Non-perpetual data natural language large model interaction method

Abstract

The invention is suitable for the technical field of non-perpetual data intelligent processing and natural language large model interaction, and provides a non-perpetual data natural language large model interaction method, which comprises the following steps: after a large model provides the answer content of the query to a target user, according to a reference, adopting a conversion link, and providing an operation suggestion approaching the reference to adopt the conversion link for the target user in the content selection and application process. The invention creatively reveals that different query states can cause significant difference of application effects and a significant interval with an optimal application effect exists under a fixed adoption conversion link of a user. On the basis of historical interaction records, the optimal query state index intervals of different comparison users can be extracted, and the optimal query state index interval of which comparison user the current query state of the target user falls into is determined when the target user queries, so that the adoption conversion link corresponding to the comparison user is taken as a reference.

Description

Technical Field

[0001] This invention belongs to the field of intelligent processing of intangible cultural heritage data and interaction technology of natural language large model, and particularly relates to a method for interaction of intangible cultural heritage data with natural language large model. Background Technology

[0002] In the process of digital protection and transmission of intangible cultural heritage (ICH), natural language processing and big data analysis have become important technical means. Existing large-scale ICH data models can provide knowledge Q&A, data compilation, process descriptions, and multimodal displays based on natural language queries, which to some extent meets users' needs for retrieving and learning ICH information. However, these technologies mostly focus on static information matching and single-response generation, usually only outputting complete process flow or knowledge content, lacking in-depth analysis and targeted guidance based on users' dynamic interactive behavior.

[0003] Practice shows that when faced with intangible cultural heritage craft data involving multiple steps and stages, users often perform a first screening after receiving the complete answer, followed by a second screening and practical application of the remaining data, forming a relatively fixed adoption and conversion chain. While existing technologies can record users' query history, they struggle to accurately capture real-time states such as input rhythm, emotional inclination, and information demand intensity during the query process. Furthermore, they fail to reveal the differences in application effectiveness resulting from a fixed adoption and conversion chain under different query states, and lack the ability to provide dynamic, personalized guidance based on these differences. Summary of the Invention

[0004] The purpose of this invention is to provide a method for interaction of natural language large-scale models of intangible cultural heritage data, which aims to solve the problems mentioned in the background art.

[0005] This invention is implemented as follows: a method for interaction of a large-scale natural language model for intangible cultural heritage data, the method comprising:

[0006] Obtain the current query category of the target user for the specified category of intangible cultural heritage data, and retrieve the historical interaction records of the large model;

[0007] Select several comparison users with the same background as the target user from the historical interaction records, extract the historical interaction samples of the comparison users under the current query category, and determine the adoption conversion link of each comparison user and the query status indicators and application effect indicators of each historical interaction sample.

[0008] The application performance indicators are sorted according to the query status indicators to obtain the application performance change trends of different comparison users, identify the significant intervals in the application performance change trends, and determine the optimal query status indicator intervals corresponding to the significant intervals.

[0009] Obtain the query status index of the target user for this query, determine the optimal query status index range in which it falls, and determine the adoption conversion link of the comparison user corresponding to the optimal query status index range as the reference adoption conversion link;

[0010] After the large model provides the target user with the answer to the query, it provides operational suggestions to the target user in the process of content selection and application, based on the reference adoption conversion link.

[0011] As a further limitation of the technical solution of the present invention, the specified category of intangible cultural heritage data refers to intangible cultural heritage data of traditional crafts and process flow.

[0012] As a further limitation of the technical solution of the present invention, "consistent with user background" means that the user's age, education or employment status, regional cultural environment, and knowledge level related to the specified category of intangible cultural heritage are consistent with the target user.

[0013] As a further limitation of the technical solution of the present invention, the adoption and transformation link refers to the process in which, under the query of intangible cultural heritage data of a specified category, the user extracts and filters the answer content provided by the large model based on their own autonomous selection characteristics, and obtains a first selection ratio relative to all answer content to form a first-layer link. Then, the extracted content is further put into practical application according to a second selection ratio relative to the extracted content to form a second-layer link, thereby constituting a two-layer continuous adoption and transformation process.

[0014] As a further limitation of the technical solution of the present invention, the acquisition of the query status index includes a comprehensive quantification of the user's query focus, emotional state and information demand intensity based on the user's input rhythm, semantic features, emotional tendency and attention duration during the query process.

[0015] The acquisition of the application effectiveness indicators includes a comprehensive quantification of the effectiveness of the content application based on the actual results, effect feedback, quality assessment results, and subsequent improvements or lasting impacts generated by users after applying the answer content.

[0016] As a further limitation of the technical solution of this invention, the steps of sorting application performance indicators according to query status indicators, obtaining application performance change trends for different comparison users, identifying significant intervals in the application performance change trends, and determining the optimal query status indicator interval corresponding to the significant intervals include:

[0017] The different historical interaction samples of each comparison user are sorted according to the numerical order of the query status index to form a sorted historical interaction sample sequence, and the trend of application effect changes is obtained accordingly.

[0018] Identify significant intervals in the application performance change trends of each comparison user;

[0019] The query status indicator interval corresponding to the significant interval is mapped in the identification of the application effect change trend, and this query status indicator interval is set as the optimal query status indicator interval.

[0020] As a further limitation of the technical solution of the present invention, the significant interval refers to the segment located at the peak and its vicinity in the trend of application effect change.

[0021] As a further limitation of the technical solution of this embodiment of the invention, after the large model provides the target user with the answer to the query, the steps of providing the target user with operational suggestions that approximate the reference adoption conversion link during the content selection and application process, based on the reference adoption conversion link, include:

[0022] After the large model provides the target user with the answer to this query, the target user's operation process on the answer content is continuously observed;

[0023] The system obtains the reference adoption conversion link, identifies the first reference selection ratio and the second reference selection ratio, and generates a first prompt message based on the difference between the actual selection and the first reference selection ratio when the target user selects the answer content. It also obtains the answer content after the target user selects the answer content.

[0024] A second prompt message is generated based on the difference between the user's actual application of the selected answer and the proportion of the second selection.

[0025] As a further limitation of the technical solution of the present invention, when observing the actual selection and application of the target user, the deviation range is quantified based on the difference or percentage difference between the target user and the first and second selection ratios, and suggestion information is generated.

[0026] As a further limitation of the technical solution of this embodiment of the invention, in the process of generating the second prompt information based on the difference between the target user's actual application of the selected answer content and the reference second selection ratio, if it is detected that the target user has not made adjustments according to the first prompt information, the reference second selection ratio is enlarged or reduced accordingly, and the adjusted ratio is used as the basis for generating the second prompt information to guide the target user's subsequent application.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] This invention innovatively reveals that, by establishing a correspondence between query status indicators and application effect indicators, different query states lead to significant differences in application effect within a user's fixed adoption and conversion path, and that there exists a significant interval for optimal application effect. Based on historical interaction records, this invention can extract the optimal query status indicator intervals for different comparison users, and when a target user queries, determine which comparison user's optimal query status indicator interval their current query state falls into, thus using the adoption and conversion path corresponding to that comparison user as a reference.

[0029] After the large model provides the answer, the system dynamically compares the differences in the proportions of the user's first and second selections based on the reference link, and generates hierarchical prompts. When necessary, it adaptively adjusts the reference proportions to achieve real-time personalized guidance for different query states. This solution significantly improves the utilization rate and practical application effectiveness of the answer content, providing a new technical path for the digital inheritance of traditional crafts and similar intangible cultural heritage categories with prominent multi-step and multi-stage characteristics, demonstrating outstanding novelty and application prospects. Attached Figure Description

[0030] Figure 1 A flowchart of the method provided in the embodiments of the present invention;

[0031] Figure 2 A flowchart for identifying the trend of application effect changes and determining the optimal query state index range in the method provided in the embodiments of the present invention;

[0032] Figure 3 A flowchart is provided for the approach reference adoption transformation link operation suggestion in the method provided in the embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0034] Figure 1 A flowchart of the method provided by an embodiment of the present invention is shown.

[0035] Specifically, a method for interaction of intangible cultural heritage data natural language large model includes the following steps:

[0036] Step S100: Obtain the current query category of the target user for the specified category of intangible cultural heritage data, and retrieve the historical interaction records of the large model.

[0037] The specified category of intangible cultural heritage data refers to intangible cultural heritage data related to traditional crafts and processes.

[0038] In this embodiment of the invention, the large-scale model is a natural language model specifically designed for intangible cultural heritage data. It can employ various deep learning language model architectures commonly found in existing technologies, such as a Transformer-based conversational large-scale model, a knowledge graph-enhanced large-scale model, or a hybrid retrieval and generation large-scale model. This large-scale model has a pre-built and continuously updated knowledge system for the intangible cultural heritage data, enabling it to understand and generate natural language content related to the data. The target user refers to the object interacting with the large-scale model and possessing a history of multiple interactions, allowing for the extraction of stable behavioral and demand characteristics from its accumulated query and feedback data.

[0039] In addition to traditional crafts and process flows, the designated categories of intangible cultural heritage data can also include traditional food techniques that also have multi-step process characteristics.

[0040] The query category refers to a more detailed and actionable specific process step or procedure within a specified category. For example, in the category of traditional processes and processes, it can be further subdivided into dye preparation, loom setup and finishing in textile dyeing, body forming, glaze preparation and firing curve control in ceramic firing, and mold preparation, melting temperature control and cooling forming in metal casting. Since these intangible cultural heritage processes often involve multiple steps, large models typically cover the entire process when generating responses. However, target users often only perform an initial screening of key steps after receiving the response, and then conduct a second screening and practical application within those selected steps, thus forming two layers of continuous selection and application features. This invention combines this practical operational pattern, supporting large models to provide users with more targeted responses and interactions at the level of specific subdivided steps.

[0041] The historical interaction records include not only the target user's past interaction data but also similar query data from other users who have interacted with the large model. Specifically, this data can be obtained from the server-side database deployed with the large model, cloud storage, or client-side cache interacting with the large model, and can be synchronized from associated intangible cultural heritage databases or third-party big data platforms via interfaces. The historical interaction records should contain at least the following types of specific data: time, location, and device information for each query; the original text and key semantic features of the query input; the model's output response content and its confidence level or weight information; the user's browsing, selection, re-questioning, and subsequent feedback behavior regarding the response; and behavioral characteristics related to the query, such as emotions and attention duration. By extracting and analyzing this data, sufficient historical data support can be provided for subsequently determining query status indicators, application effect indicators, and adoption conversion paths.

[0042] Furthermore, the method for interaction of intangible cultural heritage data with natural language processing also includes the following steps:

[0043] Step S200: Select several comparison users with the same background as the target user from the historical interaction records, extract the historical interaction samples of the comparison users under the current query category, and determine the adoption conversion link of each comparison user and the query status indicators and application effect indicators of each historical interaction sample.

[0044] The phrase "consistent with user background" refers to comparing the user's age group, education or employment status, regional cultural environment, and knowledge level related to the specified category of intangible cultural heritage with the target user's background.

[0045] The adoption and conversion link refers to the process in which, under the query scenario of intangible cultural heritage data of a specified category, users, based on their own autonomous selection characteristics, first extract and filter the answer content provided by the large model to obtain the first selection ratio relative to all answer content to form the first link, and then further put the extracted content into practical application according to the second selection ratio relative to the extracted content to form the second link, thus constituting a two-layer continuous adoption and conversion process.

[0046] The acquisition of the query status indicators includes a comprehensive quantification of the user's query focus, emotional state, and information demand intensity based on the user's input rhythm, semantic features, emotional tendencies, and attention duration during the query process.

[0047] The acquisition of the application effectiveness indicators includes a comprehensive quantification of the effectiveness of the content application based on the actual results, effect feedback, quality assessment results, and subsequent improvements or lasting impacts generated by users after applying the answer content.

[0048] In this embodiment of the invention, in step S200, it is necessary to select several comparison users with backgrounds consistent with the target user from historical interaction records to provide a highly comparable reference sample for subsequent analysis. Background consistency means that the comparison users are within a similar range to the target user in key background conditions such as age, education or employment status, regional cultural environment, and knowledge level related to the specified category and its specific subcategories of intangible cultural heritage. This consistency does not require complete identicalness, but allows for a certain range of similarity to ensure representativeness and sufficient data volume in the selection process. To ensure statistical reliability, the number of comparison users should reach a set minimum scale, for example, no less than several dozen, and this rigorous selection can be supported by the massive amount of interaction data accumulated during the long-term operation of a large model. By limiting the consistency to the level of intangible cultural heritage-related knowledge in specific subcategories, it can be ensured that the comparison users and the target user have comparable understanding and operational capabilities when facing similar intangible cultural heritage issues, thereby making subsequent behavioral characteristics and effect analyses more targeted.

[0049] The adoption and conversion link refers to a two-layer continuous filtering and application process where users, based on their autonomous selection characteristics, process the responses provided by a large model when querying intangible cultural heritage data of a specified category. The first layer is the extraction and filtering link: after receiving the response, the user performs an initial screening of the steps, solutions, or data, forming an initial selection ratio relative to the entire response. For example, if the model outputs 10 key steps in a ceramic firing process, and the user retains only 7, the initial selection ratio is 70%. The second layer is the application and filtering link: the user further filters the extracted content and actually applies it in practice, forming a second selection ratio relative to the extracted content. For example, if only 4 steps are ultimately executed, the second selection ratio is 4 / 7 ≈ 57%.

[0050] In determining the adoption conversion path, both the initial and secondary selection ratios can be achieved by automatically recording multi-dimensional user behavior data. Besides behaviors such as browsing, saving, downloading, annotating, asking follow-up questions, and providing feedback, this also includes various actual operations performed by users on selected content on computers or other terminals, such as document editing, process modeling, step execution recording, image or video generation and uploading, and calls and modifications in process management software. The system can utilize existing mature technologies such as operation logs, click records, file modification monitoring, application programming interface call logs, and cross-device synchronized data tracking to comprehensively analyze these online and offline operations. This allows for accurate calculation of the actual application level and conversion effect after receiving information, providing more comprehensive and reliable data support for determining the initial and secondary selection ratios.

[0051] The quantification of the query status indicators includes: using natural language processing and behavioral analysis technologies to comprehensively process data such as the user's input rhythm (e.g., typing speed, input interval), semantic features (e.g., vocabulary density, use of technical terms), emotional tendency (identifying positive, neutral, or negative emotions through sentiment analysis models or analyzing emotional data from wearable devices), and attention duration during the query process, in order to assess query focus, emotional state, and information demand intensity. Specifically, existing mature emotion recognition algorithms, natural language understanding models, and user behavior analysis algorithms can be used to automatically calculate and generate numerical query status indicators.

[0052] The application effectiveness index is a comprehensive numerical measure of the user's final application effectiveness of the answer content. Its quantification can combine multi-dimensional user feedback during the practical stage, such as actual results (e.g., the quantity or quality of completed craft works), effect feedback (e.g., post-use satisfaction ratings), quality assessment results (e.g., expert or system-automated scoring), and the resulting subsequent improvements or lasting impact. Technical means can include image recognition (detecting the finished craft works), text analysis (analyzing feedback), and interface data with external evaluation systems to obtain a comprehensive numerical result. A higher application effectiveness index indicates better application effectiveness, which is also an important basis for subsequent ranking and significant interval identification.

[0053] The determination of the adoption conversion path, the quantification of query status indicators, and the quantification of application effect indicators can all be accomplished by relying on the rich historical interaction data recorded by the large model during its long-term operation. This data includes user input text, click behavior, interaction time series, browsing depth, collection and download records, application feedback, and subsequent performance reports, all of which can be obtained from the system database or cloud data warehouse. Through these continuously accumulated historical interaction records, sufficient and reliable support can be provided for data analysis and indicator calculation in the methodology.

[0054] Furthermore, the method for interaction of intangible cultural heritage data with natural language processing also includes the following steps:

[0055] Step S300: Sort the application performance indicators according to the query status indicators to obtain the application performance change trends of different comparison users, identify the significant intervals in the application performance change trends, and determine the optimal query status indicator intervals corresponding to the significant intervals.

[0056] Specifically, Figure 2 The flowchart illustrates the identification of application effect change trends and the determination of the optimal query state index range.

[0057] The process of sorting application performance indicators by query status metrics to obtain application performance trends for different comparison users, identifying significant intervals within these trends, and determining the optimal query status metric interval corresponding to each significant interval includes the following steps:

[0058] Step S301: Sort the different historical interaction samples of each comparison user according to the numerical order of the query status index to form a sorted historical interaction sample sequence, and obtain the application effect change trend based on it.

[0059] Step S302: Identify significant intervals in the application effect change trend of each comparison user. The significant intervals refer to the segments located at the peak and the vicinity of the application effect change trend.

[0060] Step S303: Map the query status indicator interval corresponding to the significant interval in the identified application effect change trend, and set the query status indicator interval as the optimal query status indicator interval.

[0061] In this embodiment of the invention, in step S301, different historical interaction samples of each comparison user are sorted according to the numerical order of the query status index to form a sorted historical interaction sample sequence, and the application effect change trend is obtained accordingly. In specific implementation, mature database retrieval and sorting algorithms, time series analysis, data normalization and visualization processing technologies can be used to match the query status index of each interaction with the corresponding application effect index and arrange them in the order of increasing or decreasing index values ​​to obtain a continuous curve or segmented trend chart of the application effect changing with the query status, providing basic data support for subsequent identification of significant intervals.

[0062] In step S302, significant intervals are identified in the application effect change trend of each compared user. This can be accomplished using statistical peak detection algorithms, sliding window methods, inflection point analysis, or machine learning-based anomaly segment identification algorithms. The application effect curve is scanned to automatically detect segments located at and near the peak. The existence of significant intervals indicates that, within a certain continuous range of query status indicators, the application effect reaches relative optimality when the user follows their fixed adoption conversion path. This reveals a stable coupling relationship between query status and actual application effect from a data perspective.

[0063] In step S303, the identified salient intervals are mapped to corresponding query status indicator intervals, and these intervals are set as the optimal query status indicator intervals. Specifically, this can be achieved by combining interval mapping algorithms, data clustering and matching algorithms, and statistical analysis to confirm the query status values ​​at both ends of the salient intervals, automatically generating the optimal query status indicator interval. The optimal query status indicator interval represents the range of query states that achieves the highest application effect in historical samples, and is a key basis for subsequently determining whether the target user's current state is under efficient application conditions.

[0064] This invention reveals a core technological discovery: for users of a specific category, a relatively fixed adoption conversion path typically forms based on their habits and usage patterns, exhibiting a relatively stable first and second selection ratio in the two-layer screening process. However, existing research indicates that even with the same adoption conversion path, the final application effect will differ for the same user under different query states. Furthermore, data analysis reveals a significant interval where the application effect is optimal when using the established adoption conversion path within the corresponding query state index range—that is, when interacting with a large model to obtain data on a specific category of intangible cultural heritage. Based on this pattern, those skilled in the art can infer that when a target user's current query state index falls within the optimal query state index range for comparable users with similar backgrounds, the corresponding adoption conversion path is also most suitable for the current user's query state and can be directly used as the basis for subsequent interaction suggestions.

[0065] The novelty of this technological concept lies in its approach: instead of making static recommendations based solely on users' historical behavior, it combines the dynamic changes in query status with the peak patterns of application effects to identify the optimal query status index range and dynamically match it with the corresponding adoption and conversion links, achieving personalized decision-making based on time and context. Its applicability is reflected in its ability to extract the coupling patterns between query status and application effects from historical big data, whether for traditional crafts and process flows or other intangible cultural heritage categories with multi-step processes. This allows the natural language processing model to provide more accurate content filtering and application suggestions that better meet users' actual operational needs in real-time interactions.

[0066] Furthermore, the method for interaction of intangible cultural heritage data with natural language processing also includes the following steps:

[0067] Step S400: Obtain the query status index of the target user's current query, determine the optimal query status index range in which it falls, and determine the adoption conversion link of the comparison user corresponding to the optimal query status index range as the reference adoption conversion link.

[0068] Step S500: After the large model provides the answer to the query to the target user, based on the reference adoption conversion link, it provides the target user with operation suggestions that approach the reference adoption conversion link in the process of content selection and application.

[0069] Specifically, Figure 3 A flowchart is shown that provides operational suggestions for the approach reference adoption transformation link.

[0070] After the large model provides the target user with the answer to the query, it provides operational suggestions to the target user in the process of content selection and application that approximate the reference adoption conversion link, based on the reference adoption conversion link. Specifically, this includes the following steps:

[0071] Step S501: After the large model provides the answer to the query to the target user, the target user's operation process on the answer is continuously observed.

[0072] Step S502: Obtain the reference adoption conversion link, identify the first reference selection ratio and the second reference selection ratio, generate the first prompt information based on the difference between the actual selection and the first reference selection ratio when the target user selects the answer content, and obtain the answer content after the target user selects the answer content.

[0073] Step S503: Generate a second prompt message based on the difference between the user's actual application of the selected answer and the proportion of the second selection.

[0074] When observing the actual selection and application of target users, suggestions are generated by quantifying the deviation magnitude based on the difference or percentage difference between the target user's selection ratio and the first and second selection ratios.

[0075] In the process of generating a second prompt based on the difference between the target user's actual application of the selected answer and the reference second selection ratio, if it is detected that the target user has not adjusted according to the first prompt, the reference second selection ratio is enlarged or reduced accordingly, and the adjusted ratio is used as the basis for generating the second prompt to guide the target user's subsequent application.

[0076] In this embodiment of the invention, in step S400, by acquiring the query status index of the target user's current query and determining whether it falls within the previously identified optimal query status index range, the core technical point of the invention is connected. The purpose of this step is to directly correlate the target user's current dynamic query status with the optimal historical pattern selected from different comparison users, thereby determining the adoption conversion link of the comparison user corresponding to the optimal query status index range as the reference adoption conversion link for the current target user. In this way, it can be ensured that the subsequent filtering and application suggestions conform to the target user's optimal behavioral patterns in the current query context, achieving personalized and dynamic accurate recommendations.

[0077] In step S500, after the large model provides the target user with the answer to the query, it offers operational suggestions to the target user in the content selection and application process, based on the reference adoption conversion link. Specifically, this includes the following steps:

[0078] In step S501, the system continuously observes the target user's actions regarding the answer content. Technically, mature technologies such as front-end interaction logs, page dwell time, click paths, scrolling trajectories, and multi-device synchronous recording can be used to monitor the user's browsing, saving, annotation, and filtering operations on the answer content in real time, thereby accurately tracking the user's actual selection in the first-level filtering stage.

[0079] In step S502, the system invokes the established reference adoption conversion path and identifies the first and second reference selection ratios. When the user performs the first selection, the system compares their real-time selection ratio with the first reference selection ratio. For example, if the first reference selection ratio is 70% and the user has only selected 50%, the system immediately generates a first prompt message, reminding the user to appropriately increase the selection range to improve subsequent application effectiveness. Then, the system obtains the set of answers generated after the user completes the first selection, providing a basis for the next processing step.

[0080] In step S503, when the user applies the content selected in the first round, the system compares the actual application rate with the reference rate from the second round of selection. For example, if the reference rate for the second round of selection is 60%, while the user's actual application rate is only 40%, the system will generate a second prompt message, suggesting that the user increase investment or expand the scope of implementation in the next step. This process can be achieved by combining cloud log recording, IoT device feedback, and data interfaces of online editing and execution systems, thereby ensuring accurate tracking of the actual application process.

[0081] In the process of generating a second prompt based on the difference between the target user's actual application of the selected content and the reference second selection ratio, if it is detected that the user has not adjusted according to the first prompt, the reference second selection ratio is increased or decreased accordingly, and the adjusted ratio is used as the basis for generating the second prompt to guide the user's subsequent application. The significance of this setting is to provide adaptive dynamic guidance, avoiding deviations from reality in subsequent guidance due to the user's failure to follow the first prompt. For example, when the reference second selection ratio is 60% and the user's first selection ratio is slightly lower than the suggested value, the system can automatically increase the reference second selection ratio from 40% to 45% to adapt to the user's current actual selection base and generate a more realistic second prompt, thereby guiding the user to increase content selection and investment in subsequent applications.

[0082] The present invention offers the following advantages: By introducing dynamic analysis of query status indicators and application effect indicators, and combining this with the optimal query status indicator range mined from historical interaction data, it achieves accurate identification of the coupling pattern between user query status and actual application effect. Furthermore, by providing multi-level dynamic prompts based on the adoption conversion chain, it not only guides users to approach the historically optimal conversion model in content selection and practical application, but also automatically adjusts the prompt strategy according to the user's real-time operation.

[0083] In terms of application prospects, this invention can be widely applied to the intelligent protection, revitalization, and inheritance of intangible cultural heritage data, especially suitable for traditional crafts and processes, as well as other intangible cultural heritage categories with multi-step and multi-stage characteristics. For example, in scenarios such as folk ceramic firing, textile dyeing and weaving, metal processing, and traditional food preparation, the large model can not only provide complete process knowledge but also guide practitioners or researchers to gradually adopt and practice according to the optimal path, improving learning efficiency and practical effectiveness. In the future, it can be further expanded to various digital application scenarios of intangible cultural heritage, such as education and training, cultural tourism, and digital inheritance platforms, providing strong technical support for the dissemination and innovation of intangible cultural heritage skills.

[0084] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-heritage data natural language large model interaction method, characterized in that, The method comprises: acquiring the current query category of the target user for the specified category of non-heritage data, and calling the historical interaction records of the large model; selecting several comparison users consistent with the background of the target user from the historical interaction records, extracting the historical interaction samples of the comparison users under the current query category, determining the adoption conversion link of each comparison user and the query state indicators and application effect indicators of each historical interaction sample; sorting the application effect indicators according to the query state indicators to obtain the application effect change trend of different comparison users, identifying the significant interval existing in the application effect change trend, and determining the best query state indicator interval corresponding to the significant interval; acquiring the query state indicator of the current query of the target user, determining the best query state indicator interval it falls into, and determining the adoption conversion link of the comparison user corresponding to the best query state indicator interval as the reference adoption conversion link; after the large model provides the answer content of the current query to the target user, providing operation suggestions for the target user to approach the reference adoption conversion link in the content selection and application process according to the reference adoption conversion link.

2. The non-heritage data natural language large model interaction method according to claim 1, characterized in that, The specified category of non-heritage data refers to traditional crafts and process flow type non-heritage data.

3. The non-heritage data natural language large model interaction method according to claim 1, characterized in that, The consistency with the user background refers to the consistency of the comparison user with the target user in terms of age level, learning or working conditions, regional cultural environment and knowledge level related to the specified category of non-heritage, which are key background conditions affecting query and application.

4. The non-heritage data natural language large model interaction method according to claim 2, characterized in that, The adoption conversion link refers to the adoption and conversion process of two layers in sequence, that is, the first selection ratio of the extracted content relative to the total answer content forms the first layer link, and the second selection ratio of the extracted content relative to the extracted content is further put into actual application to form the second layer link, thereby forming the adoption and conversion process of two layers in sequence.

5. The non-heritage data natural language large model interaction method according to claim 4, characterized in that, The acquisition of the query state indicators comprises comprehensively quantifying the query focus, emotional state and information demand intensity of the user based on the input rhythm, semantic features, emotional tendency and attention duration of the user in the query process; The acquisition of the application effect indicators comprises comprehensively quantifying the content application effect based on the actual results, effect feedback, quality evaluation results and subsequent improvement or continuous influence generated after the user applies the answer content.

6. The non-heritage data natural language large model interaction method according to claim 5, characterized in that, The steps of sorting the application effect indicators according to the query state indicators to obtain the application effect change trend of different comparison users, identifying the significant interval existing in the application effect change trend, and determining the best query state indicator interval corresponding to the significant interval comprise: sorting the different historical interaction samples of each comparison user according to the numerical order of the query state indicators to form a sequence of sorted historical interaction samples, and obtaining the application effect change trend accordingly; identifying the significant interval existing in the application effect change trend of each comparison user; mapping the query state indicator interval corresponding to the significant interval in the identified application effect change trend, and setting the query state indicator interval as the best query state indicator interval.

7. The non-heritage data natural language large model interaction method according to claim 6, characterized in that, The significant interval refers to a section located at a peak value and its adjacent range in an application effect change trend.

8. The non-heritage data natural language large model interaction method according to claim 6, characterized in that, After the large model provides the answering content of the current query to the target user, the operation suggestion of the approaching reference adoption conversion link is provided to the target user in the content selection and application process according to the reference adoption conversion link. After the large model provides the answering content of the current query to the target user, the operation process of the target user on the answering content is continuously observed. The reference adoption conversion link is acquired, the reference first selection proportion and the reference second selection proportion are identified, the first prompt information is generated according to the difference between the actual selection of the target user and the reference first selection proportion when the target user selects the answering content, and the answering content selected by the target user is acquired. The second prompt information is generated according to the difference between the actual application of the selected answering content by the user and the reference second selection proportion.

9. The non-heritage data natural language large model interaction method according to claim 8, characterized in that, When the actual selection and the actual application of the target user are observed, the deviation range is quantified according to the difference or the difference percentage between the reference first selection proportion and the reference second selection proportion.

10. The non-heritage data natural language large model interaction method according to claim 8, characterized in that, In the process of generating the second prompt information according to the difference between the actual application of the selected answering content by the target user and the reference second selection proportion, if it is detected that the target user does not adjust according to the first prompt information, the reference second selection proportion is enlarged or reduced accordingly, and the adjusted proportion is taken as the basis for generating the second prompt information to guide the subsequent application of the target user.

Images