Natural resource planning intelligent question and answer analysis method and system

Abstract

The invention relates to an intelligent question and answer analysis method and system for natural resource planning, and relates to the technical field of intelligent question and answer. Converting the document data and constructing a planning vector library; after user question information is received, vectorizing the user question information into question vector information; searching in a planning vector library by using the question vector information to generate related text blocks; reordering the related text blocks to obtain selected background knowledge; and inputting the selected background knowledge and the user question information into a preset language large model, and generating and outputting reply result information. The method has the effect of improving the answering accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent question answering technology, and in particular to an intelligent question answering analysis method and system for natural resource planning. Background Technology

[0002] Intelligent question answering is an important branch of artificial intelligence. It aims to enable computers to understand questions posed by humans in natural language and return accurate, concise answers that conform to human language habits through analysis, retrieval, and reasoning.

[0003] Current intelligent question answering systems typically first store relevant knowledge to form a knowledge base, then use natural language processing to analyze the semantics of user questions, and retrieve relevant information from the knowledge base or text to provide an answer based on the semantics of the question.

[0004] Because the field of natural resources is highly specialized and ambiguous, existing intelligent question-answering systems are prone to knowledge biases when asking questions about natural resource planning, resulting in low accuracy of answers. Summary of the Invention

[0005] To improve the accuracy of answers, this invention provides an intelligent question-answering analysis method and system for natural resource planning.

[0006] Firstly, the present invention provides an intelligent question-answering analysis method for natural resource planning, which adopts the following technical solution:

[0007] A natural resource planning intelligent question-answering analysis method includes:

[0008] S1: Upload documents related to natural resource planning;

[0009] S2: Convert the document data and construct a planning vector library;

[0010] S3: Upon receiving a user's question, the user's question is vectorized into question vector information;

[0011] S4: Use the question vector information to retrieve relevant text blocks from the planning vector library;

[0012] S5: Reorder the relevant text blocks to obtain selected background knowledge;

[0013] S6: Input the selected background knowledge and the user's question information into a preset language model, generate and output the response result information.

[0014] Optionally, after uploading natural resource planning documents, the following may also be included:

[0015] S11: Retrieve the document name from the document data;

[0016] S12: Query and retrieve the document data from the historical time period and use it as historical data;

[0017] S13: Retrieve the historical name from the aforementioned historical data;

[0018] S14: Compare the historical name with the document name to obtain the name overlap.

[0019] S15: Determine whether the name overlap is lower than the preset complete overlap.

[0020] S16: If yes, continue uploading the document materials;

[0021] S17: If not, determine and output the duplicate upload prompt information based on the historical name and the document name.

[0022] Optionally, the method for generating the planning vector library includes:

[0023] S21: Retrieve the document format and topic type from the document data;

[0024] S22: Convert the document data according to the document format to form a document with a uniform format;

[0025] S23: Extract text from the uniform format document to obtain the extracted text;

[0026] S24: Cut the extracted text into blocks to form text paragraphs;

[0027] S25: Combine the text paragraph with the topic type to generate a topic paragraph;

[0028] S26: Select and transform vector models for the aforementioned thematic paragraphs to form a thematic vector library;

[0029] S27: Merge all the aforementioned thematic vector libraries into the planning vector library.

[0030] Optionally, the method for generating the thematic paragraphs includes:

[0031] S251: Determine the topic area and topic scenario according to the topic type;

[0032] S252: Determine the domain keywords based on the aforementioned topic domain;

[0033] S253: Compare the text paragraph with the domain keywords to determine the domain relevance;

[0034] S254: Select the text paragraphs according to the domain relevance to form domain paragraphs;

[0035] S255: Determine the scenario keywords based on the aforementioned thematic scenario;

[0036] S256: Compare the text paragraph with the scene keywords to determine the scene relevance;

[0037] S257: Select text paragraphs according to the scene relevance to form scene paragraphs;

[0038] S258: Merge the domain paragraph and the scene paragraph to form the topic paragraph.

[0039] Optionally, after merging the domain paragraph and the scene paragraph to form the topic paragraph, the method further includes:

[0040] S2581: When the name overlap exceeds the preset version overlap, the corresponding historical data is used as historical version data, and the corresponding unified format document is used as the current format document.

[0041] S2582: Retrieve historical format documents from the historical version data;

[0042] S2583: Compare the historical format document with the current format document to determine the distinguishing paragraphs and the identical paragraphs;

[0043] S2584: Combine the identical paragraphs and the different paragraphs to generate a version paragraph, and add the topic paragraph.

[0044] Optionally, the method for generating the relevant text blocks includes:

[0045] S41: Use the question vector information to retrieve the initial text block from the planning vector library;

[0046] S42: Retrieve the initial quantity and text relevance from the initial text block;

[0047] S43: Determine the selection quantity according to the initial quantity;

[0048] S44: Combine the question vector information with the initial text block to generate a question difficulty value;

[0049] S45: Determine the difficulty adjustment coefficient based on the question difficulty value;

[0050] S46: Determine the adjustment quantity using the selected quantity and the difficulty adjustment coefficient;

[0051] S47: Select an initial text block based on the text relevance and the adjustment quantity, and use it as the relevant text block.

[0052] Optionally, the method for generating the question difficulty value includes:

[0053] S441: Retrieve the amount of question data from the question vector information;

[0054] S442: Determine the reference value for data difficulty based on the amount of data asked;

[0055] S443: Retrieve the number of vectors from the initial text block;

[0056] S444: Determine the reference value for the difficulty of the vector library based on the number of vector libraries mentioned;

[0057] S445: Calculate the sum of the data difficulty reference value and the vector library difficulty reference value and use it as the comprehensive difficulty reference value, and use the comprehensive difficulty reference value as the question difficulty value.

[0058] Optionally, after using the comprehensive difficulty reference value as the question difficulty value, the method further includes:

[0059] S4451: Collect user account information and question input parameters;

[0060] S4452: Determine the user type and input baseline parameters from the user account information;

[0061] S4453: Compare the input reference parameters with the problem input parameters to determine the input deviation parameters;

[0062] S4454: Determine the parameter deviation range and unit adjustment value based on the user type;

[0063] S4455: Determine the user similarity value according to the input deviation parameter and the parameter deviation range;

[0064] S4456: Determine the input adjustment reference value by combining the user similarity value and the unit adjustment value;

[0065] S4457: Update and replace the question difficulty value with the input adjustment reference value.

[0066] Optionally, the method for generating the selected background knowledge includes:

[0067] S51: Retrieve the relevant quantity from the relevant text block;

[0068] S52: Determine the relevant ratio value using the relevant quantity and the initial quantity;

[0069] S53: Determine the re-ranking ratio by combining the relevant ratio value and the user similarity value;

[0070] S54: Determine the number of reorderings using the reordering ratio value and the relevant quantity;

[0071] S55: Combine the question vector information with the relevant text block to re-search in the planning vector library to obtain the reordered text block;

[0072] S56: Retrieve the reordering relevance from the reordered text block;

[0073] S57: Select the reordered text block according to the reordering relevance and the reordering quantity, and use it as the selection background knowledge.

[0074] Secondly, this invention provides an intelligent question-and-answer analysis system for natural resource planning, employing the following technical solution:

[0075] A natural resource planning intelligent question-answering analysis system includes:

[0076] The data collection module is used to collect document information, user account information, and question input parameters.

[0077] The memory stores a program for implementing a natural resource planning intelligent question-answering analysis method as described in any one of the first aspects;

[0078] The processor loads and executes programs stored in memory.

[0079] In summary, the present invention has at least one of the following beneficial technical effects:

[0080] 1. The uploaded documents are converted and constructed into a planning vector library. When a user's question is received, it is vectorized into a question vector and retrieved from the planning vector library to generate relevant text blocks. Then, the relevant text blocks are reordered to obtain selected background knowledge. The selected background knowledge and the user's question are input into a preset language model to generate and output the answer result information. In this way, the planning-related issues of natural resources are first queried and matched with the professional planning vector library, and then the answer result is generated by the language model, thereby improving the accuracy of the answer.

[0081] 2. By retrieving the document name from the document data, querying and retrieving historical data and historical names, the name overlap is obtained by comparison. By judging whether the name overlap is lower than the preset complete overlap, it is determined whether to continue uploading, thereby avoiding duplicate uploads.

[0082] 3. By retrieving document formats and topic types from document data, analyzing them separately to form text paragraphs and topic types, and then combining them to generate topic paragraphs, vector models are selected from topic paragraphs and transformed to form a topic vector library. Finally, all topic vector libraries are merged to form a planning vector library, thereby improving the accuracy of the obtained planning vector library. Attached Figure Description

[0083] Figure 1 This is a flowchart of the intelligent question-answering analysis method for natural resource planning;

[0084] Figure 2 This is a flowchart illustrating the process after uploading documents related to natural resource planning.

[0085] Figure 3 This is a flowchart illustrating the method for generating a planning vector library;

[0086] Figure 4 This is a flowchart illustrating the method for generating the relevant text blocks. Detailed Implementation

[0087] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0088] A smart question-answering analysis method for natural resource planning is proposed. By uploading natural resource planning documents and constructing a planning vector library, the method performs vectorized retrieval of user questions, reorders relevant text blocks, and generates answers by combining them with a large language model. It also incorporates optimizations such as document deduplication, version management, user type adaptation, and question difficulty adjustment, thereby achieving accurate, efficient, and user-relevant smart question-answering analysis in the field of natural resource planning, thus improving the accuracy of answers.

[0089] Reference Figure 1 This invention discloses an intelligent question-answering analysis method for natural resource planning, which includes:

[0090] S1: Upload documents related to natural resource planning.

[0091] Documents and materials refer to various documents and materials related to the field of natural resource planning. These documents and materials cover textual materials in professional fields such as territorial spatial planning, ecological protection red line delineation, mineral resource development and utilization planning, and urban and rural construction land planning. The content of these documents and materials typically includes professional information such as planning principles, technical standards, policy provisions, and case studies.

[0092] Documents are generally uploaded in advance by the operator, but they can also be uploaded by the user during the Q&A session. The uploaded documents can be in various formats such as PDF, PPT, and DOC.

[0093] To further ensure the rationality of the uploaded natural resource planning documents, it is necessary to conduct further separate analysis and calculations on the uploaded documents. The specific steps are explained in detail below.

[0094] Reference Figure 2After uploading the natural resources planning documents, the following steps are involved:

[0095] S11: Retrieve the document name from the document data.

[0096] The document name refers to the external name information of the file used to identify the subject matter of the document. Document information includes the document name.

[0097] By retrieving the document name from the documentation, it becomes easier to use later.

[0098] S12: Query and retrieve historical documents and use them as historical data.

[0099] Historical data refers to documents that were uploaded before this upload.

[0100] By querying and defining historical data, it can be easily used in the future.

[0101] S13: Retrieve historical names from historical records.

[0102] Among them, the historical name refers to the external name information of the file used to identify the subject matter of historical data.

[0103] By retrieving historical names from historical records, it becomes easier to use them later.

[0104] S14: Compare historical names with document names to obtain the name overlap.

[0105] Name overlap refers to the degree of similarity between historical names and document names. Name overlap is expressed as a percentage or a numerical value between 0 and 1. Name overlap measures the degree of matching between two names; a higher value indicates that the names are more similar.

[0106] After preprocessing the historical names and document names by removing special characters, unifying capitalization, and segmenting words, the similarity between the historical names and document names is calculated using algorithms such as edit distance and cosine similarity. The calculated value is then used as the name overlap score for convenient subsequent use.

[0107] Algorithms such as edit distance and cosine similarity are existing technologies and will not be elaborated upon here.

[0108] S15: Determine if the name overlap is lower than the preset perfect overlap. If yes, proceed to S16; if no, proceed to S17.

[0109] The "complete overlap" threshold is a standard used to determine whether document names are completely or highly similar. The complete overlap threshold can be set to a value between 90% and 100% or between 0.9 and 1. This threshold is preset by the operator according to actual needs.

[0110] By judging whether the name overlap is lower than the preset perfect overlap, it can be determined whether the currently uploaded document has already been uploaded.

[0111] S16: Continue uploading documents and materials.

[0112] If the name overlap is lower than the preset perfect overlap, it means that the document being uploaded has not been uploaded before, so the document upload continues.

[0113] S17: Determine and output duplicate upload prompt information based on historical names and document names.

[0114] Among them, the duplicate upload prompt message refers to the prompt message used to inform users that there is a duplicate upload.

[0115] When the name overlap is not lower than the preset complete overlap, it means that the document being uploaded has already been uploaded. Therefore, by selecting the historical name and document name corresponding to the name overlap being not lower than the preset complete overlap and using them as the overlapping name, and by combining the overlapping name with the preset upload duplicate prompt words to form upload duplicate prompt information and output it, it is convenient for users to understand the upload duplicate situation in a timely manner.

[0116] The duplicate upload warning text refers to the preset text used to indicate when an upload is duplicated. The duplicate upload warning text is preset by the operator according to actual needs.

[0117] S2: Convert the document data and build a planning vector library.

[0118] The planning vector database refers to a dedicated vector database built for documents and materials in the field of natural resource planning. It stores high-dimensional vector data of these documents after vectorization. These vectors correspond to text paragraphs within the documents, reflecting the semantic features of the text and supporting quick location of content related to user queries through vector similarity retrieval.

[0119] By converting the documents and building a vector library, a planned vector library is formed for easy use later.

[0120] To further ensure the rationality of the planning vector library, it is necessary to perform further separate analysis and calculation on the planning vector library, which will be explained in detail through the following steps.

[0121] Reference Figure 3 The method for generating the planning vector library includes the following steps:

[0122] S21: Retrieve document format and topic type from document data.

[0123] Document format refers to the file type of the document, such as PDF, PPT, DOC, TXT, etc. Thematic type refers to the specific area of ​​natural resource planning to which the document belongs. Thematic types include territorial spatial planning, ecological protection red line planning, mineral resource development planning, etc. Thematic type reflects the professional scope and content theme of the document.

[0124] The document format is obtained by reading the file extensions from the document data, and the core terms in the document data are read and used as topic types for easy subsequent use.

[0125] S22: Convert document data according to document format to form a document with a uniform format.

[0126] Among them, a unified format document refers to a standardized format document formed by converting documents of different original formats.

[0127] By recognizing document formats and calling the corresponding conversion tools accordingly, for uneditable PDFs or scanned documents, OCR technology is used to recognize the text content and convert it into an editable PDF; for PPT, DOC and other formats, the content is completely exported and converted into an editable PDF format through a format conversion engine; during the conversion process, the core text information of the document is retained and redundant format marks are removed, ultimately forming an editable document with a unified structure, laying the foundation for subsequent text extraction and segmentation.

[0128] S23: Extract text from a uniformly formatted document to obtain the extracted text.

[0129] Extracting text refers to the plain text content that is extracted from a document with a uniform format.

[0130] By filtering out non-text elements such as images and table frames, as well as redundant formatting information, from a uniformly formatted document, and correcting any garbled characters or grammatical errors that may occur during the extraction process using text cleaning technology, the final result is clear and coherent plain text content, which is then used as the extracted text to prepare for subsequent text segmentation and thematic paragraph generation.

[0131] Text extraction and cleaning are existing technologies, so they will not be elaborated on here.

[0132] S24: Extract the text and cut it into blocks to form text paragraphs.

[0133] Here, a text segment refers to several independent text fragments formed after the extracted text is divided according to a preset segment length. The segment length refers to the length of the extracted text segment, which is preset by the operator according to actual needs.

[0134] By using a preset segmentation length to segment the extracted text into segments, the inefficiency of long text processing is avoided, while ensuring the relative integrity of the semantics, which facilitates subsequent vectorization processing.

[0135] S25: Combine text paragraphs with topic types to generate topic paragraphs.

[0136] Among them, a thematic paragraph refers to a collection of text fragments that focus on a specific thematic area.

[0137] By analyzing text paragraphs and topic types, topic paragraphs are generated. The content of the topic paragraphs is highly matched with the professional scope of the corresponding topic, which facilitates the subsequent construction of the topic vector library.

[0138] To further ensure the rationality of the thematic paragraphs, it is necessary to conduct further separate analysis and calculations on the thematic paragraphs, which will be explained in detail through the steps shown below.

[0139] The method for generating thematic paragraphs includes the following steps:

[0140] S251: Determine the subject area and subject scenario according to the subject type.

[0141] In this context, "topic domain" refers to the professional field to which the topic type belongs. "Topic scenario" refers to the specific application scenario or business process under the topic type. Different topic types correspond to different topic domains and topic scenarios.

[0142] By inputting the topic type into a preset topic type database, the topic field and topic scenario can be matched and obtained, which facilitates subsequent use.

[0143] The topic type database pre-stores a mapping table of different topic types and their corresponding topic fields and topic scenarios. The topic type database is pre-configured by the operator.

[0144] For example, if the topic type is the overall land and space planning, the topic field can be land and space planning, and the topic scenario can be specific scenarios such as planning process, indicator calculation, and boundary delineation under the land and space planning field.

[0145] S252: Determine domain keywords based on the specific topic area.

[0146] Domain keywords refer to specialized terms, core concepts, or high-frequency words that are highly relevant to a specific topic area. Domain keywords accurately identify the professional scope and content characteristics of that field, serving as the core basis for selecting domain-related content from text paragraphs.

[0147] By extracting core terms from the standard terminology database of the specific subject area and supplementing them by analyzing high-frequency professional words in the historical documents of the specific subject area, the subject area keywords are obtained.

[0148] For example, when the topic area is territorial spatial planning, by calling the "Terminology of Territorial Spatial Planning", terms such as the pattern of territorial spatial development and protection and the three control lines are extracted. Then, the historical documents corresponding to the topic area are retrieved and high-frequency professional terms are identified to obtain terms such as urban development boundary and flexible space. Finally, the obtained terms are combined to obtain the domain keywords.

[0149] S253: Compare text paragraphs with domain keywords to determine domain relevance.

[0150] Domain relevance refers to the degree of matching between the content of a text paragraph and domain keywords. Domain relevance is a value between 0% and 100% or between 0 and 1.

[0151] By comparing text paragraphs with domain keywords, extracting overlapping words, calculating the number of each overlapping word, and then weighting the overlapping words according to their respective weights in the domain, the domain relevance is obtained for subsequent use.

[0152] In the domain keyword field, the weight corresponding to core terms is greater than the weight corresponding to high-frequency professional terms. The weight values ​​corresponding to core terms and high-frequency professional terms are preset by the operator.

[0153] S254: Select text paragraphs based on domain relevance to form domain paragraphs.

[0154] Among them, domain paragraphs refer to a set of text fragments selected from text paragraphs that are related to a specific topic domain.

[0155] By selecting text paragraphs whose domain relevance is greater than the preset domain benchmark relevance, and then using these selected text paragraphs as domain paragraphs, it becomes easier to use them later.

[0156] Domain benchmark relevance refers to the pre-set threshold used when selecting domain segments. Domain benchmark relevance is preset by the operator based on actual needs.

[0157] S255: Determine scene keywords based on the specific scenario.

[0158] Scenario keywords refer to professional terms, core concepts, or high-frequency words that are highly relevant to a specific topic or scenario. Scenario keywords accurately identify the professional scope and content characteristics of the scenario and are the core basis for selecting scenario-related content in text paragraphs.

[0159] The core terms are extracted by calling the standard terminology library of the specific scenario, and supplemented by analyzing the high-frequency professional words in the historical documents of the specific scenario, so as to obtain the scenario keywords.

[0160] For example, when the topic scenario is the planning process, core process terms such as current situation survey, target setting, and scheme demonstration are extracted from the business standards and operation manual "Regulations for the Compilation of Territorial Spatial Planning" corresponding to the scenario. Then, by analyzing the high-frequency business terms in the historical documents of the scenario, terms such as public participation, expert review, and result approval are obtained. Finally, the obtained terms are combined to obtain scenario keywords.

[0161] S256: Compare text paragraphs with scene keywords to determine scene relevance.

[0162] Contextual relevance refers to the degree of matching between the content of a text paragraph and the keywords of the context. Contextual relevance is a value between 0% and 100% or between 0 and 1.

[0163] By comparing text paragraphs with scene keywords, extracting overlapping words, calculating the number of each overlapping word, and then weighting the overlapping words according to their weight in the scene to obtain the scene relevance, it is convenient for subsequent use.

[0164] In the scenario keywords, the weight corresponding to core terms is greater than the weight corresponding to high-frequency professional terms. The weight values ​​corresponding to core terms and high-frequency professional terms are preset by the operator.

[0165] S257: Select text paragraphs based on scene relevance to form scene paragraphs.

[0166] Among them, scene paragraphs refer to a collection of text fragments selected from text paragraphs that are related to a specific topic scene.

[0167] By selecting text paragraphs whose scene relevance is greater than the preset scene baseline relevance, and using these selected text paragraphs as scene paragraphs, it is convenient for subsequent use.

[0168] Scene baseline relevance refers to the threshold value preset when selecting scene segments. Scene baseline relevance is preset by the operator according to actual needs.

[0169] S258: Merge domain paragraphs and scene paragraphs into a topic paragraph.

[0170] In this method, the accuracy of the acquired thematic paragraphs is improved by merging domain paragraphs and scenario paragraphs to form a paragraph set, and then using the resulting paragraph set as thematic paragraphs.

[0171] To further ensure the rationality of the thematic paragraphs, it is necessary to conduct further separate analysis and calculations on the thematic paragraphs, which will be explained in detail through the steps shown below.

[0172] After merging the domain paragraphs and scenario paragraphs into a single topic paragraph, the following steps are also included:

[0173] S2581: When the name overlap exceeds the preset version overlap, the corresponding historical data will be used as historical version data, and the corresponding unified format document will be used as the current format document.

[0174] Version overlap refers to the threshold standard when documents are different versions of the same topic. Version overlap is preset by the operator.

[0175] When the name overlap exceeds the preset version overlap, it indicates that there are different versions of the same topic. Therefore, historical version data and current format documents are defined separately for easy use later.

[0176] S2582: Retrieve historical format documents from historical version data.

[0177] Among them, historical format documents refer to unified format documents formed after historical versions of data have been converted.

[0178] By retrieving the unified format documents corresponding to historical versions of the data and using them as historical format documents, it is convenient to use them later.

[0179] S2583: Compare historical format documents with current format documents to identify distinguishing and identical paragraphs.

[0180] Distinguishing paragraphs refer to text paragraphs in historical and current format documents where content differs. Distinguishing paragraphs include added, deleted, or modified content, reflecting the core changes in the document version update. Identical paragraphs refer to text paragraphs in two documents that are completely identical or highly similar in content.

[0181] By establishing paragraph correspondences between historical format documents and current format documents according to sequence or semantic association, and then using text similarity algorithms such as cosine similarity and edit distance to compare content paragraph by paragraph, the matching degree between paragraphs is calculated. The matching degree is then compared with a preset matching threshold. Paragraphs with a matching degree greater than the preset matching threshold are judged as the same paragraphs, while paragraphs with a matching degree less than the preset matching threshold are judged as different paragraphs, thus facilitating subsequent use.

[0182] S2584: Combine identical paragraphs and different paragraphs to generate version paragraphs, and add topic paragraphs.

[0183] Among them, the version paragraph refers to a collection of text paragraphs that fully reflect the document version update trajectory.

[0184] By retaining identical paragraphs as the basis for version inheritance, the differences between the paragraphs are analyzed to determine whether they are additions, deletions, or modifications, and these differences are marked. Then, the marked difference paragraphs are rearranged with the identical paragraphs according to the textual logic to ensure semantic coherence and merge them to form complete version paragraphs. Finally, the version paragraphs are added to thematic paragraphs to improve the accuracy of the obtained thematic paragraphs.

[0185] S26: Select vector models for specific paragraphs and transform them to form a specific vector library.

[0186] Vector models refer to algorithmic models used to transform text paragraphs into high-dimensional vectors. Thematic vector libraries can capture the semantic features, contextual relationships, and professional connotations of text through mathematical forms. Common thematic vector libraries include embedding models such as qwen3-embedding-8b and BGE-M3. Different thematic vector libraries vary in semantic understanding accuracy and domain adaptability. A thematic vector library is a database that stores vectors of thematic paragraphs. The distance between vectors reflects the degree of semantic similarity of the text, supporting efficient semantic retrieval and association analysis.

[0187] By selecting suitable vector models based on the professional characteristics and text features of the thematic paragraphs, priority is given to pre-trained models that support long text processing and have high semantic understanding accuracy, such as qwen3-embedding-8b. When the thematic paragraphs contain a large number of professional terms or special expressions, BGE-M3 is used to vectorize the thematic paragraphs and store them according to the classification method of thematic paragraphs, thereby obtaining a thematic vector library for convenient subsequent use.

[0188] S27: Merge all thematic vector libraries into a single planning vector library.

[0189] In this process, by merging all thematic vector libraries, a planning vector library is obtained, thereby improving the accuracy of the obtained planning vector library.

[0190] S3: After receiving a user's question, vectorize the user's question into a question vector.

[0191] User question information refers to the text information corresponding to a user's question. Question vector information refers to the high-dimensional vector data formed after the user question information is transformed by a vector model.

[0192] Upon receiving a user's question, the system first preprocesses the question by removing irrelevant information such as punctuation and interjections. Then, it calls the same vector model used when building the planning vector library to transform the preprocessed question text into a high-dimensional vector consistent with the topic vector dimension, and uses this vector as the question vector information.

[0193] S4: Use the query vector information to search the planning vector library and generate relevant text blocks.

[0194] Among them, the relevant text block refers to the set of original text paragraphs retrieved from the planning vector library that correspond to the topic vectors with high semantic similarity to the user's question vector information.

[0195] By comparing the user's query vector information with all thematic vectors stored in the planning vector library one by one, the cosine similarity between the two is calculated. The higher the value, the stronger the semantic relationship. Then, relevant text blocks are selected for subsequent use.

[0196] To further ensure the rationality of the relevant text blocks, it is necessary to perform further separate analysis and calculation on the relevant text blocks, which will be explained in detail through the steps shown below.

[0197] Reference Figure 4 The method for generating the relevant text blocks includes the following steps:

[0198] S41: Use the query vector information to retrieve the initial text block from the planning vector library.

[0199] The initial text block refers to the set of original text paragraphs that correspond to the topic vectors that have a basic correlation with the semantics of the question, after preliminary retrieval.

[0200] By comparing the user's query vector information with all thematic vectors stored in the planning vector library one by one, the cosine similarity between the two is calculated. Thematic vectors with a cosine similarity greater than a preset similarity threshold are selected, and the text paragraphs corresponding to the selected thematic vectors are used as initial text blocks for subsequent use.

[0201] S42: Retrieve the initial quantity and text relevance from the initial text block.

[0202] Here, "initial quantity" refers to the number of initial text blocks. "Text relevance" refers to the semantic connection between each initial text block and the user's question.

[0203] By counting the number of initial text blocks and retrieving the count results as the initial quantity, and then retrieving the cosine similarity corresponding to the initial text blocks as the text relevance, it is convenient for subsequent use.

[0204] S43: Determine the selection quantity based on the initial quantity.

[0205] The selection quantity refers to the number of text blocks that are ultimately selected and retained from the initial text block.

[0206] The product between the initial quantity and the preset selection ratio is calculated, and the result is used as the selection quantity for convenient subsequent use.

[0207] The selection ratio refers to the proportional value used when making a selection. The selection ratio is preset by the operator according to actual needs.

[0208] S44: Combine the question vector information with the initial text block to generate a question difficulty value.

[0209] The question difficulty value refers to an indicator used to quantify the complexity of a user's question or the difficulty of answering it.

[0210] By analyzing the question vector information and the initial text block, a question difficulty value is generated for convenient subsequent use.

[0211] To further ensure the reasonableness of the question difficulty value, it is necessary to conduct a further separate analysis and calculation of the question difficulty value, which will be explained in detail through the steps shown below.

[0212] The method for generating the difficulty value of a question includes the following steps:

[0213] S441: Retrieve the amount of question data from the question vector information.

[0214] Among them, the amount of question data refers to the quantitative indicator of the amount of effective semantic information contained in user question information after preprocessing and vector transformation.

[0215] By extracting the effective semantic information from the question vector and using it as the question data, it is easier to use later.

[0216] S442: Determine the reference value for data difficulty based on the amount of data asked.

[0217] The data difficulty reference value refers to a quantitative indicator of difficulty generated based on the scale of the question data. Different question data volumes correspond to different data difficulty reference values.

[0218] The product of the question data volume and the preset data difficulty coefficient is calculated, and the calculation result is used as a reference value for data difficulty for convenient subsequent use.

[0219] The data volume difficulty coefficient is a coefficient used to convert the amount of question data into a reference value for data difficulty. The data volume difficulty coefficient is set in advance by the operator according to actual needs.

[0220] S443: Retrieve the number of vectors from the initial text block.

[0221] The number of vectors in the vector database refers to the number of vectors matched by the initial text block from the vector database.

[0222] The vectors retrieved from the planning vector library are queried from the initial text block, and the retrieved vectors are counted. The count result is then used as the number of vectors in the vector library for later use.

[0223] S444: Determine the reference value for vector library difficulty based on the number of vector libraries.

[0224] Among them, the vector library difficulty reference value refers to the difficulty quantification index generated based on the scale of the number of vector libraries.

[0225] The product of the number of vector libraries and the preset difficulty coefficient of the number of libraries is calculated, and the calculation result is used as a reference value for the difficulty of the vector libraries for convenient use later.

[0226] The difficulty coefficient of the number of vector libraries refers to the coefficient used to convert the number of vector libraries into a reference value for the difficulty of the vector libraries. The difficulty coefficient of the number of vector libraries is set in advance by the operator according to actual needs.

[0227] S445: Calculate the sum of the data difficulty reference value and the vector library difficulty reference value and use it as the comprehensive difficulty reference value, and use the comprehensive difficulty reference value as the question difficulty value.

[0228] The comprehensive difficulty reference value refers to the quantitative difficulty index generated after integrating the amount of question data and the number of vector libraries.

[0229] By calculating the sum of the data difficulty reference value and the vector library difficulty reference value, and using the calculation result as the comprehensive difficulty reference value, and then using the comprehensive difficulty reference value as the question difficulty value, the accuracy of the obtained question difficulty value is improved.

[0230] To further ensure the reasonableness of the question difficulty value, it is necessary to conduct a further separate analysis and calculation of the question difficulty value, which will be explained in detail through the steps shown below.

[0231] After using the overall difficulty reference value as the question difficulty value, the following steps are also included:

[0232] S4451: Collect user account information and question input parameters.

[0233] User account information refers to a user's identity data within the system, including account ID, user type, and historical interaction records. Question input parameters refer to the technical parameters generated when a user asks a question, including parameters such as the proportion of long sentences and terminology density.

[0234] User account information is automatically obtained after the user logs in, and question input parameters are automatically collected after the user enters a question.

[0235] S4452: Determine the user type and input baseline parameters from the user account information.

[0236] User type refers to the identity category of a user, including professional users, public users, student users, etc. Input baseline parameters refer to the baseline technical parameters generated when users of different user types ask questions.

[0237] The system retrieves user type and historical interaction records based on user account information, calculates the average proportion of long sentences and terminology density in the historical interaction records, and combines the calculated results as input benchmark parameters for convenient subsequent use.

[0238] S4453: Compare the input baseline parameters with the problem input parameters to determine the input deviation parameters.

[0239] The input deviation parameter refers to the parameter that corresponds to the situation where there is a deviation between the input parameters.

[0240] By comparing the input baseline parameters with the problem input parameters and analyzing the deviation between them, the input deviation parameters are used as input deviation parameters for convenient subsequent use.

[0241] S4454: Determine the parameter deviation range and unit adjustment value based on the user type.

[0242] Here, the parameter deviation range refers to the allowable deviation range of the question parameters when a user of a given user type uses the question normally. The unit adjustment value refers to the adjustment value required when the question difficulty value needs to be adjusted under the condition that the current question meets the normal questioning requirements of the user type. Different user types correspond to different parameter deviation ranges and unit adjustment values.

[0243] By inputting the user type into a preset user type database, the parameter deviation range and unit adjustment value are obtained for easy subsequent use.

[0244] The user type database pre-stores a table showing the correspondence between different user types and their corresponding parameter deviation ranges and unit adjustment values. The user type database is retrieved after the operator has pre-entered the information.

[0245] S4455: Determine the user similarity value based on the input deviation parameter and the parameter deviation range.

[0246] Among them, user similarity value refers to the quantitative indicator that there is similarity between the user who asked the question and the user corresponding to the user type. The higher the user similarity value, the more significant the similarity.

[0247] By judging whether the input deviation parameter falls within the parameter deviation range, if it falls within the range, 1 is directly output as the deviation reference value. If it does not fall within the range, the median value between the parameter deviation ranges is calculated and used as the median value of the range. Then, the ratio between the input deviation parameter and the median value of the range is calculated as the deviation reference value. Finally, the deviation reference value is weighted according to the weight of different parameters to obtain the user similarity value.

[0248] The weighting of different parameters is preset by the operator for different user types according to actual needs.

[0249] For example, when the user type is a professional user, the weight of long sentences is pre-set to 0.2 and the term density to 0.3. If the deviation reference value corresponding to the calculated long sentence ratio is 0.6 and the deviation reference value corresponding to the calculated term density is 0.8, then the user similarity value is 0.36.

[0250] S4456: Determine the input adjustment reference value by combining the user similarity value and the unit adjustment value.

[0251] The input adjustment reference value refers to the adjustment value corresponding to the difficulty value of the question when it is adjusted based on similar situations.

[0252] The product of the user's similarity value and the unit adjustment value is calculated, and the result is used as an input adjustment reference value for convenient subsequent use.

[0253] S4457: Update and replace the question difficulty value with the input adjustment reference value.

[0254] Specifically, the accuracy of the obtained question difficulty value is improved by calculating the sum between the input adjustment reference value and the question difficulty value, and then replacing and updating the question difficulty value with the calculation result.

[0255] S45: Determine the difficulty adjustment coefficient based on the difficulty value of the question.

[0256] The difficulty adjustment coefficient refers to the coefficient used when adjusting the number of selections based on the difficulty level.

[0257] By analyzing whether the difficulty value of the question falls within the preset coefficient difficulty range, the coefficient corresponding to the coefficient difficulty range is used as the difficulty adjustment coefficient for convenient subsequent use.

[0258] The difficulty range of the coefficients and the corresponding coefficients are all preset by the operator according to actual needs.

[0259] For example, if the pre-defined difficulty range is less than 0.3, the corresponding coefficient is 0.8. If the difficulty range is between 0.3 and 0.7, the corresponding coefficient is 1. If the difficulty range is greater than 0.7, the corresponding coefficient is 1.3.

[0260] S46: Determine the adjustment quantity using the selection quantity and difficulty adjustment coefficient.

[0261] The adjusted quantity refers to the quantity corresponding to the selected quantity after adjustment.

[0262] The product between the selected quantity and the difficulty adjustment coefficient is calculated, and the result is used as the adjustment quantity for convenient subsequent use.

[0263] S47: Select the initial text block as the relevant text block based on text relevance and adjustment quantity.

[0264] The method involves sorting the initial text blocks from highest to lowest text relevance, and then selecting the initial text blocks corresponding to the previous adjustment quantities based on the sorting results as relevant text blocks, thereby improving the accuracy of the obtained relevant text blocks.

[0265] S5: Reorder the relevant text blocks to obtain the selected background knowledge.

[0266] Among them, selecting background knowledge refers to the structured knowledge set formed by reordering relevant text blocks after they have been filtered through semantic logic.

[0267] By reordering and analyzing relevant text blocks, we can obtain the background knowledge for subsequent use.

[0268] To further ensure the rationality of the selected background knowledge, it is necessary to conduct further separate analysis and calculation on the selected background knowledge, which will be explained in detail through the steps shown below.

[0269] The method for generating background knowledge includes the following steps:

[0270] S51: Retrieve the relevant quantity from the relevant text block.

[0271] The relevant quantity refers to the number of relevant text blocks.

[0272] By counting the relevant text blocks and retrieving the counting results as relevant quantities, it is convenient to use them later.

[0273] S52: Determine the relevant ratio value using the relevant quantity and the initial quantity.

[0274] The relevant ratio value refers to the ratio between the relevant quantity and the initial quantity.

[0275] The quotient between the relevant quantity and the initial quantity is calculated, and the result is used as the relevant ratio value for convenient subsequent use.

[0276] S53: Determine the re-ranking ratio by combining the relevant ratio value and the user similarity value.

[0277] The reordering ratio refers to the ratio required when reordering related text blocks.

[0278] By inputting user similarity values ​​into a preset similarity adjustment database to obtain similarity adjustment values, the product between the similarity adjustment values ​​and the relevant ratio values ​​is calculated, and the calculation result is used as the re-sorting ratio value for convenient subsequent use.

[0279] The similarity adjustment database pre-stores different user similarity values ​​and corresponding similarity adjustment values, which are preset by the operator according to requirements.

[0280] S54: Determine the number of reorderings using the reordering ratio and related quantities.

[0281] The number of reorderings refers to the number of elements that need to be selected when performing a reordering.

[0282] The product of the reordering ratio and the relevant quantity is calculated, and the result is used as the reordering quantity for convenient subsequent use.

[0283] S55: Combine the question vector information with the relevant text blocks to re-retrieve in the planning vector library and obtain the reordered text blocks.

[0284] The reordered text block refers to the text block selected during reordering.

[0285] By vectorizing the relevant text blocks to obtain relevant text vector information, and then combining the relevant text vector information with the question vector information, a new search is performed in the planning vector library. The text blocks corresponding to the retrieved vectors are then used as reordered text blocks for convenient subsequent use.

[0286] S56: Retrieve reorder relevance from reordered text blocks.

[0287] Among them, reorder relevance refers to the degree of semantic association between the reordered text block and the user's question.

[0288] By retrieving the cosine similarity corresponding to the reordered text blocks and using it as the text relevance, it is convenient for subsequent use.

[0289] S57: Select reordered text blocks based on reorder relevance and reorder quantity, and use them as background knowledge for selection.

[0290] Specifically, by sorting the reordered text blocks from largest to smallest according to their reordering relevance, and then combining the reordered text blocks corresponding to the number of previous reordered text blocks based on the sorting results as selection background knowledge, the accuracy of the obtained selection background knowledge is improved.

[0291] S6: Input the selected background knowledge and user question information into the preset language model, generate and output the response result information.

[0292] Among them, the large language model refers to an artificial intelligence model trained on massive amounts of text data. The large language model is pre-configured by the operator according to the requirements, and can be a series of models such as GPT, BERT, and T5.

[0293] The response result information refers to the complete answer to the user's question.

[0294] By inputting background knowledge and user questions into a pre-defined language model, the language model generates responses, thus improving the accuracy of answers. This approach involves first querying and matching natural resource planning-related questions with a professional planning vector library before generating responses through the language model.

[0295] Based on the same inventive concept, embodiments of the present invention provide an intelligent question-answering analysis system for natural resource planning, comprising:

[0296] The data collection module is used to collect document information, user account information, and question input parameters.

[0297] The memory stores a program for implementing a natural resource planning intelligent question-answering analysis method as described above;

[0298] The processor loads and executes programs stored in memory.

[0299] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0300] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent question-answering analysis in natural resource planning, characterized in that, include: S1: Upload documents related to natural resource planning; S2: Convert the document data and construct a planning vector library; S3: Upon receiving a user's question, the user's question is vectorized into question vector information; S4: Use the question vector information to retrieve relevant text blocks from the planning vector library; S5: Reorder the relevant text blocks to obtain selected background knowledge; S6: Input the selected background knowledge and the user's question information into a preset language model, generate and output the response result information; After uploading the documents related to natural resource planning, the following are also included: S11: Retrieve the document name from the document data; S12: Query and retrieve the document data from the historical time period and use it as historical data; S13: Retrieve the historical name from the aforementioned historical data; S14: Compare the historical name with the document name to obtain the name overlap. S15: Determine whether the name overlap is lower than the preset complete overlap. S16: If yes, continue uploading the document materials; S17: If not, determine and output the duplicate upload prompt information based on the historical name and the document name.

2. The intelligent question-answering analysis method for natural resource planning according to claim 1, characterized in that, The method for generating the planning vector library includes: S21: Retrieve the document format and topic type from the document data; S22: Convert the document data according to the document format to form a document with a uniform format; S23: Extract text from the uniform format document to obtain the extracted text; S24: Cut the extracted text into blocks to form text paragraphs; S25: Combine the text paragraph with the topic type to generate a topic paragraph; S26: Select and transform vector models for the aforementioned thematic paragraphs to form a thematic vector library; S27: Merge all the aforementioned thematic vector libraries into the planning vector library.

3. The intelligent question-answering analysis method for natural resource planning according to claim 2, characterized in that, The method for generating the thematic paragraphs includes: S251: Determine the topic area and topic scenario according to the topic type; S252: Determine the domain keywords based on the aforementioned topic domain; S253: Compare the text paragraph with the domain keywords to determine the domain relevance; S254: Select the text paragraphs according to the domain relevance to form domain paragraphs; S255: Determine the scenario keywords based on the aforementioned thematic scenario; S256: Compare the text paragraph with the scene keywords to determine the scene relevance; S257: Select text paragraphs according to the scene relevance to form scene paragraphs; S258: Merge the domain paragraph and the scene paragraph to form the topic paragraph.

4. The intelligent question-answering analysis method for natural resource planning according to claim 3, characterized in that, After merging the domain paragraph and the scene paragraph to form the topic paragraph, it also includes: S2581: When the name overlap exceeds the preset version overlap, the corresponding historical data is used as historical version data, and the corresponding unified format document is used as the current format document. S2582: Retrieve historical format documents from the historical version data; S2583: Compare the historical format document with the current format document to determine the distinguishing paragraphs and the identical paragraphs; S2584: Combine the identical paragraphs and the different paragraphs to generate a version paragraph, and add the topic paragraph.

5. The intelligent question-answering analysis method for natural resource planning according to claim 1, characterized in that, The method for generating the relevant text blocks includes: S41: Use the question vector information to retrieve the initial text block from the planning vector library; S42: Retrieve the initial quantity and text relevance from the initial text block; S43: Determine the selection quantity according to the initial quantity; S44: Combine the question vector information with the initial text block to generate a question difficulty value; S45: Determine the difficulty adjustment coefficient based on the question difficulty value; S46: Determine the adjustment quantity using the selected quantity and the difficulty adjustment coefficient; S47: Select an initial text block based on the text relevance and the adjustment quantity, and use it as the relevant text block.

6. The intelligent question-answering analysis method for natural resource planning according to claim 5, characterized in that, The method for generating the question difficulty value includes: S441: Retrieve the amount of question data from the question vector information; S442: Determine the reference value for data difficulty based on the amount of data asked; S443: Retrieve the number of vectors from the initial text block; S444: Determine the reference value for the difficulty of the vector library based on the number of vector libraries mentioned; S445: Calculate the sum of the data difficulty reference value and the vector library difficulty reference value and use it as the comprehensive difficulty reference value, and use the comprehensive difficulty reference value as the question difficulty value.

7. The intelligent question-answering analysis method for natural resource planning according to claim 6, characterized in that, Following the use of the overall difficulty reference value as the question difficulty value, the following is also included: S4451: Collect user account information and question input parameters; S4452: Determine the user type and input baseline parameters from the user account information; S4453: Compare the input reference parameters with the problem input parameters to determine the input deviation parameters; S4454: Determine the parameter deviation range and unit adjustment value based on the user type; S4455: Determine the user similarity value according to the input deviation parameter and the parameter deviation range; S4456: Determine the input adjustment reference value by combining the user similarity value and the unit adjustment value; S4457: Update and replace the question difficulty value with the input adjustment reference value.

8. The intelligent question-answering analysis method for natural resource planning according to claim 7, characterized in that, The method for generating the selected background knowledge includes: S51: Retrieve the relevant quantity from the relevant text block; S52: Determine the relevant ratio value using the relevant quantity and the initial quantity; S53: Determine the re-ranking ratio by combining the relevant ratio value and the user similarity value; S54: Determine the number of reorderings using the reordering ratio value and the relevant quantity; S55: Combine the question vector information with the relevant text block to re-search in the planning vector library to obtain the reordered text block; S56: Retrieve the reordering relevance from the reordered text block; S57: Select the reordered text block according to the reordering relevance and the reordering quantity, and use it as the selection background knowledge.

9. A natural resource planning intelligent question-answering analysis system, characterized in that, include: The data collection module is used to collect document information, user account information, and question input parameters. The memory stores a program for implementing the intelligent question-answering analysis method for natural resource planning as described in any one of claims 1 to 8; The processor loads and executes programs stored in memory.

Images