in the embodiment scheme of the context verification device based on the bubble robot of the present invention. DETAILED DESCRIPTION
[0068] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0069] It should be noted that in the financial and security fields, with the digital transformation of business processes, intelligent interactive systems (such as bubble robots) are widely used in various business scenarios to improve user experience and simplify operational processes. These systems usually rely on complex business logic and configuration data, and provide guidance, reminders, information query and other services for security business through interaction with users. However, due to the complexity of financial business, the large and diverse amount of data, and the personalization of user needs, these systems may have inconsistent or unexpected responses in actual operation, affecting user experience and the accuracy of business processing.
[0070] In the existing technology, manual testing and some automated tools are usually relied on to verify the performance of intelligent interactive systems in different business scenarios. However, these methods have several significant shortcomings:
[0071] Low data processing efficiency: The amount of data involved in the CMS platform is huge and complex. Manually processing this data and converting it into a parameter format readable by the interface is often time-consuming and labor-intensive, making it difficult to achieve batch processing.
[0072] Insufficient questions asked by simulated users: In existing technologies, it is difficult to efficiently filter out questions related to specific businesses from the knowledge base for simulation testing, resulting in insufficient test coverage and inability to fully verify the system's responsiveness.
[0073] Difficulty in verifying returned results: Existing methods for verifying returned results often lack systematicity and comprehensiveness, especially in complex security business scenarios, and are unable to effectively identify and correct possible logical errors and data inconsistencies in the system.
[0074] Therefore, in the fields of finance and security, there is an urgent need for a more automated and systematic verification method that can efficiently and accurately process large-scale data, simulate real user behavior, and comprehensively verify the system's response and behavior to ensure the stability and consistency of intelligent interactive systems in complex business scenarios.
[0075] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a context verification method based on a bubble robot provided by the present invention. It should be noted that although a logical sequence is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than that here.
[0076] like Figure 1 As shown, the context verification method based on the bubble robot proposed in the present invention includes the following steps:
[0077] S10, exporting preset configuration data of the bubble robot from the configuration platform, and using a script to extract request parameters in the preset configuration data;
[0078] In this embodiment, during the configuration process of the intelligent interactive system, the configuration platform (such as a CMS platform) stores various configuration data related to the bubble robot. These data include key parameters such as business scenarios, page IDs, message content, and display time, which are used to guide the behavior of the bubble robot. In order to verify the correctness of the bubble robot, it is necessary to export these preset configuration data from the configuration platform. After exporting, scripts are used to automatically extract request parameters related to verification. These parameters will be used as inputs for the bubble strategy interface request to ensure the efficiency and accuracy of the test process.
[0079] Export the preset configuration data from the configuration platform (CMS platform) to obtain all key parameters related to the behavior of the bubble robot. Write a script to automatically extract the request parameters in the exported data. The script will parse the exported data file, find the relevant configuration items, and extract them into a structured data format for subsequent batch requests. Based on the extracted request parameters, a series of bubble strategy interface requests are generated. These requests will be used to batch test the response and behavior of the bubble robot to verify whether it complies with the preset configuration.
[0080] In another specific implementation, when exporting preset configuration data related to the bubble robot, the focus is on extracting user interaction history and feedback data, and multi-level data extraction and preprocessing are achieved through complex scripts. The script will automatically classify and extract request parameters for different scenarios based on the user's behavior patterns. For example, when processing a contract change business, the script not only extracts the page ID and display time, but also generates personalized request parameters based on the user's past change frequency and preferences. These parameters are then used to simulate the user's actual operations, and the accuracy and consistency of the returned data are verified through a dynamic comparison algorithm, including the degree of personalization of the message, the agility of the response, etc. If the verification is successful, it is confirmed that the bubble robot can flexibly respond to the personalized needs of different users.
[0081] By extracting request parameters through automated scripts, the efficiency of data processing is significantly improved, the probability of manual operation errors is reduced, and the accuracy and consistency of batch testing are ensured. It can effectively support large-scale data processing and system verification, and provide a reliable verification tool for intelligent interactive systems in complex business scenarios.
[0082] S20, initiating a batch request to the bubble strategy interface based on the request parameters, and obtaining data returned by the bubble strategy interface;
[0083] In this embodiment, in the intelligent interactive system, the bubble strategy interface is responsible for returning corresponding bubble prompts or guidance information according to the incoming request parameters. The request parameters extracted by the script may include key information such as business type, page ID, display time, etc. These parameters will be used to generate multiple requests and sent to the bubble strategy interface in batches. The purpose of batch requests is to test the system response under multiple scenarios or different configurations at the same time to ensure that the bubble robot can work as expected in a variety of business environments. The return data obtained includes information such as bubble prompt content, display duration, page response, etc., which will be further used for comparison and verification.
[0084] First, based on the request parameters extracted from the CMS platform, multiple API requests are automatically generated through scripts. These requests may involve different business types, pages, or display configurations. The generated requests are sent in batches to the bubble strategy interface through automated tools or scripts. Batch requests can be processed in parallel to improve test efficiency, especially in high-concurrency environments. The returned data is obtained from the interface. The returned data includes bubble prompts, display time, page ID, etc. These data will be used in subsequent verification steps to check whether the system responds correctly according to the preset configuration.
[0085] In other specific embodiments, batch requests are not only based on static configuration parameters, but also combined with the output of the prediction model. For example, a machine learning model is used to predict the user's possible operation path in the next few minutes, and the corresponding request parameters are generated based on the prediction results. These parameters include the expected order of page visits, possible interaction frequency, and expected changes in user needs. Batch requests are sent to the bubble strategy interface, and the returned data will be used to compare the match between the prediction results and the actual response, and evaluate the foresight and accuracy of the bubble strategy. This implementation method can identify potential deviations in the system when predicting user behavior in advance and improve the accuracy of the model.
[0086] By sending requests in batches and obtaining data, the system testing efficiency can be greatly improved and the workload of manual testing can be reduced. It can not only cover a variety of business scenarios, but also quickly discover potential problems of the system under high concurrency conditions, thereby improving the stability and reliability of the system.
[0087] S30, comparing the returned data with the preset configuration data, and if the comparison result is consistent, confirming that the bubble robot has passed the verification;
[0088] In this embodiment, by comparing the data returned from the bubble strategy interface with the preset configuration data exported from the CMS configuration platform, it is possible to verify whether the bubble robot is operating correctly according to the predetermined business rules and configuration. The comparison process involves checking each data field item by item, such as business type, page ID, message content, display time, etc. The comparison rules are set according to the characteristics of the data field. For example, some fields require an exact match, while other fields may allow a certain tolerance or include a match. Only when the comparison results of all fields are as expected, the bubble robot is confirmed to have passed the verification.
[0089] First, export the preset configuration data related to the bubble robot from the CMS configuration platform, and then obtain the returned data from the bubble strategy interface. Define the matching rules according to the type of data field. For example, fields such as business type and page ID need to be matched exactly, while fields such as display time can have a certain tolerance range. Compare the returned data with the preset configuration data item by item, and record the comparison results of each field. If all comparison items meet the preset rules, the bubble robot is confirmed to have passed the verification; otherwise, it is necessary to analyze the cause of the failure and retest.
[0090] In another specific implementation, in a multi-step verification scenario of a security service, the returned data includes a sequence of steps, operation results, and user feedback. The system requires sequence matching, that is, the returned data must contain a preset sequence of operation steps, and the order of these steps cannot be changed. In addition to accurately matching basic fields, the system also uses feedback consistency comparison to verify whether the user's operation is correctly guided to the expected conclusion by comparing the user's real-time feedback with the system's expected feedback results. This comparison method ensures that the bubble robot can accurately guide users when processing multi-step business operations and adjust subsequent operations based on user feedback.
[0091] Through the above steps, we can ensure that the bubble robot operates strictly according to the preset configuration. This not only improves the accuracy and consistency of the system, but also effectively reduces possible operation deviations and errors, and enhances the stability of the system.
[0092] S40, when the bubble robot passes the verification, the FAQ corpus of the security service is exported from the knowledge base platform, wherein the FAQ corpus includes standard questions and similar questions of the security service;
[0093] In this embodiment, in the verification process of the intelligent interactive system, after confirming that the bubble robot has passed the verification (that is, its behavior is consistent with the preset configuration data), the next step is to further test the system's contextual understanding and user interaction capabilities. This step involves exporting FAQ corpora related to security services from the knowledge base platform. FAQ corpora usually contain two main parts: standard questions (typical user questions preset by the system) and similar questions (other question expressions that are semantically similar to standard questions). By using these corpora, the system can further verify the semantic matching and answer accuracy of the bubble robot when processing user questions, ensuring its effectiveness in practical applications.
[0094] Select FAQ corpora related to security services from the knowledge base platform and export them. The corpus contains standard questions and their corresponding similar questions, which cover various query forms that users may ask. The exported FAQ corpora are classified according to question type and business scenario to ensure that each standard question has a corresponding set of similar questions. This provides preparation for subsequent testing. These exported FAQ corpora will be used to simulate user questions. By comparing the matching degree between standard questions and similar questions, the system's performance in processing complex semantics and diverse user questions can be verified.
[0095] By exporting and using FAQ corpus containing standard questions and similar questions, the system can more comprehensively test the performance of the bubble robot in semantic understanding and user interaction. This not only enhances the system's context matching ability, but also ensures accuracy and consistency in diverse user questioning scenarios.
[0096] S50, selecting a simulated user question from the FAQ corpus, initiating a request to the intelligent interaction interface with the simulated user question as a request parameter, and obtaining the returned context question and original question;
[0097] In this embodiment, in the verification process of the intelligent interactive system, in order to test the performance of the system in actual user interaction, it is necessary to select simulated user questions from the exported FAQ corpus. The simulated user questions are generated based on standard questions and similar questions in the FAQ, with the purpose of simulating questions that real users may ask when using the system. These simulated user questions will be sent to the intelligent interactive interface as request parameters to test the system's responsiveness in different contexts. By obtaining the returned contextual questions (user questions understood by the system in context) and the original questions (the user's initial input), the semantic matching and response accuracy of the system can be evaluated to ensure that the system can correctly understand and answer user questions.
[0098] Filter out the standard questions and similar questions that are most relevant to the current business scenario from the FAQ corpus and generate simulated user questions. These questions are designed to cover possible user questioning methods to comprehensively test the system's semantic understanding capabilities. Send the selected simulated user questions as request parameters to the intelligent interaction interface. Requests can be sent one at a time or in batches to test the system's performance under high concurrency. Get the returned contextual questions and original questions from the intelligent interaction interface. Contextual questions are the system's understanding of user questions in a specific context, while original questions are the user's initial input. By comparing these two questions, you can evaluate whether the system correctly understands and responds to the user's questions.
[0099] By selecting simulated user questions and sending them to the intelligent interaction interface, the system can simulate real user interaction scenarios in the test environment and comprehensively evaluate the system's semantic matching capabilities and response accuracy. It can effectively discover potential problems in the system's semantic understanding and ensure that the system can accurately handle various complex user questions in actual applications.
[0100] S60, if the context question is equal to the original question and is successfully matched to a standard question in the FAQ corpus, confirming that the context verification has passed;
[0101] In this embodiment, during the context verification process of the intelligent interactive system, the system needs to ensure that it can accurately understand the user's input and match it to the preset FAQ corpus. Context questions are the understanding of user questions generated by the system based on the current context during the interaction with the user. If the context question is equal to the user's original question (i.e., the user's initial input) and is successfully matched to the standard question in the FAQ corpus, this indicates that the system is accurate in semantic understanding and question matching, so that it can be confirmed that the system has passed the context verification. This process is a key step in verifying whether the system can correctly understand and respond to user needs.
[0102] The system generates contextual questions based on the user's original question and the context of the current interaction. The contextual questions reflect the system's understanding of the user's question. The system matches the generated contextual questions with the standard questions in the FAQ corpus to check whether a standard question that is highly similar to the contextual question can be found. If the contextual question is equal to the original question and successfully matches the standard question, the context verification is confirmed to have passed and the system can correctly understand and respond to the user's question.
[0103] In another specific implementation, when matching standard questions in the FAQ corpus, the system will consider the contextual history, including the user's previous questions and interaction records. If the system finds that the current contextual question is related to the previous interaction history, it will prioritize matching those standard questions related to the user's history. For example, in a continuous conversation, the user may first ask "How do I modify the terms of the contract?" and then ask "What materials are needed?" The system will contextualize the second question as "What materials are needed to modify the terms of the contract?" based on contextual understanding, and match it with the corresponding standard question. This approach ensures that the system can maintain a coherent understanding of the context in a complex conversation process.
[0104] By comparing contextual questions with original questions and matching them with standard questions, the system can effectively verify its semantic understanding ability and question response accuracy in actual applications. This verification ensures that the system can provide accurate answers to questions from diverse users and improve the user experience.
[0105] S70: If the context question is not equal to the original question, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question, confirm that the context verification is passed.
[0106] In this embodiment, during the context verification process of the intelligent interactive system, there is a situation where the context question generated by the system is not exactly the same as the user's original question in terms of literal or structural terms. In this case, the system will further check whether the context question can be matched to a similar question in the FAQ corpus with a score exceeding a preset threshold. The preset threshold is usually set by the system to ensure that the matched similar questions are semantically close enough to the user's original question. If the matching score exceeds this threshold, it means that the system performs well in semantic understanding and matching, and the context verification can still be confirmed to have passed. This process is to enhance the robustness of the system in dealing with complex or diverse expressions, and to ensure that the system can flexibly respond to different user questions.
[0107] The system generates a contextual question based on the user's original question and context, which may not be exactly the same as the original question. The system matches the generated contextual question with similar questions in the FAQ corpus and calculates the matching score. If the matching score exceeds the threshold preset by the system, the match is considered successful. Even if the contextual question is not exactly the same as the original question, as long as the matching score exceeds the preset threshold, the system can confirm that the context verification has passed. This ensures that the system can maintain efficient semantic understanding and accuracy when processing questions in different expressions.
[0108] In another specific embodiment, the system not only considers the matching scores of contextual questions and similar questions in the FAQ corpus, but also introduces user sentiment analysis. In the case where the contextual question is not equal to the original question, the system will analyze the emotional state of the user when asking the question, such as nervousness, confusion, etc. If the system detects that the user is more emotional when asking the question, it will lower the matching threshold to match similar questions more loosely, thereby ensuring that the user's needs can be met. For example, in a contract change scenario, the user may ask "How can I change the contract immediately?" The contextual question generated by the system is "How do I change the terms of the contract?" At this time, the system detects that the user is in a state of urgency, so it lowers the matching threshold from 85% to 75% to ensure that relevant similar questions are matched. If the matching score exceeds the adjusted threshold, the system confirms that the context verification has passed.
[0109] By allowing similar matches between contextual questions and original questions even when they are not exactly the same, the system significantly enhances its ability to handle diverse expressions. This ensures that the system can still provide accurate and consistent responses when faced with complex contexts or non-standard questions, thus improving the system's adaptability and user experience.
[0110] The present invention relates to a context verification method based on a bubbling robot, which exports preset configuration data related to the bubbling robot from a configuration platform and uses a script to extract key request parameters. Based on these parameters, a batch request is initiated to a bubbling strategy interface to obtain the returned data. By comparing the returned data with the preset configuration data, it is confirmed whether the bubbling robot has passed the verification. After the verification is passed, the FAQ corpus related to the security business is exported from the knowledge base platform, and simulated user questions are selected and used as request parameters to initiate a request to the intelligent interaction interface to obtain and analyze the returned context questions and original questions. If the context question and the original question match successfully, or a similar question with a score exceeding a preset threshold is found in the FAQ corpus, it is confirmed that the context verification has passed. The present invention can efficiently and accurately verify the performance of the intelligent interaction system in different business scenarios, and ensure the correctness and consistency of its response.
[0111] In one embodiment, in the above S50, selecting simulated user questions from the FAQ corpus includes:
[0112] S501, analyzing common questions asked by users based on target business scenarios, extracting keywords of the target business scenarios from the common questions asked by users to form keyword groups;
[0113] S502, using a keyword matching algorithm to traverse and search the FAQ corpus, and screen out a preset number of corpora that have the highest matching degree with at least one keyword in the keyword group;
[0114] S503, performing correlation analysis on the preset number of screened corpora and the target business scenario, and generating the simulated user questions for simulating actual user questions.
[0115] In this embodiment, in the process of selecting simulated user questions in the system, the common questions asked by users are first analyzed according to specific business scenarios (such as insurance policy loans or contract changes). These questions often contain keywords that are closely related to the business scenario. By analyzing the content of these common questions, the system can extract multiple keywords that reflect the core of the business scenario, and group these keywords into keyword groups as the basis for subsequent retrieval and screening.
[0116] After forming the keyword group, the system uses the keyword matching algorithm to traverse and search the FAQ corpus. The keyword matching algorithm analyzes each question in the FAQ and calculates its matching degree with each keyword in the keyword group. Based on the matching degree, the system will filter out the first preset number of corpora with the highest matching degree with at least one keyword in the keyword group. These corpora will serve as preliminary screening results for further analysis.
[0117] After selecting the corpora with high matching degree, the system will conduct correlation analysis between these corpora and the target business scenarios. This analysis aims to ensure that the selected corpora not only match the keywords, but are also closely related to the business scenarios in terms of semantics and content. Through this analysis, the system will eventually generate simulated user questions that can reflect the actual questions asked by users for subsequent system testing and verification.
[0118] In one specific implementation, in the policy loan business scenario, the system first collects common questions from users, such as "What is the loan interest rate?", "How to repay?", etc. Different from the traditional method, this time the system not only extracts explicit keywords (such as "loan", "interest rate") in common questions, but also identifies hidden intent keywords (such as "fast" may imply that users want to know a faster loan method) through natural language processing technology. The formed keyword group includes explicit and intent keywords. Using the keyword matching algorithm, the system traverses the FAQ corpus and filters out the top 10 corpora with the highest match with at least one keyword in these keyword groups. In the filtered corpus, the system further uses contextual relevance analysis, combined with the user's historical behavior and current operation steps, to determine the most representative simulated user questions. For example, the final generated question may be "How to quickly apply for a policy loan?" or "What is the interest rate for a policy loan?".
[0119] Through the above steps, this embodiment can effectively select highly relevant simulated user questions from the FAQ corpus, which not only improves the pertinence and effectiveness of the test, but also ensures that the system can provide accurate and reliable responses when processing actual user questions.
[0120] In one embodiment, in the above S30, after confirming that the bubble robot has passed the verification, the following step is further included:
[0121] S301, collecting user context information, including user operation logs, device status, geographic location, time, and user historical behavior data;
[0122] S302, analyzing the context information in real time to identify the user's current business scenario;
[0123] S303, designing multiple dialogue strategy templates based on the business scenario, wherein the dialogue strategy templates include response logic, prompt information display order, and related suggestion push rules;
[0124] S304, selecting a target template from the dialog strategy templates according to the user context information, and dynamically adjusting the dialog strategy in the target template based on context changes;
[0125] S305, setting multiple context event triggering conditions, monitoring changes in the user context information, and when the context event triggering conditions are met, actively initiating a dialogue based on the target template through the bubble robot to provide prompt information or suggestions for the current scenario.
[0126] In this embodiment, after the system confirms that the bubble robot has passed the preliminary verification, the next step is to gain a deeper understanding of the user's actual operating environment and behavior. User context information includes multiple dimensions, such as the user's operation log in the application, the hardware status of the device, the user's current geographic location, the time period of using the application, and the user's past behavior patterns. This information will be collected and stored for subsequent analysis and decision-making.
[0127] After collecting user context information, the system will analyze this data in real time to identify the business scenario the user is currently in. For example, by analyzing the user's geographic location and operation log, the system can determine whether the user is processing specific business such as insurance policy loans or contract changes. Identifying the current business scenario is a key step to ensure that the system can provide targeted dialogue strategies.
[0128] Once the user's business scenario is identified, the system will design multiple dialogue strategy templates based on different business needs. These templates contain detailed response logic (i.e. how the system responds to different questions from users), the display order of prompt information (i.e. what information the system should display and when), and the push rules for related suggestions (i.e. what operation suggestions the system should provide to users).
[0129] The system will select the most appropriate target template from the designed dialogue strategy templates based on the user's context information. As the user's context information changes in real time, the system will also dynamically adjust the dialogue strategy in the target template to ensure that the system's response always matches the user's current needs and environment.
[0130] The system sets multiple context event trigger conditions and continuously monitors changes in user context information. When certain conditions are triggered, the system will proactively initiate a conversation through a bubble robot, providing users with prompt information or suggestions related to the current business scenario based on the selected target template. This proactively initiated conversation can help users complete operations more efficiently and improve user experience.
[0131] By collecting and analyzing the user's context information, this embodiment can provide a more accurate and personalized interactive experience in different business scenarios. The dynamic adjustment of the dialogue strategy and the active dialogue mechanism based on trigger conditions ensure that the system is more flexible and intelligent in responding to the diverse needs of users, thereby improving user satisfaction.
[0132] In one embodiment, in the above S50, a request is initiated to the intelligent interaction interface using the simulated user question as a request parameter to obtain the returned context question and the original question, including:
[0133] S504, passing the simulated user question to the intelligent interactive interface, and integrating user context information, including user operation history, device status, and geographic location;
[0134] S505, using a context-aware algorithm and a deep learning model to generate a contextual question based on the user context information and the simulated user question, wherein the user context information includes user historical behavior data, device status, and geographic location;
[0135] S506, constructing semantic representations for the generated contextual questions and standard questions and similar questions in the FAQ corpus to generate high-dimensional semantic vectors;
[0136] S507, using a similarity algorithm to calculate the semantic similarity between the high-dimensional semantic vector of the context question and the high-dimensional semantic vector of the standard question or similar question in the FAQ corpus, and generate a similarity score;
[0137] S508, selecting the standard question or similar question with the highest similarity according to the similarity score, and obtaining the context question and the corresponding original question and the standard question or similar question with the highest similarity returned by the intelligent interaction interface.
[0138] In this embodiment, during the system verification process, it is first necessary to pass the simulated user questions selected from the FAQ corpus to the intelligent interaction interface. At the same time, the system will also integrate the user's contextual information, such as the user's operation history (such as the user's previous operation path and usage habits), current device status (such as power, network connection, etc.) and geographic location (such as whether the user is in a specific place). This contextual information will be passed to the intelligent interaction interface together with the simulated user questions to ensure that the generated contextual questions can accurately reflect the user's current needs and environment.
[0139] After receiving the simulated user question and context information, the system uses context-aware algorithms and deep learning models to generate contextual questions. The context-aware algorithm adjusts its understanding of the simulated user question by analyzing context information such as the user's historical behavior data, device status, and geographic location, and then generates a contextual question that is more in line with the current context. The deep learning model is responsible for learning and identifying complex semantic relationships from large amounts of data to ensure that the generated contextual questions can accurately reflect the user's intentions.
[0140] After generating context questions, the system will construct a semantic representation of the context questions, that is, convert the context questions into high-dimensional semantic vectors. This process uses natural language processing technology to represent context questions, standard questions in FAQ corpus, and similar questions as points in a high-dimensional vector space. High-dimensional semantic vectors can capture the semantic relationship between questions, which is convenient for subsequent similarity calculations.
[0141] Next, the system uses a similarity algorithm to calculate the semantic similarity between the high-dimensional semantic vector of the generated contextual question and the high-dimensional semantic vector of the standard question or similar question in the FAQ corpus. The similarity score reflects the degree of semantic similarity between the two questions, and the higher the score, the closer the semantics of the two questions are. This step ensures that the system can identify the FAQ question that is most relevant to the contextual question.
[0142] Finally, the system selects the standard question or similar question that is most similar to the context question based on the similarity score. The system returns these results, including the generated context question, the user's original question, and the standard question or similar question with the highest similarity. This process ensures that the system can provide users with accurate answers and improve the user experience.
[0143] This embodiment combines user context information with a deep learning model to generate contextual questions, and uses high-dimensional semantic vectors and similarity algorithms for matching. The system can more accurately understand the user's intentions and provide highly relevant FAQ answers. This enhances the system's ability to understand semantics in complex situations and can provide accurate responses when faced with diverse user needs.
[0144] In one embodiment, in the above S505, using a context-aware algorithm and a deep learning model to generate a context question based on the user context information and the simulated user question includes:
[0145] S5051, performing word segmentation processing on the simulated user question and performing part-of-speech tagging to identify key words;
[0146] S5052, extracting keywords from the simulated user question based on the part-of-speech tagging result, and calculating weights for the keywords based on their sequence positions and word frequencies in the simulated user question;
[0147] S5053, adjusting the weight of the keyword based on the frequency of occurrence of the keyword in the FAQ corpus;
[0148] S5054, using a pre-trained word vector model to convert each keyword into a keyword vector, applying the adjusted keyword weight to the keyword vector, and generating a weighted keyword vector in a weighted manner;
[0149] S5055, encode the user's historical behavior data into a sequence vector, encode the device state into a one-hot vector, encode the geographic location into a floating point vector, and combine them to form a user context information vector;
[0150] S5056, combining the weighted keyword vector with the user context information vector to generate a comprehensive semantic vector, and inputting the comprehensive semantic vector into a deep learning model to generate a final contextual question.
[0151] In this embodiment, in the process of generating contextual questions, the system first performs word segmentation on the simulated user questions. This step breaks down the text in the question into individual words or phrases. Next, the system performs part-of-speech tagging on these words, that is, identifying the grammatical function of each word, such as noun, verb, adjective, etc. This tagging process helps to identify the keywords in the question, which will play an important role in the subsequent steps.
[0152] The system extracts keywords from simulated user questions through part-of-speech tagging results. These keywords are usually the core content of the question. Then, the system calculates the weight of the keyword based on its sequence position in the question (i.e. the relative position of the keyword in the sentence) and word frequency (the number of times the keyword appears in the question). The higher the weight, the more important the keyword is in the question.
[0153] In order to further refine the weight of keywords, the system also refers to the frequency of occurrence of these keywords in the FAQ corpus. Keywords with high frequency of occurrence may be important in the FAQ corpus, so the system will adjust the weight of these keywords accordingly to ensure that their importance is taken into account when generating contextual questions.
[0154] The system uses a pre-trained word vector model to convert each keyword into a vector representation (i.e., keyword vector). These vectors represent the semantic relationship between words in a high-dimensional space. Next, the system applies the weights adjusted in the previous step to these keyword vectors and generates weighted keyword vectors in a weighted manner. Weighted keyword vectors can capture the importance of keywords in context and provide a basis for subsequent semantic calculations.
[0155] The system encodes the user's historical behavior data into sequence vectors, which reflect the user's past operations and behavior patterns. Device status (such as battery level, network connection, etc.) is encoded as a one-hot vector, which is a simple but effective encoding method that represents the various possibilities of device status. The geographic location is encoded as a floating-point vector to accurately reflect the user's geographic location. Finally, the system combines these different types of vectors to form a complete user context information vector that can fully reflect the user's current status and historical behavior.
[0156] Finally, the system combines the generated weighted keyword vector with the user context information vector to form a comprehensive semantic vector. This comprehensive semantic vector captures the core content of the simulated user question and its related context information. The system inputs the comprehensive semantic vector into the pre-trained deep learning model to generate the final contextual question. This contextual question fully considers the user's historical behavior, device status, and geographic location to ensure that the system understands the user's intent more accurately.
[0157] By combining keyword weights, user context information and deep learning models, this embodiment enables the system to generate more accurate and contextual questions. This not only improves the system's ability to understand semantics in complex situations, but also ensures that each question raised by the user can receive a highly relevant response, thereby improving the overall user experience.
[0158] In one embodiment, the above S30 includes:
[0159] S306, extracting predefined key data fields consistent with the preset configuration data structure and format from the returned data;
[0160] S307, defining a corresponding comparison rule according to the type of the predefined key data field, wherein the comparison rule includes an exact match, a tolerance match, and an inclusion match;
[0161] S308, if the type of the predefined key data field is a business type, a page ID, or a half-screen display identifier, accurately matching the predefined key data field with the preset configuration data;
[0162] S309, if the type of the predefined key data field is display time, duration or page height, performing tolerance matching on the predefined key data field and the preset configuration data;
[0163] S310, if the type of the predefined key data field is message content or title, perform inclusion matching on the predefined key data field and the preset configuration data;
[0164] S311, according to the comparison rules, the predefined key data fields in the returned data are compared with the preset configuration data item by item. If all the predefined key data fields pass the comparison, it is confirmed that the bubble robot has passed the verification.
[0165] In this embodiment, during the system verification process, predefined key data fields that are consistent with the preset configuration data structure and format are first extracted from the data returned by the system. Predefined key data fields refer to data items that are focused on during the verification process. These fields must be consistent with the preset configuration data in structure and format to ensure the accuracy of data comparison. The purpose of this step is to screen out data items that have a key impact on the verification results.
[0166] Once the predefined key data fields are extracted, the system will define the corresponding matching rules according to the field type. The matching rules are the standards used to check whether the returned data is consistent with the preset configuration data. Specifically, there are three types of matching rules: exact match, tolerance match, and inclusion match. Exact match requires that the data is completely consistent, tolerance match allows slight differences in the data within a certain range, and inclusion match checks whether the returned data contains the content of the preset configuration data.
[0167] For certain predefined key data fields, such as business type, page ID, and half-screen display logo, the system requires an exact match. Exact match means that the system compares the returned data with the preset configuration data word by word to ensure that these data items are completely consistent in content. This is because these fields play a decisive role in the verification process, and any difference may affect the accuracy of the verification.
[0168] For some other fields, such as display time, duration, and page height, the system uses tolerance matching. Tolerance matching allows the returned data to have slight differences within a certain range, and these differences usually do not affect the overall function of the system. For example, a slight change in page height may be caused by differences in display effects on different devices, so a certain tolerance range is allowed.
[0169] For message content or title, the system uses inclusion matching. Inclusion matching checks whether the returned data contains keywords or phrases in the preset configuration data, but does not require exact consistency. This matching method allows for a certain degree of flexibility in information expression. For example, the expression of the title may vary depending on the specific scenario, but as long as the key content is consistent, the match is considered successful.
[0170] Finally, the system compares the predefined key data fields in the returned data with the preset configuration data one by one according to the comparison rules defined above. Only when all key data fields pass the comparison will the system confirm that the bubble robot has passed the verification. This verification process ensures that the various functions of the system are consistent with the expected settings in actual applications, improving the reliability and stability of the system.
[0171] This embodiment uses multiple comparison rules such as exact matching, tolerance matching, and inclusion matching, so that the system can more flexibly and accurately verify the consistency of the returned data with the preset configuration data. This multi-level comparison method ensures that the system can maintain a high degree of accuracy in different situations, while allowing reasonable changes within a certain range, thereby improving the system's adaptability and fault tolerance.
[0172] In one embodiment, in the above S10, using a script to extract the request parameters in the preset configuration data includes:
[0173] S101, set up the script development environment, select and configure the script language, install dependent libraries, and set up the logging mechanism;
[0174] S102, loading and parsing a preset configuration data file, and converting it into an operable data structure;
[0175] S103, determining and defining request parameter extraction rules, including page ID, business type, prompt information, display duration, and trigger conditions;
[0176] S104, extracting the request parameter from a corresponding node or key-value pair of preset configuration data based on the request parameter extraction rule;
[0177] S105, formatting the extracted request parameters, unifying the parameter format, storing the formatted request parameters in a collection, and saving them to a memory or outputting them to a file.
[0178] In this embodiment, before starting to write a script to extract request parameters, you first need to set up a script development environment. Developers should choose a suitable scripting language (such as Python, JavaScript, etc.) according to project requirements and configure the corresponding development environment. This includes installing necessary dependent libraries (such as libraries for processing JSON, XML or other data formats), and setting up a logging mechanism. The logging mechanism is used to capture and record key operations, error messages, and operating status when the script is running, for debugging and subsequent analysis.
[0179] Once the development environment is set up, the first step of the script is to load the preset configuration data files. These files may be stored in JSON, XML, or other structured data formats. The script will parse these files and convert the data into actionable data structures (such as dictionaries, lists, tree structures, etc.) for subsequent processing. The parsing operation extracts the information in the file and makes it easy for the script to operate and extract data.
[0180] Next, the script needs to determine and define the request parameter extraction rules based on business needs. The extraction rules define which data items need to be extracted from the preset configuration data, such as page ID, business type, prompt information, display duration, trigger conditions, etc. These rules can be implemented by specifying data nodes, key-value pairs, or paths to ensure that the script can accurately find and extract the required request parameters.
[0181] The script will traverse the structure of the preset configuration data according to the previously defined extraction rules and extract the request parameters from the corresponding nodes or key-value pairs. This step ensures that the script can accurately obtain all the required parameters in the configuration file and extract them for subsequent use.
[0182] After extracting the request parameters, the script needs to format them. The purpose of formatting is to unify the format of the parameters, such as data type, date format, string case, etc., to ensure the consistency and compatibility of these parameters in subsequent operations. The formatted request parameters will be stored in a collection (such as a list, array, or dictionary), and can be saved in memory for subsequent processing, or output to a file for archiving or passing to other system components.
[0183] This embodiment extracts the request parameters in the preset configuration data in a scripted manner, so that the system can automatically and accurately obtain the required configuration information and ensure the consistency and standardization of the data. This automated process reduces the risk of errors in manual operations and improves the efficiency and reliability of data processing. In addition, the flexibility of the script enables it to adapt to different data formats and structures, enhancing the adaptability and scalability of the system.
[0184] The present invention also provides a context verification device based on a bubble robot, referring to Figure 2 , Figure 2 The functional module diagram of a preferred embodiment of the context verification device based on the bubble robot of the present invention is shown in FIG. The context verification device based on the bubble robot comprises:
[0185] A configuration data management module exports preset configuration data of the bubble robot from the configuration platform and uses a script to extract request parameters in the preset configuration data;
[0186] A bubble strategy request module initiates a batch request to the bubble strategy interface based on the request parameters, and obtains data returned by the bubble strategy interface;
[0187] A data comparison and verification module compares the returned data with the preset configuration data, and if the comparison result is consistent, confirms that the bubble robot has passed the verification;
[0188] A FAQ corpus management module, when the bubble robot passes the verification, exports the FAQ corpus of the security service from the knowledge base platform, wherein the FAQ corpus contains standard questions and similar questions of the security service;
[0189] A user question simulation and interaction module selects simulated user questions from the FAQ corpus, initiates a request to the intelligent interaction interface using the simulated user questions as request parameters, and obtains returned context questions and original questions;
[0190] A context verification module confirms that the context verification has passed if the context question is equal to the original question and is successfully matched to a standard question in the FAQ corpus;
[0191] The similar question matching module confirms that the context verification has passed if the context question and the original question are not equal, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question.
[0192] The specific implementation of the context verification device based on the bubble robot of the present invention is basically the same as the above-mentioned embodiments of the context verification method based on the bubble robot, and will not be repeated here.
[0193] The present invention also provides a context verification device based on a bubble robot, such as Figure 3As shown, the context verification device based on the bubble robot may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0194] Those skilled in the art will understand that Figure 3 The hardware structure of the context verification device based on the bubble robot shown in the figure does not constitute a limitation of the context verification device based on the bubble robot, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.
[0195] like Figure 3 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a context verification program based on a bubble robot. The operating system is a program for managing and controlling a context verification device and software resources based on a bubble robot, and supports the operation of the network communication module, the user interface module, the context verification program based on a bubble robot, and other programs or software; the network communication module is used to manage and control the network interface 1004; the user interface module is used to manage and control the user interface 1003.
[0196] exist Figure 3 In the hardware structure of the bubble robot-based context verification device shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client and communicate data with the client; the processor 1001 can call the bubble robot-based context verification program stored in the memory 1005, and perform the same operation as the bubble robot-based context verification method.
[0197] The specific implementation of the context verification device based on the bubble robot of the present invention is basically the same as the above-mentioned embodiments of the context verification method based on the bubble robot, and will not be repeated here.
[0198] In addition, an embodiment of the present invention further proposes a computer storage medium, on which a context verification program based on a bubble robot is stored. When the context verification program based on a bubble robot is executed by a processor, the steps of the context verification method based on a bubble robot as described above are implemented.
[0199] The specific implementation of the computer storage medium of the present invention is basically the same as the above-mentioned embodiments of the context verification method based on the bubble robot, and will not be repeated here.
Claims
1. A context verification method based on bubble robot, It is characterized in that The following steps are involved: Export the preset configuration data of the bubble robot from the configuration platform, and use the script to extract the request parameters in the preset configuration data; Initiate a batch request to the bubble strategy interface based on the request parameters, and obtain the data returned by the bubble strategy interface; Compare the returned data with the preset configuration data, and if the comparison result is consistent, confirm that the bubble robot has passed the verification; When the bubble robot passes the verification, the FAQ corpus of the security service is exported from the knowledge base platform, and the FAQ corpus contains standard questions and similar questions of the security service; Selecting a simulated user question from the FAQ corpus, initiating a request to the intelligent interaction interface with the simulated user question as a request parameter, and obtaining the returned context question and original question; If the context question is equal to the original question and is successfully matched to the standard question in the FAQ corpus, the context verification is confirmed to be successful; If the context question is not equal to the original question, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question, it is confirmed that the context verification has passed.
2. The context verification method based on the bubble robot as claimed in claim 1, It is characterized in that Select simulated user questions from the FAQ corpus, including: Analyze common user questions based on the target business scenario, extract keywords of the target business scenario from the common user questions to form a keyword group; Using a keyword matching algorithm to traverse and search the FAQ corpus, and screen out a preset number of corpora that have the highest matching degree with at least one keyword in the keyword group; A correlation analysis is performed between the preset number of screened corpora and the target business scenario to generate the simulated user questions for simulating actual user questions.
3. The context verification method based on the bubble robot as claimed in claim 1, It is characterized in that After confirming that the bubble robot has passed the verification, it also includes: Collect user context information, including user operation logs, device status, geographic location, time, and user historical behavior data; Analyze the context information in real time to identify the user's current business scenario; Design multiple dialogue strategy templates based on the business scenario, wherein the dialogue strategy templates include response logic, prompt information display order, and related suggestion push rules; Selecting a target template from the dialogue strategy templates according to the user context information, and dynamically adjusting the dialogue strategy in the target template based on context changes; A plurality of context event triggering conditions are set to monitor changes in the user context information, and when the context event triggering conditions are met, the bubble robot actively initiates a dialogue based on the target template to provide prompt information or suggestions for the current scenario.
4. The context verification method based on the bubble robot as described in claim 1, It is characterized in that A request is initiated to the intelligent interaction interface using the simulated user question as a request parameter to obtain the returned context question and original question, including: Passing the simulated user question to the intelligent interactive interface and integrating user context information, including user operation history, device status and geographic location; Generate contextual questions based on the user context information and the simulated user questions using a context-aware algorithm and a deep learning model, wherein the user context information includes user historical behavior data, device status, and geographic location; Construct semantic representations for the generated contextual questions and standard questions and similar questions in the FAQ corpus to generate high-dimensional semantic vectors; Using a similarity algorithm to calculate the semantic similarity between the high-dimensional semantic vector of the context question and the high-dimensional semantic vector of the standard question or similar question in the FAQ corpus, and generate a similarity score; The standard question or similar question with the highest similarity is selected according to the similarity score, and the context question and the corresponding original question and the standard question or similar question with the highest similarity returned by the intelligent interaction interface are obtained.
5. The context verification method based on the bubble robot as claimed in claim 4, It is characterized in that Generate contextual questions based on the user context information and the simulated user questions using a context-aware algorithm and a deep learning model, including: Performing word segmentation on the simulated user question and performing part-of-speech tagging to identify key words; Extracting keywords from the simulated user question based on the part-of-speech tagging result, and calculating weights for the keywords based on their sequence positions and word frequencies in the simulated user question; Adjust the weight of the keyword based on the frequency of occurrence of the keyword in the FAQ corpus; Use the pre-trained word vector model to convert each keyword into a keyword vector, apply the adjusted keyword weight to the keyword vector, and generate a weighted keyword vector in a weighted manner; Encode the user's historical behavior data into a sequence vector, the device status into a one-hot vector, and the geographic location into a floating-point vector, and combine them to form a user context information vector; The weighted keyword vector is combined with the user context information vector to generate a comprehensive semantic vector, and the comprehensive semantic vector is input into a deep learning model to generate a final contextual question.
6. The context verification method based on the bubble robot as claimed in claim 1, It is characterized in that Comparing the returned data with the preset configuration data, if the comparison result is consistent, confirming that the bubble robot has passed the verification, including: Extracting predefined key data fields consistent with a preset configuration data structure and format from the returned data; Defining corresponding comparison rules according to the type of the predefined key data field, wherein the comparison rules include exact matching, tolerance matching and inclusion matching; If the type of the predefined key data field is a business type, a page ID, or a half-screen display identifier, accurately matching the predefined key data field with the preset configuration data; If the type of the predefined key data field is display time, duration or page height, performing tolerance matching on the predefined key data field and the preset configuration data; If the type of the predefined key data field is message content or title, performing inclusion matching on the predefined key data field and the preset configuration data; According to the comparison rules, the predefined key data fields in the returned data are compared with the preset configuration data item by item. If all the predefined key data fields pass the comparison, it is confirmed that the bubble robot has passed the verification.
7. The context verification method based on the bubble robot as claimed in claim 1, It is characterized in that Using a script to extract request parameters in the preset configuration data includes: Set up the script development environment, select and configure the script language, install dependent libraries, and set up the logging mechanism; Load and parse the preset configuration data file and convert it into an operational data structure; Determine and define the request parameter extraction rules, including page ID, business type, prompt information, display duration, and trigger conditions; Based on the request parameter extraction rule, extract the request parameter from the corresponding node or key-value pair of the preset configuration data; Format the extracted request parameters, unify the parameter format, store the formatted request parameters in a collection, and save them to memory or output them to a file.
8. A context verification device based on a bubble robot, It is characterized in that The context verification device based on the bubble robot includes: A configuration data management module exports preset configuration data of the bubble robot from the configuration platform and uses a script to extract request parameters in the preset configuration data; A bubble strategy request module initiates a batch request to the bubble strategy interface based on the request parameters, and obtains data returned by the bubble strategy interface; A data comparison and verification module compares the returned data with the preset configuration data, and if the comparison result is consistent, confirms that the bubble robot has passed the verification; A FAQ corpus management module, when the bubble robot passes the verification, exports the FAQ corpus of the security service from the knowledge base platform, wherein the FAQ corpus contains standard questions and similar questions of the security service; A user question simulation and interaction module selects simulated user questions from the FAQ corpus, initiates a request to the intelligent interaction interface using the simulated user questions as request parameters, and obtains returned context questions and original questions; A context verification module confirms that the context verification has passed if the context question is equal to the original question and is successfully matched to a standard question in the FAQ corpus; The similar question matching module confirms that the context verification has passed if the context question and the original question are not equal, and a similar question with a score exceeding a preset threshold is matched in the FAQ corpus based on the context question.
9. A context verification device based on a bubble robot, It is characterized in that The bubble robot-based context verification device includes a memory, a processor, and a bubble robot-based context verification program stored in the memory and executable on the processor. When the bubble robot-based context verification program is executed by the processor, the steps of the bubble robot-based context verification method as described in any one of claims 1 to 7 are implemented.
10. A computer storage medium, It is characterized in that The storage medium stores a context verification program based on a bubble robot, and when the context verification program based on a bubble robot is executed by a processor, the steps of the context verification method based on a bubble robot as described in any one of claims 1-7 are implemented.
