Scene question and answer management method and related method, apparatus, device, and storage medium

Abstract

The application discloses a scene question and answer management method and related methods, devices, equipment and storage media, wherein the scene question and answer management method comprises the following steps: selecting a standard question which is not triggered based on an inquiry template to determine a current question; analyzing response data of an inquired person to the current question to obtain a filling result of the response data to a slot set; in response to the filling result including a filling value of at least one target slot involving an associated scene of a target scene, first inquiring the inquired person based on an inquiry template of the associated scene until each target slot in the inquiry template of the associated scene is filled, and then selecting a question in the inquiry template of the target scene based on the filling result to obtain a new current question, and returning to the step of analyzing the response data until each target slot in the inquiry template of the target scene is filled. The above scheme can improve the completeness of information collection in the scene question and answer process as much as possible.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to a scenario-based question-answering management method and related methods, apparatus, devices and storage media. Background Technology

[0002] In these scenarios, staff often need to communicate with relevant personnel, and such communication is usually not a completely open chat, but rather has a clear business purpose and a relatively fixed information collection process.

[0003] In existing technologies, factors such as insufficient staff proficiency or decreased attention may prevent staff from completing the full information collection process, even when guided by relevant standards, resulting in the omission of some, or even critical, information in the scenario. Therefore, improving the completeness of information collection during scenario-based question-and-answer sessions has become an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem addressed by this application is to provide a scenario-based question-and-answer management method and related methods, devices, equipment, and storage media, which can maximize the integrity of information collection during the scenario-based question-and-answer process.

[0005] To address the aforementioned technical issues, the first aspect of this application provides a scenario-based question-and-answer management method, comprising: obtaining a question template for a target scenario; wherein the question template includes several standard questions and target slots for each standard question to be triggered for filling; selecting untriggered standard questions based on the question template to determine the current question; analyzing the response data of the respondent to the current question to obtain the filling result of the response data on the slot set; wherein the slot set includes target slots for each standard question; in response to a related scenario where the filling result includes at least one target slot filling value involving the target scenario, firstly, querying the respondent based on the question template of the related scenario until all target slots in the question template of the related scenario are filled, then selecting a question in the question template of the target scenario based on the filling result to obtain a new current question, and returning the step of analyzing the response data of the respondent to the current question to obtain the filling result of the response data on the slot set, until all target slots in the question template of the target scenario are filled.

[0006] To address the aforementioned technical problems, a second aspect of this application provides a method for querying inquiry records, comprising: obtaining the inquiry records of the respondent in a target scenario; wherein the inquiry records include the fill values ​​and traceability information of each target slot in the inquiry template of the target scenario, the traceability information being obtained from the respondent's response data during the inquiry process, and the fill values ​​being obtained from the scenario question-and-answer management method of the first aspect; and, in response to a query instruction for any target slot in the inquiry records, outputting the fill value and traceability information of the target slot.

[0007] To address the aforementioned technical problems, a third aspect of this application provides a scenario-based question-and-answer management device, comprising: a template acquisition module, a question determination module, a response analysis module, and a process switching module. The template acquisition module acquires a question template for a target scenario; wherein the question template includes several standard questions and target slots for each standard question to be triggered and filled. The question determination module selects untriggered standard questions based on the question template to determine the current question. The response analysis module analyzes the response data of the respondent to the current question to obtain the filling result of the response data on the slot set. The slot set contains target slots for each standard question; the process switching module is used to respond to the related scenarios of the target scenario whose filling results include at least one target slot filling value. First, the query is performed on the respondent based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. Then, based on the filling results, a question is selected in the query template of the target scenario to obtain a new current question. The module returns the steps of analyzing the respondent's response data to the current question to obtain the filling results of the slot set based on the response data, until all target slots in the query template of the target scenario are filled.

[0008] To address the aforementioned technical problems, a fourth aspect of this application provides an inquiry record query device, comprising: a record acquisition module and a slot query module. The record acquisition module is used to acquire the inquiry records of the respondent in a target scenario; wherein the inquiry records include the fill values ​​and traceability information of each target slot in the inquiry template of the target scenario, the traceability information being obtained from the respondent's response data during the inquiry process, and the fill values ​​being obtained from the scenario question-and-answer management device of the third aspect mentioned above; the slot query module is used to respond to a query command for any target slot in the inquiry records and output the fill value and traceability information of the target slot.

[0009] To address the aforementioned technical problems, the fifth aspect of this application provides an electronic device, which includes at least a memory and a processor coupled to each other. The memory stores at least program instructions, and the processor executes the program instructions to implement the scenario question-and-answer management method in the first aspect or the query record query method in the second aspect.

[0010] To address the aforementioned technical problems, the sixth aspect of this application provides a computer-readable storage medium storing program instructions executable by a processor. These program instructions are used to implement the scenario question-and-answer management method of the first aspect or the query record query method of the second aspect.

[0011] The above scheme obtains a query template for the target scenario, which contains several standard questions and target slots for each standard question to be triggered and filled. Based on the query template, it selects untriggered standard questions to determine the current question. Then, it analyzes the respondent's response data to the current question to obtain the filling result of the response data on the slot set, where the slot set contains the target slots for each standard question. In response to the filling result including at least one target slot filling value involving a related scenario of the target scenario, it first queries the respondent based on the query template of the related scenario until all target slots in the related scenario's query template are filled. Then, based on the filling result, it selects a question from the query template of the target scenario to obtain a new current question, and returns the steps of analyzing the respondent's response data to the current question to obtain the filling result of the response data on the slot set. Once all target slots in the query template for the target scenario are filled, the process minimizes human intervention during the scenario-based Q&A, reducing data collection omissions due to subjective factors. Furthermore, if the filling results indicate that at least one target slot's value relates to a related scenario, the system first queries the respondent using the related scenario's query template until all target slots in that template are filled. Then, it continues querying the respondent using the target scenario's query template until all target slots in the target scenario's template are filled. This approach not only collects information from the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information from those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based Q&A process. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating an embodiment of the question-and-answer management method for the scenario described in this application. Figure 2 This is a schematic diagram of a process of an embodiment of the question-and-answer management method in this application. Figure 3 This is a flowchart illustrating an embodiment of the query record retrieval method of this application; Figure 4 This is a schematic diagram illustrating the composition of an embodiment of the traceability information in this application; Figure 5This is an interactive schematic diagram of an embodiment of the query interface of this application; Figure 6 This is a schematic diagram of the framework of an embodiment of the question-and-answer management device for the present application scenario; Figure 7 This is a schematic diagram of the framework of an embodiment of the query record query device of this application; Figure 8 This is a schematic diagram of the framework of an embodiment of the electronic device of this application; Figure 9 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0013] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0014] In the following description, specific details such as particular system architectures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0015] In this paper, the terms "system" and "network" are often used interchangeably. The term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the slash " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, "many" in this paper means two or more.

[0016] In the embodiments disclosed in this application, an inquiry template for a target scenario is obtained. The inquiry template contains several standard questions and target slots for each standard question to be triggered for filling. Based on the inquiry template, untriggered standard questions are selected to determine the current question. Then, based on the respondent's response data to the current question, the filling result of the response data on the slot set is obtained. The slot set contains the target slots for each standard question. In response to a related scenario where the filling result includes at least one target slot's filling value involving the target scenario, the respondent is first questioned based on the inquiry template of the related scenario until all target slots in the related scenario's inquiry template are filled. Then, based on the filling result, a question is selected in the inquiry template of the target scenario to obtain a new current question. Finally, the process of analyzing the respondent's response data to the current question to obtain the filling result of the response data on the slot set is returned. The process continues until all target slots in the query template for the target scenario are filled. This minimizes human intervention during the scenario-based question-and-answer process, reducing the risk of data omissions due to subjective biases. Furthermore, if the filling result detects that at least one target slot's value relates to a related scenario, the system first queries the respondent based on the related scenario's query template until all target slots in that template are filled. Then, it continues querying the respondent based on the target scenario's query template until all target slots in the target scenario's template are filled. This approach not only collects information about the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information about those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based question-and-answer process.

[0017] To facilitate understanding of the embodiments disclosed in this application, the relevant terms used in the embodiments disclosed in this application will be described by way of example below.

[0018] Inquiry templates: These represent structured inquiry settings tailored to specific industry scenarios. They may include, but are not limited to: template identifiers, applicable scenarios, language sets, question nodes, field slots, answer types, follow-up question rules, jump rules, terminology glossaries, sensitive field strategies, and record output formats.

[0019] Question Node: Represents an executable query unit in the query template. Each question node can include the question intent, standard Chinese question format, target language expression (i.e., translating the standard Chinese question format into the target language), acceptable answer types (e.g., date, numeric, string, etc.), corresponding slot, required field marker (i.e., whether it is a required field), follow-up question conditions, jump conditions, and termination conditions.

[0020] Slot: Represents the fields that need to be extracted from the conversation and stored in a structured manner, such as name, nationality, ID number, destination, symptoms, contact location, companions, contact number, ticketing requirements, etc. Each slot can have field type, format constraints, whether it is required, whether it is sensitive, whether confirmation is required, optional enumeration, related slots, etc.

[0021] Inquiry Status: Indicates the current state of the dialogue relative to the template process, which may include, but is not limited to: current question node, completed node, filled slot, slot to be confirmed, missing slot, conflicting slot, free notes, for asking questions in return, next candidate question, etc.

[0022] Field Source Traceability Object: This refers to a verifiable evidence object generated for a specific structured field. It may include, but is not limited to: field name, field value, field type, source language, original text fragment in the source language, translated text fragment in the target language, audio file identifier, audio start and end time, source round, source issue node, current confirmation status, confirmer or confirmation method, sensitivity level, visible objects, review status, modification history, etc. It serves as the basis for on-site process control, as well as for subsequent review, quality inspection spot checks, or accountability tracing.

[0023] It should be noted that the above is merely an illustrative description of the relevant terms and does not limit the specific meaning of the relevant terms in actual application.

[0024] Please refer to the following: Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the question-and-answer management method for this application. It should be noted that the process operations in this embodiment can be executed by an electronic device with computing capabilities or related equipment containing an electronic device. The specific structure and type of the electronic device and related equipment containing the electronic device are not limited herein. Specifically, this embodiment may include the following steps: Step S11: Obtain the query template for the target scenario.

[0025] In this embodiment, the query template may include several standard questions and target slots that are triggered by each standard question for filling. As one possible example, the query template may be uniformly configured and distributed to the device via a cloud platform; as another possible example, the query template may be configured locally on the device; as yet another possible example, the query template may be edited by industry customers according to their own processes. Furthermore, the query template can be version-managed and supports features such as canary releases, rollbacks, and multi-language updates, which are not limited here.

[0026] In an implementation scenario, the target scenario may include, but is not limited to: disease control and epidemiological investigation scenario, hospital triage scenario, cultural and tourism service scenario, and overseas reception scenario for enterprises. The specific type of target scenario is not limited here, nor will it be listed one by one.

[0027] In one implementation scenario, the query template for the target scenario can contain several standard questions to clarify the various information slots involved in the target scenario for the respondent. For ease of understanding, taking a cultural and tourism service scenario as an example, the inquiry template for such a scenario may include, but is not limited to, several standard questions for inquiring about the following information slots: tourist needs, destination, time, number of people, ticketing, or safety tips, etc. For example, in this scenario, these standard questions may include, but are not limited to: "What is the purpose of your trip, such as pure sightseeing or shopping?", "Where are you going?", "How many days do you plan to stay?", "How many people are in total?", etc. The specific wording of these standard questions is not limited here. Alternatively, taking a disease control and epidemiological investigation scenario as an example, the inquiry template for such a scenario may include, but is not limited to, several standard questions for inquiring about the following information slots: symptoms, onset time, travel history, contact history, travel companions, contact information, etc. For example, in this scenario, these standard questions may include, but are not limited to: "What are your symptoms?", "When did you first notice them?", "Where have you been?", "Who have you been in contact with?", "Who else is traveling with you?", "What is their contact information?", etc. The specific wording of these standard questions is not limited here. Of course, the above examples are only a few possible examples of inquiry templates in practical applications; other possible scenarios are not limited here, nor will they be listed one by one.

[0028] In one implementation scenario, the standard questions in the inquiry template can include multiple expressions. The system can select the most appropriate expression from these options based on the respondent's language, region, age, cultural habits, and level of comprehension. For ease of understanding, taking a cultural tourism service scenario as an example, simple questions and answers can be used for ordinary tourists, while complete terminology can be used for professionals. As a possible example, when asking a question to a respondent, features can first be extracted from various expressions of the question to obtain first feature representations for each expression (e.g., including but not limited to: colloquialism, formality, sentence conciseness, professionalism, and friendliness). Then, features are extracted based on the respondent's basic information (e.g., identity information) to obtain a static feature representation. Finally, features are extracted based on the respondent's interaction data from the previous round of question-and-answer sessions (e.g., including the expression of the standard question and the respondent's response data) to obtain a dynamic feature representation. These static and dynamic feature representations are then fused to obtain a second feature representation of the respondent. This allows for the learning of the respondent's cognitive level, interaction preferences, and interaction atmosphere during the interaction. Based on this, the feature similarity between the first feature representations of various expressions of the question and the respondent's second feature representation can be obtained to determine the original score values ​​for each expression of the question. Simultaneously, based on the style similarity between the expression of the standard question in the previous round and various expressions of the current question, style correction coefficients can be obtained for each expression of the current question. It should be noted that the style correction coefficient is positively correlated with style similarity; the higher the style similarity, the closer the style correction coefficient is to 1, and vice versa. After obtaining the original scores and style correction coefficients for each expression of the current question, for each expression of the current question, a final score can be obtained based on the original score and style correction coefficient (e.g., the original score can be multiplied by the style correction coefficient). Based on the final scores of all expressions of the current question, one expression can be selected (e.g., the expression with the highest final score) as the target expression for asking the question to the respondent.

[0029] In one implementation scenario, as mentioned earlier, the inquiry template can also include target slots that are triggered by standard questions for filling. In other words, the target slot represents the primary purpose for which the standard question is triggered. For ease of understanding, taking the aforementioned cultural and tourism service scenario as an example, the target slot corresponding to the standard question "Where to go?" can be represented as "Destination," meaning the primary purpose of triggering the standard question "Where to go?" is to fill the target slot "Destination." Alternatively, taking the disease control and epidemiological investigation scenario as an example, the target slot corresponding to the standard question "What symptoms do you have?" can be represented as "Symptoms," meaning the primary purpose of triggering the standard question "What symptoms do you have?" is to fill the target slot "Symptoms." Of course, the above examples are merely a few possible examples of target slots in practical applications; other possible scenarios are not limited here, nor will they be listed in detail.

[0030] In one implementation scenario, an inquiry template can include not only a standard question and a target slot, but also other content, such as, but not limited to, question intent, answer type, triggering conditions, follow-up questioning rules, and multilingual expressions. The specific content that can be further included in the inquiry template is not limited here. For example, if the inquiry template includes question intent, the question intent can represent the purpose of the standard question, i.e., what the main intention of triggering the standard question is. It can be used to detect whether the subsequent response data of the respondent to the standard question meets the question intent. In a cross-language scenario, it can be used to detect whether the translated question after translating the standard question into the target language truly reflects the question intent. Or, for another example, if the inquiry template includes answer type, the answer type can represent the expected type of the standard question's answer (e.g., date, numeric, string, etc.). It can be used to detect whether the subsequent filling value of the target slot matches the answer type. Of course, the above examples are just a few possible examples of what the inquiry target can further include in practical applications. Other possible situations are not limited here, nor will they be listed one by one.

[0031] In one implementation scenario, taking the execution of this embodiment by relevant devices in the target scenario (e.g., handheld translators, dual-screen translators, window service terminals, mobile terminals, medical inquiry terminals, conference translation terminals, etc.) as an example, staff can activate the relevant devices. After activation, the devices can automatically obtain the inquiry template for the target scenario (e.g., obtain a pre-stored inquiry template from the local space of the relevant devices, or obtain an inquiry template from a cloud server) and load the inquiry template. Exemplarily, the system can load the corresponding inquiry template based on staff selection, device usage scenario, customer configuration, geographical location, business entry point, or historical sessions. For example, in a medical and disease control scenario, the system can load an epidemiological investigation template; or, in a port scenario, the system can load an entry inquiry template; or, in a cultural and tourism scenario, the system can load a tourist service or scenic spot guide template. Based on this, subsequent operations in this embodiment can be automatically executed. In addition, after loading the query template, the system can also initialize the query state, such as including but not limited to: default language direction (i.e., in cross-language scenarios, the source language -> target language is the default language direction), current question node, required slot set, optional slot set, sensitive field strategy, terminology glossary, follow-up question template, and record output format. For example, taking a cross-language scenario, the relevant devices may include, but are not limited to, the following software modules: template management module, speech recognition and translation module, dialogue type recognition module, slot extraction and mapping module, query state management module, follow-up question generation module, field source tracing module, structured record module, and interactive output module. It should be noted that the template management module can be used to create, load, update, and select query templates for target scenarios; the speech recognition and translation module can be used to convert cross-language speech into source language text and target language translation; the dialogue type recognition module can be used to determine whether the current input belongs to a template question, template answer, free addition, rhetorical question, off-topic, or flow control; the slot extraction and mapping module can be used to extract field values ​​from source language text or target language translation and map them to one or more target slots in the query template; and the query status management module can be used to maintain the current template's flow position, filled fields, and pending fields. The software includes modules for confirming fields, missing fields, conflicting fields, and jump conditions; a follow-up question generation module that can generate the next standard question, partial follow-up question, conflict confirmation, or format validation question based on the query status; a field source tracing module that can generate source tracing objects for extracted fields and record source language fragments, target language fragments, audio fragments, query nodes, confirmation status, and review markers; a structured record module that can generate bilingual question-and-answer records, field tables, confirmation statuses, audio references, and review evidence chains; and an interactive output module that can display questions, translations, fields, follow-up questions, and records on single-screen, dual-screen, or window service devices. The specific operations performed by each of the above software modules can be found in the following embodiments of this disclosure, and will not be elaborated upon here.

[0032] Step S12: Select standard questions that have not been triggered based on the query template to determine the current question.

[0033] In one implementation scenario, the inquiry template can be initialized with default standard questions and the questioning order between the standard questions. When the inquiry template is first launched to start an inquiry with the respondent, the default standard question can be selected directly to determine the current question. Afterwards, based on the respondent's response data to the current question, operations such as conducting partial follow-up questions on the respondent, launching the next standard question according to the questioning order, or skipping some standard questions can be selected. For details, please refer to the relevant descriptions below, which will not be elaborated here.

[0034] In another implementation scenario, unlike the aforementioned implementation, as another possible implementation, the inquiry template may not have initialized the default standard questions. In this case, when the inquiry template is first launched to start an inquiry with the respondent, the staff can manually trigger any standard question in the inquiry template to determine the current question. Afterwards, based on the respondent's response data to the current question, the staff can choose to conduct partial follow-up questions on the respondent, launch the next standard question according to the inquiry order, skip some standard questions, etc. For details, please refer to the relevant descriptions below, which will not be elaborated here.

[0035] It should be noted that the examples above are only a few possible examples of determining the current problem in practical applications. Other possible methods of determination are not limited here, nor will they be listed one by one.

[0036] Step S13: Analyze the response data of the respondents to the current question to obtain the filling result of the slot set by the response data.

[0037] In this embodiment of the disclosure, the slot set includes the target slots for each standard problem. The slot set may include a set of required slots, a set of optional slots, or a set of both required and optional slots; no limitation is made herein.

[0038] In one implementation scenario, before, after, or during an inquiry, some target slots can be pre-filled based on information such as the respondent's identification document, business type, and prior information. For example, scanning an identification document can obtain the fill values ​​for some target slots, such as name, nationality, and identification number.

[0039] In one implementation scenario, in a cross-language context, the current question, expressed in the source language, can be displayed on the staff side. The question can then be translated into the target language and displayed on the respondent side (or, after translation into the target language, speech synthesis can be performed and played back to the respondent; alternatively, both display and playback can be performed simultaneously). This ensures the respondent is aware of the current question. The specific method by which the respondent perceives the question is not limited here. Taking a dual-screen translator for scenario-based question-and-answer management as an example, the staff side can display information such as the question intent, field target (i.e., target slot), and current record status. The respondent side can only display the current question translated into the target language, minimizing the exposure of internal fields and process judgments. It should be noted that respondents can respond according to the requirements of the inquiry template (e.g., answering with the destination for the standard question "Where are you going?", and answering with the symptoms for the standard question "What are your symptoms?"), or they can answer freely using natural language (e.g., saying whatever comes to mind), or they can pre-fill multiple slots (e.g., answering the target slot corresponding to a standard question that has not yet been asked), ask counter-questions (e.g., raising doubts about the standard question), correct the aforementioned content, or express uncertainty, etc. Of course, the above examples are only a few possible examples of respondents' response data, and other possible response data are not limited here, nor will they be listed one by one. In addition, for the convenience of subsequent traceability, the system can also generate an audio file identifier for each round of voice input and record the start and end time of that round of language. When a specific text segment is extracted from a subsequent slot, the system can further record the approximate start and end time of that text segment in the audio or align the segment number, so as to facilitate the reverse location of the original recording from the structured fields.

[0040] In one implementation scenario, after obtaining the respondent's response data to a standard question, entity recognition, field extraction, and slot mapping can be performed sequentially based on the response data. (In a cross-language scenario, the response data can first be translated, that is, the respondent's target language can be translated into the worker's source language, and then entity recognition, field extraction, and slot mapping can be performed on the translated content. Alternatively, entity recognition, field extraction, and slot mapping can be performed directly on the target language response data.) This yields the execution result of the response data. Based on this, if the execution result indicates that there are no fill values ​​simultaneously mapped to multiple slots, the slot filling result of the response data can be directly obtained based on the execution result.

[0041] In a specific implementation scenario, as mentioned earlier, in a cross-language scenario, one can either perform translation operations on the response data first and then perform entity recognition, field extraction, and slot mapping in sequence, or directly perform entity recognition, field extraction, and slot mapping on the response data in sequence. Alternatively, one can simultaneously perform field extraction and consistency verification on both the original text and the translated text.

[0042] In a specific implementation scenario, in a cross-language scenario, during the field extraction process for the target slot, the field extraction for the target slot can be performed based on the original response data of the respondent to obtain the first candidate value of the target slot. Then, the field extraction for the target slot can be performed based on the translated data after the original response data of the respondent is translated into the language of the staff to obtain the second candidate value of the target slot. Finally, the field extraction for the target slot can be performed based on the translated data after the original response data of the respondent is translated into the language of the staff to be translated back into the language of the respondent to obtain the third candidate value of the target slot. Based on this, it is possible to verify whether the first, second, and third candidate values ​​of the target slot have the same meaning. If they do, the fill value of the target slot can be obtained based on the first, second, and third candidate values ​​(e.g., at least one of them can be selected as the fill value). Conversely, if they do not have the same meaning (e.g., some of the three have the same meaning, while some have different meanings; or, no two of the three have the same meaning), the target slot can be marked as a slot to be confirmed. As a possible example, the first, second, and third candidate values ​​can be temporarily filled into the target slot for subsequent querying and review.

[0043] In a specific implementation scenario, as mentioned earlier, in order to facilitate subsequent traceability, for each extracted field (i.e., target slot), the system can not only generate field values ​​(i.e., fill values), but also generate field source traceability objects to record which round of speech, which source language text, which target language translation, which problem node, and which audio segment the field comes from.

[0044] In a specific implementation scenario, during slot mapping, the system can determine which slot (i.e., the target slot) the extracted value should be written to based on the field type, question intent, current node, template slot definition, and context state. For example, a date entity can be mapped to "fever start date," "entry date," or "return date" based on the current question node, while a location entity can be mapped to "place of residence," "destination," "contact location," or "scenic spot destination" based on the semantic context. Of course, the above examples are just a few possible examples of slot mapping in practical applications, and other possible scenarios are not limited here, nor will they be listed one by one. In addition, during the process of the respondent responding to standard questions, the response data can simultaneously fill multiple target slots. For example, for the response data "I flew from City C to Guangdong on April 15th and stayed in District E," the system can extract the fill value "City C" for the target slot "departure point," the fill value "City A" for the target slot "destination," the fill value "April 15th" for the target slot "entry date," and the fill value "District E" for the target slot "place of residence," and write them to the corresponding target slots.

[0045] In one implementation scenario, after obtaining the execution result of the response data through the aforementioned operations, if the execution result indicates that there are fill values ​​simultaneously mapped to multiple target slots, the confidence level of each fill value mapped to multiple slots can be obtained. Then, based on the confidence level, the multiple target slots simultaneously mapped by the fill value are sorted to determine whether the fill value can uniquely map to a single target slot. If a uniquely mapped target slot cannot be determined, slots awaiting confirmation can be marked (e.g., multiple target slots simultaneously mapped by the fill value can be uniformly marked as slots awaiting confirmation). Query data is also generated to confirm the mapping relationship between the fill value and the target slot with the respondent. It should be noted that slots awaiting confirmation can represent target slots simultaneously mapped to the same fill value as other target slots. This method, when fill values ​​simultaneously mapped to multiple target slots exist, sorts them by the confidence level of the fill value mapped to multiple target slots, determines whether the fill value can uniquely map to a single target slot based on the sorting result, and further inquires with the respondent when it cannot be determined, thus minimizing potential ambiguity in slot mapping.

[0046] In a specific implementation scenario, where a fill value may map to multiple target slots, the confidence score of the fill value mapped to each of the multiple target slots can be calculated based on parameters such as the current issue node (i.e., the current issue), the most recently unfilled slots, semantic trigger words, template order, field format, and historical dialogues. For example, for any of the multiple target slots mapped to by the fill value, the semantic similarity between the fill value and the target slot can be obtained (e.g., a pre-trained semantic model can be used to extract the semantic features of both the fill value and the target slot, and then the cosine similarity between them can be calculated). sem (E,S i ): (1) In the above formula (1), E represents the fill value, and S i This represents the i-th target slot mapped by the fill value. The semantic features representing the padding values, Let represent the semantic features of the i-th target slot, and cos represent the cosine similarity function. Simultaneously, the contextual relevance score between the context fragments of the filled values ​​in the response data and the target slot can be obtained. ctx (E,S i For example, large language models can be used to analyze the contextual relevance between contextual fragments and target slots. Furthermore, the role matching score between the grammatical components (e.g., subject, object, adverbial, complement, etc.) that the filler values ​​function in the response data and the preset grammatical roles of the target slots can be obtained. syn (E,S i It can also obtain the domain constraint score of the entity type representing the fill value (e.g., location entity, date entity, etc.) to determine whether it meets the target slot requirements. rule (E,S i (For example, if the former meets the latter constraint, the domain constraint score can be directly set to the highest score; otherwise, it can be directly set to the lowest score.) It can also obtain the frequency percentage of fill values ​​of the same type as the fill value in historical question-and-answer sessions that are mapped to the target slot, as a corpus distribution score. his (E,S i ): (2) In the above formula (2), This represents the statistical number of times a fill value of the same type as the fill value in historical question-and-answer sessions has been mapped to the target slot. Based on this, the weighting coefficients for semantic similarity, contextual relevance, role matching, domain constraint, and corpus distribution can be adaptively determined according to the scenario type of the target scenario (e.g., for spoken dialogue scenarios, the weighting coefficient for contextual relevance can be increased, while for normative question-and-answer scenarios, the weighting coefficient for semantic similarity can be increased). Then, based on the weighting coefficients for semantic similarity, contextual relevance, role matching, domain constraint, and corpus distribution, these factors are weighted separately to obtain the initial confidence score (RawScore(E,S)) for mapping the fill value to the target slot. i ): (3) In the above formula (3), w1, w2, w3, w4, and w5 are the weighting coefficients of semantic similarity, contextual relevance, role matching, domain constraint, and corpus distribution, respectively. After obtaining the initial confidence of the multiple target slots mapped by the filling values, the standard deviation of the initial confidence of the multiple target slots can be obtained. For any of the multiple target slots mapped by the filling values, the mutual exclusion correction coefficient of the target slot can be obtained based on the confidence ratio between the initial confidence and the maximum initial confidence and the standard deviation. The mutual exclusion correction coefficient is positively correlated with the standard deviation and the confidence ratio. For example, for the i-th target slot, its mutual exclusion correction coefficient α(S i This can be represented as: (4) In the above formula (4), σ represents the standard deviation (e.g., the normalized standard deviation, which ranges from 0 to 1). λ represents the confidence ratio, and λ represents the mutual exclusion strength coefficient (e.g., it can be set to a value between 0 and 1). Therefore, the closer the initial confidence levels (i.e., the smaller the standard deviation) of the multiple target slots mapped by the fill values, the more differentiated the attack needs to be applied to the multiple target slots. Furthermore, the smaller the confidence ratio, the better the mutual exclusion correction effect. Based on this, the initial confidence levels can be corrected using the mutual exclusion correction coefficients of the target slots (e.g., for any target slot, its mutual exclusion correction coefficient can be multiplied by the initial confidence level) to obtain the corrected confidence level of the target slot. Finally, the corrected confidence levels of each target slot can be normalized (e.g., using the softmax function) to obtain the final confidence level of each target slot. In other words, the highest initial confidence level can remain unaffected (the mutual exclusion correction coefficient is 1 at this point). For other target slots, differential decay can be applied based on their relative difference from the highest initial confidence level, but this will not fundamentally change the original initial confidence level ranking (e.g., target slots with higher initial confidence levels will still be higher than target slots with lower initial confidence levels after correction). If the initial confidence levels of multiple target slots meet the preset confidence conditions (e.g., the initial confidence levels are exactly the same, or the difference between the initial confidence levels is very small), it indicates that the existing information is insufficient, and query data can be generated to confirm the slot mapping relationship. Furthermore, as another possible example, the ambiguity bias of the padding value can also be obtained (e.g., for the case where the padding value is a monosemous word, the ambiguity bias of the padding value can approach 1, and for the case where the padding value is a polysemous word, the ambiguity bias of the padding value can be less than 1). In this case, for any target slot, the initial confidence of the target slot can be corrected based on the mutual exclusion correction coefficient of the target slot and the ambiguity bias of the padding value, to obtain the corrected confidence score FixScore(E,S) of the target slot. i ): (5) In the above formula (5), This represents the initial confidence level for the i-th target slot. This represents the mutual exclusion correction coefficient for the i-th target slot. The ambiguity bias of the fill value is represented. For ease of description, the following examples are two target slots (with initial confidence levels of 0.8 and 0.7 respectively). The mutual exclusion strength coefficient λ is 0.5, the standard deviation σ is 0.05, and the ambiguity bias β is temporarily set to 1. Then, by using the above formulas (4) and (5), the mutual exclusion correction coefficient of the first target slot is 1 and the corrected confidence level is 0.8. The mutual exclusion correction coefficient of the second target slot is 0.941 and the corrected confidence level is 0.659. That is, after correction, the difference between the confidence levels increases from 0.1 to 0.141, while the original order remains unchanged. In addition, in some cases (e.g., when the accuracy requirements are relatively relaxed and the computing resources are relatively limited), the aforementioned formulas (4) and (5) can be skipped. After obtaining the initial confidence level, the initial confidence level can be normalized directly, and then the difference between the highest initial confidence level and the second highest initial confidence level after normalization can be compared to determine whether to automatically map or confirm follow-up questions (e.g., when the difference is large, automatic mapping can be performed, and when the difference is small, follow-up questions can be confirmed).

[0047] In a specific implementation scenario, after obtaining the final confidence scores of the target slots mapped to the fill value and sorting them, the confidence difference between the target slot with the highest confidence score and the target slot with the second highest confidence score can be obtained based on the sorting result. If the confidence difference meets the confidence condition, it can be determined that the fill value uniquely maps to the target slot with the highest confidence score. The confidence condition can include any of the following: the confidence difference is not lower than the confidence threshold, or the confidence difference is higher than the confidence threshold. Conversely, if the confidence difference does not meet the confidence condition, the target slot to which the fill value uniquely maps cannot be determined. It should be noted that when the accuracy requirement for slot differentiation is high, the confidence threshold can be set appropriately large; when the accuracy requirement for slot differentiation is relatively lenient, the confidence threshold can be set moderately. The specific value of the confidence threshold is not limited here. The above method, by sorting the confidence scores and using the confidence difference between the highest and second highest confidence scores after sorting, determines whether the fill value can uniquely map to a target slot, which helps improve the accuracy of slot mapping.

[0048] In a specific implementation scenario, when the target slot to which the fill value uniquely maps cannot be determined, as mentioned earlier, query data can be generated to confirm the mapping relationship between the fill value and the target slot with the respondent. After obtaining the respondent's response to the query data, the target slot to which the fill value uniquely maps can be determined based on that response. Furthermore, the marked status of the slot to be confirmed (i.e., the pending confirmation status) can also be written with traceability information. As a possible example, the pending confirmation status written in the traceability information ensures that subsequent processes do not easily determine that the corresponding problem node has been completed simply because a slot has a fill value.

[0049] Step S14: In response to the filling result including the filling value of at least one target slot involving the related scenarios of the target scenario, firstly, the respondent is queried based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. Then, based on the filling result, a question is selected in the query template of the target scenario to obtain a new current question. The step of analyzing the respondent's response data to the current question to obtain the filling result of the response data on the slot set is returned until all target slots in the query template of the target scenario are filled.

[0050] In an implementation scenario, the associated scenarios of the target scenario represent other scenarios related to the target scenario. For ease of understanding, taking a cultural and tourism service scenario as an example, if the fill value of the target slot "Items Carried" is "Power Bank", then the associated scenario "Item Supervision Scenario" related to the target scenario "Cultural and Tourism Service Scenario" can be identified. Alternatively, taking an entry management scenario as an example, if the fill value of the target slot "Place of Departure" is "Country A", then considering that "Country A" is currently experiencing an epidemic of infectious disease B, the associated scenario "Disease Control and Epidemiological Investigation Scenario" related to the target scenario "Entry Management Scenario" can be identified. Of course, the above examples are only a few possible examples of associated scenarios in practical applications, and other possible situations are not limited here, nor will they be listed one by one.

[0051] In one implementation scenario, if the populated value involves related scenarios of the target scenario, the respondent can be questioned based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. It should be noted that the specific content of the query template of the related scenario can be found in the aforementioned description of query templates, and will not be repeated here. Furthermore, the specific process of questioning the respondent based on the query template of the related scenario is similar to that of questioning the respondent based on the query template of the target scenario, and will not be repeated here either.

[0052] In one implementation scenario, when the filling result includes a situation where the filling value of any target slot does not involve any related scenarios of the target scenario, a question can be selected from the query template of the target scenario based on the filling result to obtain a new current question. Then, the process can return to analyze the response data of the respondent to the current question to obtain the filling result of the slot set based on the response data, until all target slots in the query template of the target scenario are filled. This approach, by directly determining a new current question and returning to the aforementioned analysis operation when the filling result includes a situation where the filling value of any target slot does not involve any related scenarios of the target scenario, improves the orderliness of the scenario-based question-and-answer process.

[0053] In a specific implementation scenario, regardless of whether the fill result includes any target slot whose fill value does not involve any related scenarios of the target scenario, or includes at least one target slot whose fill value involves related scenarios of the target scenario, the new current question must be determined based on the fill result. Specifically, the response data can be identified based on the fill result to obtain the dialogue type of the response data. The dialogue type can include, but is not limited to: a template answer to the current question, a pre-answer to at least some target slots corresponding to a standard question that has not yet been asked, and flow control. Based on this, a question can be selected from the query template of the target scenario based on the dialogue type to obtain the new current question. It should be noted that, in the above dialogue types, "template answer to the current question" means that the response data is exactly the answer to the current question (i.e., neither adding extra information, such as content corresponding to other slots, nor omitting information, such as content corresponding to the target slot in the current question); "advance answer to at least some target slots corresponding to standard questions that have not yet been asked" means that the response data also includes answer information for at least some slots corresponding to standard questions that have not yet been asked (i.e., advance answer); "flow control" means that the response data is answer information used to control the flow of the process, such as skip, don't know, refuse to answer, end, return to the previous question, etc. Of course, in practical applications, the dialogue types are not limited to these (as several other possible scenarios are given in the following descriptions), which will not be listed here. The above method identifies the response data based on the filling results to obtain the dialogue type of the response data, and then selects a question from the query template of the target scenario based on the dialogue type to obtain a new current question. Therefore, controlling the flow of the scenario question-and-answer process and the switching of questions through the dialogue type helps to improve the orderliness of scenario question-and-answer.

[0054] As a possible example, the dialogue type belongs to at least one of several preset types (such as the aforementioned types and the types mentioned later). In order to identify the dialogue type of the response data, based on the semantic overlap between the response data and the current question (e.g., if the response data revolves around the target slot of the current question, the semantic overlap is high, and vice versa), the question-answer semantic alignment feature is obtained. Based on the sentence structure information of the response data (e.g., sentence structure, punctuation, modal particles, sentence pattern) and behavioral identifiers (e.g., whether it contains question-leading words, rejection words, small talk words, etc.), feature extraction is performed to obtain sentence structure and behavioral features. The interaction role reversal feature of the response data is extracted with reference to the current question (e.g., under normal circumstances, the system asks a question and the person being asked answers, and the interaction role reversal degree is 0; conversely, if the person being asked asks a question, the interaction role reversal degree is 1; if the person being asked only questions and does not ask questions, the interaction role reversal degree is between the two). The proportion of effective words in the response data can also be counted to obtain the semantic validity feature. Based on this, the initial matching scores of response data belonging to various preset types can be predicted using the aforementioned question-and-answer semantic alignment features, sentence structure and behavioral features, interaction role reversal features, and semantic validity features. For example, if the question-and-answer semantic alignment features are high, the interaction role reversal features are low, and the semantic validity features are high, then the initial matching score for the preset type "template answer" is high; if the semantic validity features are low, then the initial matching score for the preset type "off-topic" is high. Of course, the above examples are merely a few possible instances of combining the aforementioned features to determine the initial matching score in practical applications. Other possible scenarios are not limited here, nor will they be listed one by one. In addition, the state correction coefficients for each preset type can be determined based on the current stage, historical interaction data, and whether different preset types are mutually exclusive. For example, if the current stage is the normal slot collection stage, a positive coefficient (i.e., a state correction coefficient greater than 1) can be applied to the preset type "template answer"; or, for another example, if historical interaction data shows that the interaction has deviated from the topic multiple times, a positive coefficient can be applied to the preset type "deviation from topic"; or, for yet another example, since the response data cannot simultaneously belong to mutually exclusive preset types (e.g., cannot simultaneously belong to "template answer" and "free supplement"), the scores of mutually exclusive preset types can be mutually inhibited (i.e., when it is determined that a positive coefficient needs to be applied to a preset type, a negative coefficient can be applied to its mutually exclusive preset type, and vice versa). After obtaining the initial matching score and state correction coefficient of each preset type, the corrected matching score of the corresponding preset type can be obtained based on the initial matching score and state correction coefficient of the same preset type (e.g., the state correction coefficient of the preset type can be multiplied by its initial matching score). fix (R,Ti ): (6) In the above formula, R represents the response data, and T i θ represents the i-th preset type. i The score represents the state correction coefficient for the i-th preset type. base (R,T i Let represent the initial matching score of the i-th preset type. Based on this, normalization can be performed on the modified matching scores of each preset type (e.g., normalization using the softmax function) to obtain the final matching score of each preset type. Then, based on the final matching scores of each preset type, at least one preset type can be selected as the dialogue type for the response data.

[0055] As one possible example, in response to dialogue types including template answers, a new current question can be obtained by sequentially selecting from standard questions that have not been triggered in the query template. For instance, the query template can be pre-set with the questioning order of each standard question, so that a new current question can be obtained by sequentially selecting from the questioning order of each standard question that has not been triggered in the query template.

[0056] As one possible example, in response to dialogue types including early responses, standard questions that have been answered before being asked can be skipped based on an inquiry template, resulting in a new current question. For instance, the inquiry template could be pre-set with the questioning order of each standard question, thus skipping the standard questions that have been answered beforehand and selecting them sequentially according to the questioning order of the standard questions that have not yet been triggered in the inquiry template to obtain a new current question.

[0057] As one possible example, in response to dialogue types including flow control, inapplicable standard questions can be skipped based on the query template to arrive at a new current question. For instance, the query template can also have configuration requirements for each standard question (e.g., mandatory questions, skippable questions, etc.). Based on the configuration requirements of each standard question in the query template and the questioning order of each standard question, inapplicable standard questions can be skipped, and a new current question can be obtained by sequentially selecting from the questioning order of each untriggered standard question in the query template.

[0058] As a possible example, dialogue types can also include free supplements that are related to the target scenario but do not directly answer the current question. For ease of understanding, let's take the cultural tourism service scenario as an example again. For the current question "Where did you come from?", the response data "I came from a previous tourist destination" is related to the target scenario but does not directly answer the current question; therefore, its dialogue type can be considered a free supplement. In this case, the response type including free supplements can prompt the respondent to answer the current question. In addition, free supplements can be divided into "business-related notes" and "irrelevant small talk." The former can be attached to subsequent structured data, while the latter may not be included in the formal record or may only be retained in the original dialogue log.

[0059] As one possible example, the dialogue type could also include a correction of the already filled target slot. For ease of understanding, let's continue with the cultural tourism service scenario. For the current question "Where are you from?", the response data might include "I misspoke, I plan to stay here for 3 days," meaning the dialogue type of the response data is a correction of the already filled target slot "stay duration". In this case, since the response dialogue type includes correction, we can first generate a question based on the response data to confirm the correction with the respondent, and then prompt the respondent to answer the current question.

[0060] As one possible example, dialogue types can also include proactive rhetorical questions. For ease of understanding, let's continue with the cultural tourism service scenario. For the current question "Where are you from?", the response data might include "Does visiting places before the last tourist destination count?", meaning the dialogue type of the response data is a proactive rhetorical question. In this case, since the response dialogue type includes proactive rhetorical questions, we can first generate answer data to reply to the respondent based on the response data, and then prompt the respondent to answer the current question.

[0061] As one possible example, dialogue types can also include off-topic responses. For ease of understanding, let's continue with the cultural tourism service scenario. For the current question "Where did you come from?", the response data might include "I heard this place is fun, so I came," meaning the dialogue type of the response data is off-topic. In this case, the response's off-topic dialogue type prompts the respondent to answer the current question.

[0062] It should be noted that the above examples are only a few possible examples of dialogue types in actual applications. Other possible dialogue types are not limited here, nor will they be listed one by one.

[0063] In a specific implementation scenario, before selecting a question from the query template of the target scenario based on the dialogue type to obtain a new current question, in response to the detection that some target slots of the current question are not yet filled, follow-up question data can be generated to ask the respondent about the missing slots. It should be noted that missing slots can represent the target slots of the current question that are not yet filled. For ease of understanding, let's take the cultural tourism service scenario as an example. The current question is "Where do you come from, and where are you going?", and the response data includes "From country A". While the target slot "originating location" in the current question is filled (i.e., filled with the value "country A"), the target slot "destination" is not yet filled. Therefore, follow-up question data can be generated to ask the respondent about the missing "destination" slot, such as "Where are you going?". Of course, the above example is only one possible example of missing slots and the generation of follow-up question data in practical applications. Other possible scenarios are not limited here, nor will they be listed in detail.

[0064] In a specific implementation scenario, before selecting a question from the query template of the target scenario based on the dialogue type and obtaining a new current question, in response to the detection that a target slot that has already been filled has been newly filled with a different value, a conflicting slot can be marked, and the staff can be prompted to confirm the conflicting slot with the respondent. It should be noted that a conflicting slot can represent a target slot that already has a filled value but has been newly filled with a different value. For ease of understanding, let's take the cultural tourism service scenario as an example. For the target slot "origin," there was already a filled value "Country C" in the previous dialogue. However, when faced with the current question "Where do I come from?", the respondent outputs the response data "I come from Country A," meaning that the target slot "origin" has a new different filled value "Country A." In this case, the target slot "origin" can be marked as a conflicting slot, and the staff can be prompted to confirm the conflicting slot with the respondent. Of course, the above example is only one possible example of conflicting slots and confirming conflicting slots with the respondent in practical applications. Other possible scenarios are not limited here, nor will they be listed in detail.

[0065] In a specific implementation scenario, before selecting a question from the query template of the target scenario based on the dialogue type and obtaining a new current question, in response to the detection that the fill value of the target slot does not meet the format conditions of the target slot (e.g., the ID number length does not meet the rules, the phone number has too many or too few digits, the date is missing the year or month, the duration of symptoms cannot be parsed, etc.), the format slot can be marked, and the staff can be prompted to confirm the format slot with the respondent. It should be noted that the format slot can represent the target slot whose fill value does not meet the format conditions. For ease of understanding, let's take the cultural and tourism service scenario as an example. The current question is "Please enter your contact information." The respondent's response data includes "My contact number is: 133XXXXXXX." Since the fill value "133XXXXXXX" of the target slot "Contact Number" does not meet the format conditions of the target slot "Contact Number" (i.e., 11 digits), the target slot "Contact Number" can be marked as a format slot, and the staff can be prompted to confirm the format slot with the respondent. Of course, the above examples are just one possible example of format slots in actual applications and confirming format slots with the person being queried. Other possible situations are not limited here, nor will they be listed one by one.

[0066] In a specific implementation scenario, before selecting a question from the query template of the target scenario based on the dialogue type to obtain a new current question, in response to the detection that the filled value of the target slot is related to an unfilled target slot, the related slot can be marked, and question data for follow-up questions to the respondent about the related slot can be generated. It should be noted that a related slot can represent a target slot that is not yet filled but is related to a filled value that has already been filled. For ease of understanding, let's take the cultural tourism service scenario as an example. For the current question "Where do you come from?", if the respondent's response data includes "from country A", and a certain infectious disease is currently prevalent in country A, then because the filled value "country A" is related to unfilled target slots such as "travel duration" and "contact persons", it can be marked as a related slot, and question data for follow-up questions to the respondent about the related slot can be generated. Of course, the above example is only one possible example of related slots and follow-up questions to the respondent in practical applications; other possible scenarios are not limited here, nor will they be listed in detail.

[0067] In one implementation scenario, during the process of writing fill values ​​for target slots, traceability information can also be recorded for the target slots based on the response data. It should be noted that the traceability information can be used to verify whether the fill values ​​for the target slots are correct. This method, recording traceability information for target slots based on response data, enables "slot-level" information traceability, significantly improving the convenience of subsequent information queries and verifications.

[0068] In a specific implementation scenario, traceability information may include information related to the issue to which the target slot belongs, such as the issue node number, issue intent, standard question format, etc., without limitation.

[0069] In a specific implementation scenario, traceability information may include the information recorded by staff during the Q&A process for the question belonging to the target slot, such as the original question, the target language question, and the playback time on the staff's side, without any limitations.

[0070] In a specific implementation scenario, traceability information may include the recorded information on the respondent's side during the question-and-answer process for the question to which the target slot belongs, such as the respondent's original answer, translation, audio citation, etc., without limitation.

[0071] In a specific implementation scenario, traceability information may include the target slot's own record information, such as the target slot itself, the source round of the target slot, the source node of the problem to which the target slot belongs, and the confirmation status of the target slot (e.g., pending confirmation, confirmed status, etc.), etc., without limitation.

[0072] In a specific implementation scenario, traceability information may include record information related to the source of the target slot's fill value, such as the original language text fragment, the target language translated text fragment, the audio file identifier, the audio start and end time, and the fragment alignment information, etc., without limitation.

[0073] In a specific implementation scenario, traceability information may include missing slots, conflicting slots, slots pending review and their processing results, etc., without limitation.

[0074] In a specific implementation scenario, traceability information may include slot-level record information, such as review status, modification history, confirmer, confirmation method, etc., without limitation.

[0075] In a specific implementation scenario, traceability information may include optional supplementary content, remarks, etc., which are not limited here.

[0076] In a specific implementation scenario, traceability information may include the results of sensitive field desensitization, permission tags, local retention policies, etc., without limitation.

[0077] In a specific implementation scenario, traceability information can be exported as structured data. In practical applications, traceability information can also be exported as tables, PDFs, business system interface data, email attachments, and locally encrypted files. For sensitive scenarios, the system can upload only the anonymized fields or structured summaries, while the original audio and complete text can be stored on the local device or an authorized intranet system.

[0078] In one implementation scenario, the system can update the inquiry template based on on-site inquiry records and staff corrections. For example, if a standard question frequently triggers follow-up questions, the system can suggest optimizing the multilingual expression of that standard question; or, if a free-response question often includes multiple slots in advance, the system can adjust the order of standard questions in the inquiry template; or, if a target slot is frequently left blank, the system can increase its follow-up priority. Of course, the above examples are just a few possible scenarios for updating the inquiry template. Other possible update methods are not limited here, nor will they be listed in detail. As one possible example, the update can be performed locally on the client side, or it can be performed after connecting to the network and uploading anonymized statistical data to the cloud. The cloud then generates new template versions, terminology versions, follow-up question rules, and field mapping rules, which are then distributed according to industry customers, regions, device groups, and languages.

[0079] In one implementation scenario, as described above, the relevant operations in this embodiment can be performed by devices such as single-screen translators and dual-screen translators. As one possible example, for a dual-screen device, the screen on the worker's side can display information such as the template name, current question, filled slots, missing slots, slots awaiting confirmation, and the next question. The screen on the respondent's side can display information such as the translation of the current question into the target language, a confirmation button, candidate options, and necessary explanations. As another possible example, for a single-screen device, the system can display questions, answers, slot filling results, and follow-up suggestions through partitions, cards, pop-ups, etc. The worker can click on a slot to view its source text and translation, or manually modify the filling value. During voice interaction, the system can automatically play the next question or follow-up question, such as "Please confirm whether your final destination is City A or City B," "When did your cough start," "Can this question be skipped?" etc. The system can output source or target language prompts based on the current language direction. When outputting business records, the system can provide different recording modes such as "save only structured fields," "save fields and bilingual text," and "save fields, bilingual text, and audio references." In other words, the system can provide different visible content for different roles. For example, staff can query template processes, slot status, missing item alerts, and conflict prompts; those being questioned can query the current question, confirmation options, and necessary explanations; and the management platform can view the anonymized summary data. In addition, the system also supports offline operation. In offline mode, devices can use local templates, local terminology lists, local entity recognition models, and local rules to complete the process. After connecting to the internet, template updates or anonymized records can be synchronized.

[0080] In one implementation scenario, please refer to the following: Figure 2 , Figure 2This is a schematic diagram illustrating the process of one embodiment of the question-and-answer management method for this application. For example... Figure 2 As shown, after obtaining the query template for the target scenario (e.g., it may contain standard questions and their corresponding target slots), questions can be asked to the respondent (e.g., in the first language, questions such as "Have you had a fever, been bitten by mosquitoes, or recently visited any places?") and the respondent's response data to the standard questions can be obtained (e.g., users freely answering in the second language, such as "I was bitten by mosquitoes," "I had a high fever days later," or "I did not have persistent vomiting," which translate to "I was bitten by mosquitoes," "I had a high fever a few days later," or "I did not have persistent vomiting" in the first language, respectively). After a series of operations such as sentence and fragment alignment, cross-node field decomposition, and node status determination (see the aforementioned descriptions and subsequent examples for details), a bilingual structured record can finally be formed (e.g., it may contain fill values, source language fragments, translated fragments, audio recordings, confirmation / review status, etc.). For ease of understanding, several different target scenarios are used as examples for illustration below. Of course, the examples below are merely simple illustrations of the question-and-answer management method for this application scenario, and do not constitute a limitation of the question-and-answer method for this application scenario.

[0081] As a possible example, taking the target scenario as a disease control and epidemiological investigation scenario, the system can load a "disease control and epidemiological investigation template" as the query template for the target scenario. It can include symptom inquiry nodes, mosquito bite exposure nodes, travel history nodes, onset time nodes, severe warning symptom nodes, contact information nodes, etc. The above different nodes can be regarded as different standard questions. In addition, target slots such as "whether there is fever", "fever start time", "mosquito bite exposure", "bite location", "bite date", "travel location", "gingival bleeding", "persistent vomiting" can also be defined. During the initial inquiry, staff can select the "Symptom Inquiry Node" as the starting point. This allows for the translation of the questions corresponding to the starting node into the respondent's language in a cross-language context. For example, staff could ask the respondent, "Have you had a fever recently? Have you been bitten by mosquitoes? Where have you been recently?" The respondent would then respond in their native language, such as English, to obtain their response data: "I was bitten by mosquitoes near the hotel pool last Friday. I had a high fever two days later, but I did not have bleeding gums or persistent vomiting. I also visited a night market in District E." Of course, the above examples are just a few possible scenarios in practical application; other possible situations are not limited here, nor will they be listed in detail.Based on this, a series of operations such as translation, entity recognition, field extraction, and slot mapping can be performed on the above response data to obtain the fill value and traceability information of the target slot. For example, if the fill value of the target slot "Mosquito bite exposure" is "Yes", its traceability information may include: "Source language" is "English", "Source original text fragment" is "bitten by mosquitoes", "Target language translated text fragment" is "bitten by insects", "Audio file identifier" is "Recording A of this round of inquiries", "Audio fragment" is "00:03.2-00:05.1", "Source node" is "Initial screening of symptoms and exposure history", and "Confirmation status" is "Pending confirmation"; if the fill value of the target slot "Bite location" is "Near the hotel pool", its traceability information may include: "Source original text fragment" is "near the hotel pool", "Audio fragment" is "00:05.2-00:07.0", and "Confirmation status" is "Pending confirmation"; if the fill value of the target slot "Bite time" is "Last Friday", its traceability information may include: "Source original text fragment" is "last The target slot's "Friday" and "Audio clip" values ​​are "00:07.1-00:08.0", and "Confirmation status" is "Pending conversion" (i.e., converted to date). The target slot's "Fever start time" is filled with "two days after the bite," and its traceability information can include: "Source original clip" as "two days later", "Audio clip" as "00:10.0-00:11.4", and "Confirmation status" as "Pending conversion" (i.e., converted to date). The target slot's "Gingival bleeding" value is "No," and its traceability information can include: "Source original clip" as "did not have bleeding gums", "Audio clip" as "00:12.5-00:14.2", and "Confirmation status" as "Pending confirmation." The target slot's "Persistent vomiting" value is "No," and its traceability information can include: "Source original clip" as "did not have ... persistent." The "vomiting" and "audio clip" are set to "00:12.5-00:16.0", and the "confirmation status" is "pending confirmation". The target slot "location of residence or activity" is filled with the value "E District Night Market", and its source information can include: "original source clip" as "a night market in E District", "audio clip" as "00:17.0-00:20.5", and "confirmation status" as "pending confirmation". The system can then control the template process based on the source source object of the field (i.e., the aforementioned source information).For "mosquito bite exposure nodes" and "severe warning symptom nodes," since the corresponding slots already have source text, translations, audio clips, and round records, but have not yet been confirmed by staff, the system does not directly mark these nodes as completed. Instead, it merges them into a confirmation question on the staff side, such as "Please confirm: You were bitten by mosquitoes near the hotel pool last Friday; you developed a high fever two days later; you did not have bleeding gums or persistent vomiting, is that correct?" Furthermore, for "travel history nodes," "E-zone night market" can be added as a pending travel location, and when the travel history node is reached in subsequent processes, the system skips the ordinary open-ended question and instead asks about the length of stay and the people traveling with the individual. After confirmation, the status of the above slots can be updated from "pending confirmation" to "confirmed," and the corresponding template node can be marked as completed or partially completed. If subsequent reviewers need to verify the accuracy of slots such as "persistent vomiting," they can directly access the traceability information of that slot from the structured records for review, such as playing back audio clips and comparing them with the original English text and Chinese translation.

[0082] As a possible example, taking the target scenario as an entry inquiry scenario, the system can load an "entry purpose and itinerary inquiry template" as the inquiry template for the target scenario. It can include entry purpose node, departure point node, entry city node, stay period node, accommodation address node, subsequent destination node, and items carried node, etc. The above different nodes can be regarded as different standard questions. In addition, target slots can also be defined, which will not be elaborated here. During the initial inquiry, staff can select the "Purpose of Arrival Node" as the starting point. This allows for the translation of the question corresponding to the starting point into the respondent's language in a cross-language context. For example, staff could ask, "What is the purpose of your trip?" The respondent can respond in their own language, such as English, to obtain their response data: "Tourism. I flew from city C to city A this morning. I will stay at the Garden Hotel for three nights, then go to city B for a trade fair. I have two spare lithium batteries in my carry-on." Of course, the above examples are just a few possible scenarios in practical application; other possible situations are not limited here, nor will they be listed in detail.Based on this, a series of operations such as translation, entity recognition, field extraction, and slot mapping can be performed on the above response data to obtain the fill value of the target slot and its traceability information. For example, if the fill value of the target slot "Purpose of Entry" is "Tourism", its traceability information may include: "Source Node" is "Purpose of Entry" and "Confirmation Status" is "Confirmed"; if the fill value of the target slot "Place of Departure" is "City C", its traceability information may include: "Source Node" is "Purpose of Entry (Free Answer)" and "Confirmation Status" is "Pending Confirmation"; if the fill value of the target slot "City of Entry" is "City A", its traceability information may include: "Source Text Fragment" is "flew from City C to City A this morning" and "Confirmation Status" is "Pending Confirmation"; if the fill value of the target slot "Accommodation Address" is "Garden Hotel", its traceability information may include: "Source Text Fragment" is "stayat the Garden Hotel for three nights" and "Confirmation Status" is "Pending Confirmation"; if the fill value of the target slot "Duration of Stay" is "Three Nights", its traceability information may include: "Source Text Fragment" is "for three The target slot "Subsequent Destination" is filled with "City B", and its traceability information can include: "Source text fragment" as "then go to City B", and "Confirmation Status" as "Pending Confirmation"; the target slot "Subsequent Activities" is filled with "trade fair", and its traceability information can include: "Source text fragment" as "for a trade fair", and "Confirmation Status" as "Pending Confirmation"; the target slot "Carried Items" is filled with "Two spare lithium batteries in my carry-on", and its traceability information can include: "Source text fragment" as "two spare lithium batteries in my carry-on", "Sensitivity Level" as "Regulated Item", and "Confirmation Status" as "Pending Verification". Based on the above records, the system can adjust the subsequent process: when the process reaches the nodes of departure point, entry city, accommodation address, length of stay, and subsequent destination, the system does not repeat the complete standard questions, but displays the slot source and confirmation button on the staff side, and generates a merged confirmation statement for the respondent to confirm.In addition, for "trade fair", the system identifier may have a business conflict with the "tourism" entry purpose. The target slot "entry purpose" can be marked as a "conflicting slot", and staff can be prompted to confirm whether the entry purpose needs to be changed from "tourism" to "business / exhibition" or recorded as "tourism and exhibition". For "spare lithium battery", the system does not treat it as a normal itinerary field, but enters the carry-on item sub-process (i.e. the related scenario "regulated item scenario" of the target scenario "entry inquiry scenario") to load the inquiry template of the corresponding scenario and continue to ask the person being questioned about things such as battery capacity, whether it is checked in, quantity, etc.

[0083] It should be noted that the above examples are merely two possible illustrations of scenario-based inquiry management using inquiry templates for target scenarios in practical applications. Other possible scenarios are not limited here, nor will they be listed in detail. Furthermore, any matters not covered in the above examples can be referred to other relevant descriptions in the embodiments disclosed in this application, which will not be described in detail here.

[0084] The above scheme obtains a query template for the target scenario, which contains several standard questions and target slots for each standard question to be triggered and filled. Based on the query template, it selects untriggered standard questions to determine the current question. Then, it analyzes the respondent's response data to the current question to obtain the filling result of the response data on the slot set, where the slot set contains the target slots for each standard question. In response to the filling result including at least one target slot filling value involving a related scenario of the target scenario, it first queries the respondent based on the query template of the related scenario until all target slots in the related scenario's query template are filled. Then, based on the filling result, it selects a question from the query template of the target scenario to obtain a new current question, and returns the steps of analyzing the respondent's response data to the current question to obtain the filling result of the response data on the slot set. Once all target slots in the query template for the target scenario are filled, the process minimizes human intervention during the scenario-based Q&A, reducing data collection omissions due to subjective factors. Furthermore, if the filling results indicate that at least one target slot's value relates to a related scenario, the system first queries the respondent using the related scenario's query template until all target slots in that template are filled. Then, it continues querying the respondent using the target scenario's query template until all target slots in the target scenario's template are filled. This approach not only collects information from the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information from those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based Q&A process.

[0085] Please see Figure 3 , Figure 3 This is a flowchart illustrating an embodiment of the inquiry record query method of this application. It should be noted that the process operations in this embodiment can be performed by an electronic device with computing capabilities or related equipment containing an electronic device. The specific structure and type of the electronic device and related equipment containing the electronic device are not limited herein. Specifically, this embodiment may include the following steps: Step S31: Obtain the questioning records of the person being questioned in the target scenario.

[0086] In this embodiment of the disclosure, the inquiry record may include the fill value and traceability information of each target slot in the inquiry template of the target scenario. The traceability information is obtained from the response data of the person being questioned during the inquiry process, and the fill value is obtained from the above-mentioned scenario question and answer management method embodiment. For details, please refer to the aforementioned scenario question and answer management method embodiment, which will not be repeated here.

[0087] In one implementation scenario, the query template and the specific process of querying the respondent based on the query template to obtain information including fill-in values ​​and traceability information can be found in the above-described scenario question and answer management method implementation examples, which will not be repeated here.

[0088] In one implementation scenario, please refer to the following: Figure 4 , Figure 4 This is a schematic diagram illustrating the composition of an embodiment of the traceability information in this application. For example... Figure 4As shown, traceability information may include, but is not limited to, the following layers: text evidence layer, audio evidence layer, confirmation control layer, and process status layer, etc., without limitation here. Specifically, the text evidence layer records the source evidence of the target slot's fill value at the text level; the audio evidence layer records the source evidence of the target slot's fill value at the audio level; the confirmation control layer records information regarding whether the target slot's fill value has been confirmed; and the process status layer records the source evidence of the target slot's fill value at the process level. Taking the target slot "Persistent vomiting" (with a fill value of "No") as an example, the text evidence layer can include the original source language fragment (e.g., "did not have persistent vomiting", i.e., no persistent vomiting) and the target language translated fragment (e.g., "no persistent vomiting"). The audio evidence layer can include the recording ID (e.g., "A"), start and end times (e.g., "00:12.5-00:16.0"), and segment number (e.g., "seg-07"). The confirmation control layer can include the confirmation status (e.g., "pending confirmation"), sensitivity level (e.g., "CDC sensitive"), and visible objects (e.g., "staff side"). The process status layer can include the source round (e.g., "round 3"), inquiry node (e.g., "initial screening of exposure history"), and source type (e.g., "free supplementation"). Furthermore, in practical applications, such as... Figure 4 As shown, the target slot, its fill value, and slot type can also be included in the fill value layer of traceability information (e.g., target slot "persistent vomiting", fill value "no", slot type "warning symptom negation"). As a possible example, information recorded during subsequent queries and reviews can also be included in the review and traceability layer of source information (e.g., review status "not reviewed", confirmer "staff confirmed", modification history "version record"). Of course, the above examples are just a few possible examples of source information, and the specific composition of source information is not limited here, nor will it be listed one by one.

[0089] Step S32: In response to a query command for any target slot in the query record, output the fill value and traceability information of the target slot.

[0090] In one implementation scenario, please refer to the following: Figure 5 , Figure 5 This is an interactive schematic diagram of one embodiment of the query interface of this application. For example... Figure 5 As shown, query commands can be triggered and generated within the query interface. For example, the target slot and its fill value can be displayed in the first area of ​​the query interface, and in response to a query command for any target slot in the first area (e.g., clicking on any target slot), the traceability information of the target slot can be displayed in the second area of ​​the query interface (see details). Figure 5 (and the aforementioned description regarding traceability information).

[0091] In one implementation scenario, in response to the review instructions for the populated values ​​based on the reference traceability information, the review information for the target slot can also be recorded. It should be noted that review instructions can include any of the following: approval instructions, modification instructions, and instructions pending review. Please refer to the following further details. Figure 5 If, based on the audio clip, it is confirmed that the fill value for the target slot "continuous vomiting" has passed the review, then the review information for the target slot "continuous vomiting" can be recorded as "passed". Of course, the above example is only one possible example of recording review information in actual application. Other possible situations are not limited here (e.g., it is also possible to support modifying the fill value based on the traceability information, in which case the triggered modification command can be recorded in the review information), and will not be listed one by one.

[0092] The above solution obtains the query records of the respondent in the target scenario. The query records include the fill values ​​and traceability information of each target slot in the query template of the target scenario. The traceability information is obtained from the respondent's response data during the query process, and the fill values ​​are obtained from the above-described scenario question and answer management method embodiment. Then, in response to the query command for any target slot in the query record, the fill value and traceability information of the target slot are output. Therefore, with the support of collecting information as completely as possible during the scenario question and answer process, the query requirements of the query records can be met.

[0093] Please see Figure 6 , Figure 6 This is a schematic diagram of the framework of an embodiment of the scenario question-and-answer management device of this application. The scenario question-and-answer management device 60 includes: a template acquisition module 61, a question determination module 62, a response analysis module 63, and a process switching module 64. The template acquisition module 61 is used to acquire the query template of the target scenario; wherein, the query template includes several standard questions and target slots for each standard question to be triggered for filling; the question determination module 62 is used to select untriggered standard questions based on the query template to determine the current question; the response analysis module 63 is used to analyze the response data of the respondent to the current question to obtain the filling result of the response data on the slot set; wherein, the slot set The process switching module 64 is used to respond to the related scenarios of the target scenario in which the filling result includes the filling value of at least one target slot. First, the respondent is queried based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. Then, based on the filling result, a question is selected in the query template of the target scenario to obtain a new current question. The process switching module 64 is used to respond to the related scenarios of the target scenario in which the filling result of the target scenario is filled. Then, the process switching module 64 is used to respond to the related scenarios of the target scenario in which the filling result of the target slot is filled. The process switching module 64 is used to respond to the related scenarios of the target scenario in which the filling result of the target slot is filled. The process switching module 64 is used to respond to the related scenarios of the target scenario in which the filling result of at least one target slot in the query template of the target scenario is filled.

[0094] In the above scheme, the scenario question-answering management device 60 obtains the query template of the target scenario. The query template contains several standard questions and target slots for each standard question to be triggered for filling. Based on the query template, it selects standard questions that have not been triggered to determine the current question. Then, it analyzes the response data of the respondent to the current question to obtain the filling result of the response data on the slot set. The slot set contains the target slots for each standard question. In response to the filling result including at least one target slot filling value involving the target scenario in a related scenario, it first queries the respondent based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. Then, based on the filling result, it selects a question in the query template of the target scenario to obtain a new current question and returns the filling result of the response data on the slot set based on the analysis of the respondent's response data to the current question. The process involves several steps until all target slots in the query template for the target scenario are filled. This minimizes human intervention during the scenario-based question-and-answer process, reducing the risk of data loss due to subjective biases. Furthermore, if the filling result indicates that at least one target slot's value relates to a related scenario, the respondent is first questioned using the query template of that related scenario until all target slots in that scenario's template are filled. Then, the process continues using the query template of the target scenario until all target slots in the target scenario's template are filled. This approach not only collects information about the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information about those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based question-and-answer process.

[0095] In some disclosed embodiments, the scenario question-answering management device 60 includes a question selection module, used to select a question from the query template of the target scenario based on the filling result in response to the filling result including any target slot filling value not involving the associated scenario of the target scenario, to obtain a new current question; the scenario question-answering management device 60 includes a loop iteration module, used to return the step of analyzing the response data of the respondent to the current question to obtain the filling result of the response data on the slot set, until all target slots in the query template of the target scenario are filled.

[0096] In some disclosed embodiments, the scenario question-and-answer management device 60 includes a type identification module, used to identify the response data based on the filling result to obtain the dialogue type of the response data; wherein, the dialogue type includes at least one of the following: a template answer to the current question, an advance answer to at least part of the target slots corresponding to a standard question that has not yet been asked, and flow control; the scenario question-and-answer management device 60 includes a question determination module, used to select a question from the query template of the target scenario based on the dialogue type to obtain a new current question.

[0097] In some disclosed embodiments, the question determination module includes a first response submodule, used to sequentially select standard questions that have not been triggered in the query template in response to a dialogue type including template answer, to obtain a new current question; the question determination module includes a second response submodule, used to skip standard questions that have not yet been asked but have been answered in advance based on the query template in response to a dialogue type including early answer, to obtain a new current question; the question determination module includes a third response submodule, used to skip inapplicable standard questions based on the query template in response to a dialogue type including flow control, to obtain a new current question.

[0098] In some disclosed embodiments, the dialogue type further includes at least one of the following: free supplementation related to the target scenario but not directly answering the current question, filling correction of already filled target slots, proactive counter-questioning, and deviation from the topic. The question determination module includes a fourth response submodule, used to prompt the respondent to answer the current question in response to the dialogue type including free supplementation; the question determination module includes a fifth response submodule, used to generate question data for confirming the filling correction with the respondent based on the response data in response to the dialogue type including filling correction, and then prompt the respondent to answer the current question; the question determination module includes a sixth response submodule, used to generate answer data for replying to the respondent based on the response data in response to the dialogue type including proactive counter-questioning, and then prompt the respondent to answer the current question; the question determination module includes a seventh response submodule, used to prompt the respondent to answer the current question in response to the dialogue type including deviation from the topic.

[0099] In some disclosed embodiments, the scenario question-and-answer management device 60 includes a missing detection module, which generates follow-up question data for asking the respondent about the missing slots in response to detecting that some target slots of the current question have not been filled; wherein, the missing slots represent the target slots of the current question that have not been filled.

[0100] In some disclosed embodiments, the scenario question-and-answer management device 60 includes a conflict detection module, which, in response to detecting that a target slot that has already been filled has been newly filled with a different filling value, marks the conflict slot and prompts the staff and the person being asked to confirm the conflict slot; wherein, the conflict slot represents a target slot that has already been filled with a filling value but is now filled with a different filling value.

[0101] In some disclosed embodiments, the scenario question-and-answer management device 60 includes a format detection module, which, in response to detecting that the fill value of a target slot does not conform to the format conditions of the target slot, marks the format slot and prompts the staff and the person being asked to confirm the format slot; wherein, the format slot represents the target slot whose fill value does not conform to the format conditions.

[0102] In some disclosed embodiments, the scenario question-and-answer management device 60 includes an association detection module, which, in response to detecting that the fill value of a target slot is associated with an unfilled target slot, marks the associated slot and generates question data for asking the respondent about the associated slot; wherein, the associated slot represents a target slot that is not yet filled but is associated with a fill value that has already been filled.

[0103] In some disclosed embodiments, the response analysis module 63 includes a data analysis submodule, used to sequentially perform entity recognition, field extraction, and slot mapping based on the response data to obtain the execution result of the response data; the response analysis module 63 includes a slot mapping submodule, used to, in response to the execution result indicating the existence of fill values ​​simultaneously mapped to multiple target slots, obtain the confidence level of the fill values ​​mapped to multiple target slots respectively, sort the multiple target slots simultaneously mapped by the fill values ​​based on the confidence level, determine whether the fill value can be uniquely mapped to a target slot, and, if the target slot cannot be uniquely mapped, mark the slot to be confirmed, and generate query data for confirming the mapping relationship between the fill value and the target slot to the respondent; wherein, the slot to be confirmed represents the target slot that is simultaneously mapped to the same fill value as other target slots.

[0104] In some disclosed embodiments, the slot mapping submodule includes a difference calculation unit for obtaining a confidence difference between the target slot with the highest confidence and the target slot with the second highest confidence based on the ranking result; the slot mapping submodule includes a first determination unit for determining that the fill value uniquely maps to the target slot with the highest confidence in response to the confidence difference satisfying a confidence condition; wherein the confidence condition includes any one of the following: the confidence difference is not lower than a confidence threshold, or the confidence difference is higher than a confidence threshold; the slot mapping submodule includes a second determination unit for determining that the target slot uniquely mapped to the fill value cannot be determined in response to the confidence difference not satisfying a confidence condition.

[0105] In some disclosed embodiments, the scenario question-and-answer management device 60 includes a traceability recording module for recording traceability information for a target slot based on response data; wherein the traceability information is used to verify whether the filling value of the target slot is correct.

[0106] Please see Figure 7 , Figure 7This is a schematic diagram of the framework of an embodiment of the inquiry record query device of this application. The inquiry record query device 70 includes: a record acquisition module 71 and a slot query module 72. The record acquisition module 71 is used to acquire the inquiry record of the respondent in the target scenario; wherein, the inquiry record includes the fill value and traceability information of each target slot in the inquiry template of the target scenario, the traceability information is obtained from the response data of the respondent during the inquiry process, and the fill value is obtained from the above-mentioned scenario question and answer management device embodiment; the slot query module 72 is used to respond to the query command for any target slot in the inquiry record and output the fill value and traceability information of the target slot.

[0107] In the above solution, the inquiry record query device 70 obtains the inquiry record of the respondent in the target scenario. The inquiry record includes the fill value and traceability information of each target slot in the inquiry template of the target scenario. The traceability information is obtained from the respondent's response data during the inquiry process, and the fill value is obtained from the above-mentioned scenario question and answer management device embodiment. Then, in response to the query instruction for any target slot in the inquiry record, the fill value and traceability information of the target slot are output. Therefore, with the support of collecting information as completely as possible during the scenario question and answer process, the query requirements of the inquiry record can be met.

[0108] In some disclosed embodiments, the query instruction is triggered and generated on the query interface. The slot query module is specifically used to display the target slot and its fill value in the first area of ​​the query interface, and to display the traceability information of the target slot in the second area of ​​the query interface.

[0109] In some disclosed embodiments, the query record query device 70 further includes an information review module, which is used to record the review information of the target slot in response to the review instruction of the fill value based on the reference traceability information; wherein the review instruction includes any of the following: pass instruction, modify instruction, or pending review instruction.

[0110] Please see Figure 8 , Figure 8 This is a schematic diagram of a framework of an embodiment of the electronic device of this application. The electronic device 80 includes at least a memory 81 and a processor 82 coupled to each other. The memory 81 stores at least program instructions, and the processor 82 is used to execute the program instructions to implement the steps in any of the above-described scenarios of question-and-answer management method embodiments, or the steps in any of the above-described queries record query method embodiments. For details, please refer to the foregoing disclosed embodiments, which will not be repeated here. It should be noted that the electronic device 80 can be a computer device that executes computer program instructions. Exemplarily, the electronic device 80 can include, but is not limited to, a computer, a dual-screen translator, a single-screen translator, etc., and the specific type of the electronic device 80 is not limited here.

[0111] Specifically, processor 82 controls itself and memory 81 to implement the steps in any of the above-described scenarios of the question-and-answer management method embodiments, or the steps in any of the above-described queries record query method embodiments. Processor 82 can also be called a CPU (Central Processing Unit). Processor 82 may be an integrated circuit chip with signal processing capabilities. Processor 82 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. Furthermore, processor 82 can be implemented using integrated circuit chips.

[0112] In the above scheme, the electronic device 80 acquires an inquiry template for the target scenario. The inquiry template contains several standard questions and target slots for each standard question to be triggered and filled. Based on the inquiry template, it selects untriggered standard questions to determine the current question. Then, it analyzes the respondent's response data to the current question to obtain the filling result of the response data on the slot set. The slot set includes the target slots for each standard question. In response to a related scenario where the filling result includes at least one target slot's filling value, the electronic device 80 first queries the respondent based on the inquiry template of the related scenario until all target slots in the related scenario's inquiry template are filled. Then, based on the filling result, it selects a question from the inquiry template of the target scenario to obtain a new current question and returns the step of analyzing the respondent's response data to the current question to obtain the filling result of the response data on the slot set. The process continues until all target slots in the query template for the target scenario are filled. This minimizes human intervention during the scenario-based question-and-answer process, reducing the risk of data omissions due to subjective biases. Furthermore, if the filling result detects that at least one target slot's value relates to a related scenario, the system first queries the respondent based on the related scenario's query template until all target slots in that template are filled. Then, it continues querying the respondent based on the target scenario's query template until all target slots in the target scenario's template are filled. This approach not only collects information about the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information about those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based question-and-answer process. Furthermore, the system acquires the query records of the respondent in the target scenario, and the query records contain the fill values ​​and traceability information of each target slot in the query template of the target scenario. The traceability information is obtained from the respondent's response data during the query process, and the fill values ​​are obtained from the above-mentioned scenario question and answer management method embodiment. Then, in response to the query instruction for any target slot in the query record, the system outputs the fill value and traceability information of the target slot. Therefore, with the support of collecting information as completely as possible during the scenario question and answer process, the system can meet the query requirements of the query records.

[0113] Please see Figure 9 , Figure 9 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. The computer-readable storage medium 90 stores program instructions 91 that can be executed by a processor. The program instructions 91 are used to implement the steps in any of the above-described scenarios of the question-and-answer management method embodiments, or to implement the steps in any of the above-described queries record query method embodiments.

[0114] In the above scheme, the computer-readable storage medium 90 acquires the query template of the target scenario. The query template contains several standard questions and target slots for each standard question to be triggered for filling. Based on the query template, it selects standard questions that have not been triggered to determine the current question. Then, it analyzes the response data of the respondent to the current question to obtain the filling result of the response data on the slot set. The slot set contains the target slots for each standard question. In response to a related scenario where the filling result includes at least one target slot filling value involving the target scenario, the respondent is first queried based on the query template of the related scenario until all target slots in the query template of the related scenario are filled. Then, based on the filling result, a question is selected in the query template of the target scenario to obtain a new current question. Finally, it returns the result of the filling of the slot set based on the response data of the respondent to the current question. The process involves several steps, continuing until all target slots in the query template for the target scenario are filled. This minimizes human intervention during the scenario-based question-and-answer process, reducing the risk of data loss due to subjective biases. Furthermore, if the filling result indicates that at least one target slot's value relates to a related scenario, the system first queries the respondent using the related scenario's query template until all target slots in that template are filled. Then, it continues querying the respondent using the target scenario's query template until all target slots in the target scenario's template are filled. This approach not only collects information about the target scenario itself but also proactively detects whether the respondent's response data relates to related scenarios and, if so, collects additional information about those related scenarios. Therefore, it maximizes the completeness of information collection during the scenario-based question-and-answer process. Furthermore, the system acquires the query records of the respondent in the target scenario, and the query records contain the fill values ​​and traceability information of each target slot in the query template of the target scenario. The traceability information is obtained from the respondent's response data during the query process, and the fill values ​​are obtained from the above-mentioned scenario question and answer management method embodiment. Then, in response to the query instruction for any target slot in the query record, the system outputs the fill value and traceability information of the target slot. Therefore, with the support of collecting information as completely as possible during the scenario question and answer process, the system can meet the query requirements of the query records.

[0115] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0116] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0119] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the user through pop-up information or by asking the user to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A scenario-based question-and-answer management method, characterized in that, include: Obtain the query template for the target scenario; wherein, the query template contains several standard questions and target slots that are triggered by each of the standard questions to be filled; Based on the query template, select the standard questions that have not been triggered to determine the current question; Based on the analysis of the respondent's response data to the current question, the result of the response data filling the slot set is obtained; wherein, the slot set contains the target slots for each of the standard questions; In response to the filling result including at least one of the target slots having a filling value related to the target scene's associated scene, the respondent is first queried based on the associated scene's query template until all the target slots in the associated scene's query template are filled. Then, based on the filling result, a question is selected in the target scene's query template to obtain a new current question. The step of analyzing the respondent's response data to the current question to obtain the filling result of the response data for the slot set is returned until all the target slots in the target scene's query template are filled.

2. The method according to claim 1, characterized in that, The method further includes: In response to the filling result including any filling value of the target slot not involving the associated scenario of the target scenario, a question is selected from the query template of the target scenario based on the filling result to obtain a new current question; The process returns to the step of analyzing the response data of the respondent to the current question to obtain the filling result of the slot set by the response data, until all the target slots in the query template of the target scenario are filled.

3. The method according to claim 1 or 2, characterized in that, The step of selecting a new current question from the query template of the target scenario based on the filling result includes: Based on the filling result, the response data is identified to obtain the dialogue type of the response data; wherein, the dialogue type includes at least one of the following: a template answer to the current question, an advance answer to at least part of the target slots corresponding to a standard question that has not yet been asked, and flow control; Based on the dialogue type, a question is selected from the query template of the target scenario to obtain the new current question.

4. The method according to claim 3, characterized in that, The step of selecting a question from the query template in the target scenario based on the dialogue type to obtain the new current question includes at least one of the following: In response to the dialogue type including the template answer, the new current question is obtained by sequentially selecting from the standard questions that have not been triggered in the query template. In response to the dialogue type including the early response, the standard questions that have not yet been asked but have been answered in advance are skipped based on the inquiry template, and the new current question is obtained; In response to the dialogue type including the flow control, inapplicable standard questions are skipped based on the query template to obtain the new current question.

5. The method according to claim 3, characterized in that, The dialogue types also include at least one of the following: free supplementation related to the target scenario but not a direct answer to the current question, filling and correcting already filled target slots, proactively asking questions, and deviating from the topic. The method also includes: In response to the dialogue type including the free supplement, the respondent is prompted to answer the current question; In response to the dialogue type including the fill correction, firstly based on the response data, generate question data for confirming the fill correction with the respondent, and then prompt the respondent to answer the current question; In response to the dialogue type including the proactive question, firstly, based on the response data, generate answer data for replying to the respondent, and then prompt the respondent to answer the current question; In response to the dialogue type including the off-topic, the respondent is prompted to answer the current question.

6. The method according to claim 3, characterized in that, Before selecting a question from the query template in the target scenario based on the dialogue type to obtain the new current question, the method further includes at least one of the following: In response to detecting that a portion of the target slots for the current question are not yet filled, follow-up question data is generated to ask the respondent about the missing slots; wherein, the missing slots represent the portion of the target slots for the current question that are not yet filled; In response to the detection that a target slot that has already been filled has been newly filled with a different fill value, a conflicting slot is marked, and the staff is prompted to confirm the conflicting slot with the person being questioned; wherein, the conflicting slot represents a target slot that has already been filled with the fill value but is now filled with a different fill value; In response to the detection that the fill value of the target slot does not conform to the format conditions of the target slot, the format slot is marked, and the staff is prompted to confirm the format slot with the person being questioned; wherein, the format slot represents the target slot where the fill value does not conform to the format conditions; In response to detecting that the fill value of the target slot is associated with an unfilled target slot, the associated slot is marked, and question data for asking the respondent about the associated slot is generated; wherein, the associated slot represents a target slot that is not yet filled but is associated with a fill value that has already been filled.

7. The method according to claim 1, characterized in that, The step of analyzing the response data of the respondent to the current question to obtain the filling result of the slot set based on the response data includes: Based on the response data, entity recognition, field extraction, and slot mapping are performed sequentially to obtain the execution result of the response data; In response to the execution result indicating the existence of fill values ​​simultaneously mapped to multiple target slots, the confidence level of each fill value mapped to multiple target slots is obtained. Based on the confidence level, the multiple target slots simultaneously mapped by the fill value are sorted to determine whether the fill value can uniquely map to a target slot. If a uniquely mapped target slot cannot be determined, a slot to be confirmed is marked, and query data is generated to confirm the mapping relationship between the fill value and the target slot with the respondent. The slot to be confirmed represents a target slot that is simultaneously mapped to the same fill value as other target slots.

8. The method according to claim 7, characterized in that, The sorting result of sorting multiple target slots simultaneously mapped to the fill value based on the confidence level, and determining whether the fill value can uniquely map to a target slot, includes: Based on the sorting results, the confidence difference between the target slot with the highest confidence level and the target slot with the second highest confidence level is obtained; In response to the confidence difference satisfying the confidence condition, it is determined that the fill value is uniquely mapped to the target slot with the highest confidence level; wherein, the confidence condition includes any one of the following: the confidence difference is not lower than the confidence threshold, or the confidence difference is higher than the confidence threshold; If the confidence difference does not meet the confidence condition, the target slot that the fill value uniquely maps to cannot be determined.

9. The method according to any one of claims 1 to 8, characterized in that, Also includes: Based on the response data, traceability information is recorded for the target slot; wherein, the traceability information is used to verify whether the filling value of the target slot is correct.

10. A method for querying inquiry records, characterized in that, include: Obtain the inquiry record of the respondent in the target scenario; wherein the inquiry record includes the fill value and traceability information of each target slot in the inquiry template of the target scenario, the traceability information is obtained from the response data of the respondent during the inquiry process, and the fill value is obtained by the scenario question and answer management method according to any one of claims 1 to 9; In response to a query command for any of the target slots in the query record, the fill value and traceability information of the target slot are output.

11. The method according to claim 10, characterized in that, The query command is generated by triggering the query interface, and the output of the target slot's fill value and traceability information includes: The target slot and its fill value are displayed in the first area of ​​the query interface, and the traceability information of the target slot is displayed in the second area of ​​the query interface. And / or, the method further includes: in response to an audit instruction on the fill value with reference to the traceability information, recording audit information of the target slot; wherein the audit instruction includes any of the following: pass instruction, modify instruction, or pending review instruction.

12. A scenario-based question-and-answer management device, characterized in that, include: The template acquisition module is used to acquire the query template for the target scenario; wherein, the query template contains several standard questions and target slots that are triggered by each of the standard questions to be filled; The problem determination module is used to select the standard problems that have not been triggered based on the query template and determine the current problem; The response analysis module is used to analyze the response data of the respondent to the current question to obtain the filling result of the slot set of the response data; wherein, the slot set includes the target slots of each of the standard questions; The process switching module is used to respond to the fact that the filling result includes at least one of the target slots' filling values ​​involving the target scenario's associated scenarios. First, it queries the respondent based on the query template of the associated scenarios until all the target slots in the query template of the associated scenarios are filled. Then, based on the filling result, it selects a question in the query template of the target scenario to obtain a new current question and returns to the step of analyzing the respondent's response data to the current question to obtain the filling result of the response data for the slot set, until all the target slots in the query template of the target scenario are filled.

13. An inquiry record query device, characterized in that, include: The record acquisition module is used to acquire the inquiry record of the respondent in the target scenario; wherein, the inquiry record includes the fill value and traceability information of each target slot in the inquiry template of the target scenario, the traceability information is obtained from the response data of the respondent during the inquiry process, and the fill value is obtained by the scenario question and answer management device as described in claim 12. The slot query module is used to respond to a query command for any of the target slots in the query record and output the fill value and traceability information of the target slot.

14. An electronic device, characterized in that, It includes at least a memory and a processor coupled to each other, wherein the memory stores at least program instructions, and the processor is used to execute the program instructions to implement the scenario question-answering management method according to any one of claims 1 to 9, or the question record query method according to any one of claims 10 to 11.

15. A computer-readable storage medium, characterized in that, The system stores program instructions that can be executed by a processor, the program instructions being used to implement the scenario question-answering management method according to any one of claims 1 to 9, or the question record query method according to any one of claims 10 to 11.

Images