Data management method and system for agricultural planning

Abstract

The present application discloses a data management method and system for agricultural planning, the method comprising: accessing a web page including first content, the web page being associated with a first data set related to the first content, the first data set being inaccessible through a link on the web page or a web crawler, the first data set including agricultural planning data; obtaining a first question input by a user in a dialog box of the web page; determining data screening conditions; obtaining a second data set from the first data set based on user permissions and the data screening conditions; asking a question-answering model the second data set and the first question according to a first template; and returning the content output by the question-answering model to the dialog box.

Description

Technical Field

[0001] This application relates to artificial intelligence applications, and in particular to a data management method and system for agricultural planning. Background Art

[0002] Traditional resource planning websites typically only offer a few fixed pages for public access. However, professionals or internal staff who need to access the actual planning data behind a particular page must access other internal systems. Furthermore, different users have varying access rights to the data.

[0003] With the maturity of large-scale model solutions, they have demonstrated impressive material analysis capabilities. Many systems are attempting to integrate these models for application. However, most companies lack the ability to train these models, or even struggle to achieve good results with limited training resources. Consequently, model deployments often struggle to achieve effective integration with the system. Most deployments are limited to setting up a single conversation entry point.

[0004] If the model needs to be used for deep access, the model provider may need to provide a separate solution, which is expensive, and issues such as project collaboration will extend the project cycle.

[0005] At the same time, attempting to use large models to analyze and apply the system's own data also encounters various problems. For one thing, excessive data input can lead to confusing model outputs. For another, it can cause various data security issues. This is particularly problematic for businesses where some data is not suitable for public disclosure to all users. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a data management system for agricultural planning, which improves the traditional planning website to achieve rapid intelligence while ensuring data access security.

[0007] In one aspect, an embodiment of the present application provides a data management method for agricultural planning, comprising:

[0008] Accessing a webpage including first content, the webpage being associated with a first data set related to the first content, the first data set being inaccessible via a link on the webpage or a web crawler, the first data set including agricultural planning data;

[0009] Obtain the first question input by the user in the dialog box of the web page;

[0010] Determine data screening conditions;

[0011] A second data set is obtained from the first data set based on user permissions and data screening conditions;

[0012] Ask the question-answering model questions using the second data set and the first question according to the first template;

[0013] Return the content output by the question-answering model to the dialog box.

[0014] In some embodiments, the first template includes content instructing the question-answering model to use the second data set and the first content as the basis for answering, as well as content answering the first question.

[0015] In some embodiments, determining the data screening conditions specifically includes:

[0016] The first filtering condition is collected through the UI component of the user access page, and / or the second filtering condition is obtained through the first question, and the combination of the first filtering condition and / or the second filtering condition is used as the data filtering condition.

[0017] In some embodiments, based on the first question, a question is asked to the question-answering model according to a second template, so that the question-answering model outputs a specific type of information according to a method specified by the second template, and the information output by the question-answering model is used as a second screening condition;

[0018] The specific type of information includes a time range, a geographical range, and / or a preset data tag;

[0019] Each data file in the first data set is provided with one or more preset data labels.

[0020] In some embodiments, the second template is provided with a preset access address, the access address includes all preset data tags, and the second template is provided with a statement requiring the question-answering model to determine the preset data tags related to the first question, and requiring the question-answering model to output the relevant preset data tags in a preset format;

[0021] The preset data tag in the access address is mapped and updated based on the tag of the data file in the first data set.

[0022] In some embodiments, the preset data tag corresponding to each data file is provided with a correlation degree between the tag and the data file;

[0023] The method further comprises the following steps:

[0024] Determining the data volume of the second data set, if the data volume is greater than a preset value;

[0025] Without reducing the number of preset data tag coverages, the data volume of the second data set is reduced starting from the data file with the smallest sum of correlation degrees according to the sum of correlation degrees of the preset data in the data files.

[0026] In some embodiments, the following steps are also included: when the user asks a second question based on the first question, the third data set filtered out from the first data set based on the second question is filtered out before being sent to the question-answering model, and the data belonging to the second data set in the third data set is eliminated.

[0027] In some embodiments, when the question-answering model is called for the first time, the third template is used to require the question-answering model to output results according to preset rules, where the preset rules include output length.

[0028] On the other hand, an embodiment of the present application discloses a data management system for agricultural planning, comprising:

[0029] an access module, configured to access a webpage including first content, wherein the webpage is associated with a first data set related to the first content, the first data set being inaccessible via a link on the webpage or a web crawler, the first data set including agricultural planning data;

[0030] An acquisition module, configured to acquire a first question input by a user in a dialog box of a web page;

[0031] Determination module, used to determine data screening conditions;

[0032] a screening module, configured to obtain a second data set from the first data set based on user permissions and data screening conditions;

[0033] a questioning module, configured to ask the question-answering model questions using the second data set and the first question according to the first template;

[0034] The return module is used to return the content output by the question-answering model to the dialog box.

[0035] On the other hand, an embodiment of the present application discloses a data management system for agricultural planning, comprising:

[0036] Memory, used to store programs;

[0037] A processor is used to load the program to execute the data management method for agricultural planning.

[0038] By setting an association between a web page and a first data set through an embodiment of the present application, a user is allowed to access agricultural planning data that is within their own authority and of interest through a question-and-answer model, and the question-and-answer model outputs answers to relevant questions; the user can use the window of the web page to ask questions, and the system indirectly asks questions through the question model based on the user's question content and the data filtered based on the user's authority and filtering conditions, and returns the questions answered by the model to the user; in this way, the deployer does not need to train the model too much, nor does it need to provide data to the model in advance, thereby achieving rapid deployment and application of the model, and providing data to the model based on the scope required by the user. On the one hand, the data can be updated in a timely manner, and the output results are less erroneous than letting the model learn the data in advance. On the other hand, filtering the data through methods such as authority control can speed up the model response and reduce the risk of data leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0040] Figure 1 is a flow chart of the method provided in the embodiment of the present application;

[0041] Figure 2 This is a schematic diagram of the data flow provided in the embodiment of the present application;

[0042] Figure 3 This is a system block diagram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of this application more clear, the following will refer to the drawings in the embodiments of this application to clearly and completely describe the technical solutions of this application through implementation methods. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0044] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0046] Question-answering models primarily refer to currently mainstream natural language models such as Wenxin Yiyan, GPT, and DeepSeek. These models can be remotely accessed through methods such as purchasing Tonken, and some allow users to deploy them locally. Local deployment, in particular, provides greater security for user data and privacy.

[0047] These models are typically pre-trained to handle most common-sense problems and can be learned instantly through web searches. However, these models currently generally adhere to web crawler protocols. When crawling web content, crawlers ignore these pages to avoid copyright infringement. Similarly, if a web page is private or has permissions set, these models cannot retrieve it through their search capabilities.

[0048] In this case, if the model needs to be quickly deployed within an existing website system, content data or content must be used to train the model, which involves training software, hardware, and personnel costs. Most companies do not have these capabilities. Simple deployments typically only leave a Q&A model interface for users to answer questions, preventing users from accessing private data through documentation.

[0049] Reference Figure 1 The present application provides a data management method for agricultural planning. In the embodiment, a dialog window component is added to an existing website as a front-end access, and a three-party question-and-answer model is called in the background based on the embodiment to answer user questions. The method includes:

[0050] S1. Accessing a webpage containing first content. The webpage is associated with a first dataset related to the first content. The first dataset is not accessible via a link on the webpage or a web crawler, and the first dataset includes agricultural planning data. In this embodiment, the webpage is a website related to agricultural planning, and the first content is primarily publicly accessible information, readily accessible to most users. This content is public, but may lack some data or information that is inconvenient or unnecessary to disclose.

[0051] The backend of this webpage is associated with a first dataset, which includes one or more data tables. For example, the webpage content introduces the planning of a certain place, while the first dataset records the monthly yield information of various crops in the area in detail. This information is generally not disclosed to the public, and the website content may only provide a general description of the situation. However, professional users such as scholars or staff may want to know more detailed information when browsing the webpage. The traditional method only requires users to log in to professional databases to query, which is not convenient. Since this data cannot be crawled by crawlers, it is also difficult to interpret the data using third-party question-and-answer models.

[0052] S2. Obtaining a first question input by the user in a dialog box of the webpage.

[0053] By setting up a dialog component on the web page, the chat window will transmit the user's input, namely the "first question", to the backend, which will call the model based on this information (the first question) for processing. It is worth noting that the first question is not directly sent to the third-party question-and-answer model. Instead, it is first transmitted to the backend for processing, that is, embedded in the question template, and then the backend calls the question-and-answer model.

[0054] S3. Determine data screening conditions.

[0055] In this step, the amount of data associated with some web pages may be quite large, spanning many years. It may also involve data from a variety of different aspects. Therefore, it is necessary to filter the data to reduce the amount. This allows the question-answering model to focus on the content of the user's question while also preventing excessive data from being fed into the model, which could lead to incorrect answers due to limited memory capacity.

[0056] In this step, there are multiple ways to determine data screening conditions.

[0057] Among them, some data can be determined through the UI components of the web page itself, such as maps, time selection items, region selection items, etc. At the same time, the model's own capabilities can also be used to determine what information the user wants to know from the user's first question.

[0058] In some embodiments, determining the data screening conditions specifically includes:

[0059] The first filtering condition is collected through the UI component of the user access page, and / or the second filtering condition is obtained through the first question, and the combination of the first filtering condition and / or the second filtering condition is used as the data filtering condition.

[0060] As can be seen, there are multiple ways to determine data filtering conditions, and they can be combined. Generally speaking, if the UI component collects information and the information obtained through the first question overlaps, the conditions of the first question can be prioritized or combined. For example, if a user selects a map of location A through the UI component and then asks information about location B, a strategy can be adopted to prioritize location B as the filtering condition. Alternatively, the collected information can be arranged in a parallel relationship, that is, when filtering, selecting either A or B. Time and location are particularly useful filtering information for agricultural planning content. A similar approach can be adopted for time. For example, the user can select a certain time period on the page before asking a question. It is understood that the UI component can be used to select an area on the map. The UI component can pre-set the map area into partitions and set the selected areas to correspond to place names in the data file. Therefore, after the user selects an area and then asks a question, it is equivalent to directly obtaining information about an area that the user may not be able to describe verbally.

[0061] Based on the first question, questions are asked to the question-answering model according to the second template, so that the question-answering model outputs a specific type of information in the manner specified by the second template, and the information output by the question-answering model is used as the second screening condition. Specifically, questions can be set in the second template to require the question-answering model to obtain this information and output it in a specific format. The background will filter out these formatted data based on the response of the question-answering model and use them as conditions for data screening. For example, questions are set in the template to let the template output at least one place name mentioned by the user in the form of XX city or XX district. If it is not mentioned, it will not be output. If there are multiple places, they will be separated by commas. When outputting, set the "Place Name:" identifier before the output result. In this way, the model can output the information in the first question obtained through semantic understanding in a specific format.

[0062] Regarding time, you can set a statement to require the Q&A model to output the time range the user wants to know about. You can also convert unified requirements into data ranges. For example, for today, you can set it to "A month B day 2025 - A month B day 2025." If it's 2024, then output will be "January 1, 2024 - December 31, 2024." By setting output requirements in the template, you can format the corresponding data.

[0063] It is understandable that the above-mentioned multiple screening conditions can be set individually or in combination, and the method of obtaining them can be obtained through the UI component of the website itself or through the question-answer model. It can be seen that the embodiment includes multiple methods, and the methods are not mutually exclusive.

[0064] Specific types of information include time ranges, geographical ranges, and / or pre-set data tags. It is understood that the time range and geographical range can be used to filter out a rough range of data. Users may be interested in certain types of data, which may exist in the data table or may not exist in the data table and require calculations using the data table.

[0065] Filtering by time range requires setting the following conditions. If the time period output by the model doesn't fall within the range of recorded data, set a rule to provide data from a preset period close to the relevant time period. For example, if a user asks to predict next year's crop yield, next year clearly doesn't fall within the time period of the existing data, but rather in the future. To analyze data for a future time period, only historical data can be used, so data from the previous year can be provided. This means that when filtering data using time conditions, rules must be set. When the user's time period is in the future, data is filtered according to the preset time range. For past data that doesn't exist, the model will inform the user that the data doesn't exist or that analysis is impossible. In the template, you can instruct the model to output specific information if the user wishes to analyze future data, allowing the backend to filter data based on the output results.

[0066] Then, each data file in the first dataset can be assigned one or more preset data tags. For example, if a data table records the yield information of various crops in a certain region, then the field information in the table itself is the tag of this data table. At the same time, some values ​​that can be indirectly calculated from the data table can be set, such as planting area, planting efficiency, etc. By setting common data types, you can associate them with the data table that can calculate this information.

[0067] In order to facilitate the management of these label information and avoid repeatedly modifying the template, a preset access address can be set in the second template, and the access address includes all preset data labels. The second template is provided with a statement requiring the question-answering model to determine the preset data labels related to the first question, and requires the question-answering model to output the relevant preset data labels in a preset format.

[0068] The preset data tags in the access address are mapped and updated based on the tags of the data files in the first dataset. It is understood that by updating in the above manner, these tags can be updated without frequently rewriting scripts. When the data table is deleted, the related content can also be quickly deleted. In this embodiment, the question instruction question-answering model can be set to output the content that the user is interested in in the first question and that belongs to these preset data tags in a specific format.

[0069] S4. Obtain a second data set from the first data set based on user permissions and data screening conditions.

[0070] Now, you can use these time and location tags to filter data ranges. Specifically, you can filter data based on the time range. Based on user data needs, you'll generally expand the data range of interest. Based on region, you can filter data by region. By filtering out relevant tags, you can roughly determine which data tables and data are involved. This will form a second data set.

[0071] Of course, if the permission requirements for a data table are higher than the user's current permission, this data table will not be filtered out even if it is related to the content the user is interested in. This ensures that users cannot bypass their permissions and obtain data they should not know through the question-and-answer model.

[0072] S5. Ask the question-answering model the second data set and the first question according to the first template.

[0073] The first template includes content instructing the question-answering model to use the second dataset and the first content as the basis for its answer, as well as content for its response to the first question. It can be understood that this first template feeds the filtered dataset into the model and instructs the model to answer the first question based on the second dataset. For example, "Please use the data table to answer the average monthly corn yield in region XX this year and predict next month's data." The model will output "Location: XX District, XX City, XX Province" and "Time: January 1, 2025 to December 31, 2025, April 1, 2025 to April 30, 2025" as required. When the time passes through the condition library, the judgment condition is set to determine if a time period is completely in the future. Therefore, data from the most recent year will be used as the filtering condition. If the filtering time period falls within the past, no data will be filtered. If no data is filtered, a default filtering period can be set, such as the most recent year, or data from a year that exists and is close to the target time period can be filtered.

[0074] S6. Return the content output by the question-answering model to the dialog box.

[0075] It is understood that the data returned based on the question-answer model is returned to the user's chat box component. Of course, the data can also be processed based on the answer mode of the question-answer model before being returned.

[0076] In some embodiments, the preset data tag corresponding to each data file is provided with a correlation degree between the tag and the data file.

[0077] The method further comprises the following steps:

[0078] Determining the data volume of the second data set, if the data volume is greater than a preset value;

[0079] Without reducing the number of preset data tag coverages, the data volume of the second data set is reduced starting from the data file with the smallest sum of correlation degrees according to the sum of correlation degrees of the preset data in the data files.

[0080] You can set a correlation for tags. For example, if some tags are highly correlated with data files, you can set a higher correlation. For example, if a field in a data file has a high correlation with a device, the correlation can be lower as a result of indirect calculation. In this case, if the amount of matching data is large, you can prioritize excluding data files with a low sum of correlations. A low sum of correlations indicates that there are few associated tags or that the associated tags are all indirectly calculated, which reduces the AI's response speed. While this simple rule won't achieve optimal optimization, it can improve system response speed and reduce the model's tonken cost by reducing the amount of input data. For example, File A contains monthly production data, and File B records daily production data. If the user wants to know the monthly production data, File A is directly correlated, while File B requires calculation. Assume that the correlation for directly associated tags is set to 1, and the indirect correlation is set to 0.5. In this example, if the data volume is large, File B's data can be excluded first. Based on this example, assuming that the relevance of another tag that hits file B is 1, the sum of the tag relevances of file B is greater than that of file A, so A can be excluded.

[0081] Regarding the issue of label coverage, assume that the screening conditions include four preset labels T1, T2, T3, and T4. Among them, file A matches T1 and T2, file B matches T3 and T4, and file C matches files T1, T2, and T3. It can be understood that in this example, assuming that the amount of data has exceeded the threshold, at this time, in order to ensure that the number of hit label coverage does not decrease. File A and file C can be removed, depending on the degree of correlation between the two. Assuming that the sum of the correlations of file A is greater than that of file C, the data of file C is discarded. When the correlation between the two is the same, the actual amount of data of the two can be considered, and the data file with a large amount of data can be discarded first. Therefore, the number of hit label coverage does not decrease means that after removing some data files, all labels that could be hit before removing the data can still be hit.

[0082] Some embodiments further include the following step: when a user asks a second question based on the first question, before inputting a third dataset filtered from the first dataset based on the second question into the question-answering model, data belonging to the second dataset is removed from the third dataset. In this solution, data may be fed into the model multiple times during multiple question-answering sessions. To reduce tonken consumption, data already fed into the question-answering model may be removed.

[0083] To further reduce the number of tonkens, the system uses a third template to set the Q&A model's answer mode to reduce the number of output tonkens. When the Q&A model is first called, the third template instructs the model to output results according to pre-set rules, including output length. Of course, you can also set a statement to require the model to output a concise answer.

[0084] Reference Figure 2 , you can understand how this solution is implemented. In this solution, users can ask questions about the content of a web page through the chat window in the client (such as a browser). After the user asks a question, the background will use the question-answering model to extract specific information based on the user's question, and then continue to use this information as a screening condition to obtain data related to the web page from the database. Combined with this data and the user's question, the third-party model will be called again to complete the answer to the user's question and reply to the user's chat window.

[0085] Referring to Figure 3, an embodiment of the present application discloses a data management system for agricultural planning, comprising:

[0086] an access module, configured to access a webpage including first content, wherein the webpage is associated with a first data set related to the first content, the first data set being inaccessible via a link on the webpage or a web crawler, the first data set including agricultural planning data;

[0087] An acquisition module, configured to acquire a first question input by a user in a dialog box of a web page;

[0088] Determination module, used to determine data screening conditions;

[0089] a screening module, configured to obtain a second data set from the first data set based on user permissions and data screening conditions;

[0090] a questioning module, configured to ask the question-answering model questions using the second data set and the first question according to the first template;

[0091] The return module is used to return the content output by the question-answering model to the dialog box.

[0092] The present application discloses a data management system for agricultural planning, including:

[0093] Memory, used to store programs;

[0094] A processor is used to load the program to execute the data management method for agricultural planning.

[0095] Note that the above are only preferred embodiments of the present application and the technical principles employed. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments and may include many other equivalent embodiments without departing from the scope of the present application. The scope of the present application is determined by the scope of the appended claims.

Claims

1. A data management method for agricultural planning, characterized in that: include: Accessing a webpage including first content, the webpage being associated with a first data set related to the first content, the first data set being inaccessible via a link on the webpage or a web crawler, the first data set including agricultural planning data; Obtain the first question input by the user in the dialog box of the web page; Determine data screening conditions; A second data set is obtained from the first data set based on user permissions and data screening conditions; Ask the question-answering model questions using the second data set and the first question according to the first template; Return the content output by the question-answering model to the dialog box; The data screening conditions are determined, specifically including: Collecting a first filtering condition through a UI component of a user-accessed page and / or a second filtering condition obtained through a first question, and using a combination of the first filtering condition and / or the second filtering condition as a data filtering condition; Based on the first question, ask the question-answering model a question according to the second template, so that the question-answering model outputs a specific type of information in a manner specified by the second template, and uses the information output by the question-answering model as a second screening condition; The specific type of information includes a time range, a geographical range, and / or a preset data tag; Each data file in the first data set is provided with one or more preset data labels; The second template is provided with a preset access address, which includes all preset data tags. The second template is provided with a statement requiring the question-answering model to determine the preset data tags related to the first question, and requiring the question-answering model to output the relevant preset data tags in a preset format; The preset data tag in the access address is mapped and updated based on the tag of the data file in the first data set.

2. The data management method for agricultural planning according to claim 1, characterized in that: The first template includes content that instructs the question-answering model to use the second data set and the first content as the basis for answering, as well as content that answers the first question.

3. The data management method for agricultural planning according to claim 1, characterized in that: The preset data tag corresponding to each data file is set with the correlation degree between the tag and the data file; The method further comprises the following steps: Determining the data volume of the second data set, if the data volume is greater than a preset value; Without reducing the number of preset data tag coverages, the data volume of the second data set is reduced starting from the data file with the smallest sum of correlation degrees according to the sum of correlation degrees of the preset data in the data files.

4. The data management method for agricultural planning according to claim 1, characterized in that: The following steps are also included: When the user asks a second question based on the first question, the third data set filtered from the first data set based on the second question is filtered out before being sent to the question-answering model, and the data belonging to the second data set in the third data set is eliminated.

5. The data management method for agricultural planning according to claim 1, characterized in that: When the question-answering model is called for the first time, the third template is used to require the question-answering model to output results according to preset rules, where the preset rules include output length.

6. A data management system for agricultural planning, characterized in that: include: Memory, used to store programs; A processor, configured to load the program to execute the data management method for agricultural planning according to any one of claims 1 to 5.

Images