Low-altitude intelligent question and answer construction method and system based on dynamic parameters

Abstract

The invention relates to a low-altitude intelligent question and answer construction method and system based on dynamic parameters. The method comprises the following steps: collecting low-altitude domain data, cleaning the low-altitude domain data, generating a semantic vector index, and constructing a low-altitude domain knowledge base based on the semantic vector index; receiving a natural language query of a user, analyzing a query intention, extracting keywords in the natural language query, and matching a corresponding candidate word quantity based on query types of the natural language query of the user, the query types at least comprising high-frequency phrase query and low-frequency long-tail query; and respectively carrying out fusion semantic retrieval and keyword retrieval, carrying out secondary sorting on the candidate results based on a preset resorter, preferentially sorting the candidate results related to the query intention, and outputting the corresponding candidate results. By adopting the method, a combined domain retrieval enhancement generation mechanism is provided, so that the professionality and accuracy of answers are improved; and the retrieved knowledge base content is re-screened to increase the hit probability of the knowledge base.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent question answering, and in particular to a method and system for constructing low-altitude intelligent question answering based on dynamic parameters. Background Art

[0002] With the rapid development of artificial intelligence (AI), natural language processing (NLP) and question-answering systems are becoming increasingly commonplace across various fields. The core process of existing NLP and question-answering systems is "input parsing → knowledge matching → answer generation." These systems use natural language understanding to decipher user needs, leverage retrieval technology to locate relevant knowledge, and ultimately generate accurate responses.

[0003] Although generative models (such as the GPT series) have shown outstanding performance in question-answering systems in recent years, their application in the low-altitude domain still faces many challenges. However, existing generative question-answering systems often lack effective integration of domain-specific knowledge, resulting in insufficient accuracy and practicality in their responses. Summary of the Invention

[0004] 1. Problem to be solved Based on this, it is necessary to provide a low-altitude intelligent question-answering construction method and system based on dynamic parameters that can improve knowledge question-answering in specific fields to address the above technical problems.

[0005] 2. Technical solution In the first aspect, this application provides a method for constructing low-altitude intelligent question-answering based on dynamic parameters. The method includes: Collecting low-altitude field data and cleaning the low-altitude field data, performing block processing on the cleaned low-altitude field data and generating a semantic vector index, and building a low-altitude field knowledge base based on the semantic vector index; Receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; Based on the keywords, extended list and number of selected words, fused semantic retrieval and keyword retrieval are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

[0006] In one embodiment, the block processing of the cleaned low-altitude area data includes: Parent-child segmentation and block processing and sliding window block processing; The parent-child segmentation and block processing specifically includes: Cutting the cleaned low-altitude area data into sub-text blocks of preset sizes; Uniquely identify each sub-text block and record its location information in the original text; Identify the parent text unit containing the child text block and set it as the parent module; Establish a mapping relationship between the child text block and the parent module.

[0007] In one embodiment, the block processing of the cleaned low-altitude area data includes: The sliding window block processing specifically includes: Setting window parameters and performing a block operation on the low-altitude area data based on the window parameters, wherein the window parameters include at least the size and step size of the sliding window; Intercept the first text block according to the set window parameters and record the starting and ending positions; intercepting subsequent text blocks by sliding the window according to a set step size, where there are overlapping texts between adjacent text blocks; When the length of the remaining text block is less than the window length, the window length is adjusted to generate the final text block.

[0008] In one embodiment, generating a semantic vector index includes: Normalize each text block; Storing semantic vectors of text blocks based on a preset vector database; Adding metadata tags to the semantic vector, wherein the metadata tags include at least the block type, the original data source, and the associated keywords; Establish a vector index structure and output the semantic vector index.

[0009] In one embodiment, performing fusion semantic search and keyword search based on the keywords, the expansion list, and the number of selected words includes: The semantic retrieval calculates the correlation between the text vector and the query vector based on cosine similarity; The keyword search locates text containing keywords based on an inverted index; The relevance of the results is evaluated according to the preset weights.

[0010] In one embodiment, performing secondary sorting on the candidate results based on a preset rearranger includes: After pre-processing the candidate structure, a core element set for correlation evaluation is preset, wherein the core element set includes at least entity elements, process elements, and environment elements; Dynamically assign factor weights based on user query intent; Count the frequency of core elements in the document and calculate the frequency matching score based on the frequency ratio; Use pre-trained language models to calculate the semantic matching scores between documents and core elements; Calculate the scene matching score based on the user query scene and the preset document scene label; The comprehensive relevance score is calculated based on the frequency matching score, the semantic matching score and the scene matching score, the candidate results are sorted in descending order based on the comprehensive relevance score and the sorted candidate results are output.

[0011] Secondly, this application also provides a low-altitude intelligent question-answering construction system based on dynamic parameters. The system includes: A low-altitude knowledge base construction module is used to collect low-altitude field data and clean the low-altitude field data, perform block processing on the cleaned low-altitude field data and generate semantic vector indexes, and construct a low-altitude field knowledge base based on the semantic vector indexes; A query intent parsing module is configured to receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an expanded list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; The query intent output module is used to perform fusion semantic retrieval and keyword retrieval based on the keywords, extension list and number of selected words, quickly screen candidate results from the low-altitude domain knowledge base, perform secondary sorting on the candidate results based on a preset rearranger, prioritize candidate results related to the query intent and output the corresponding candidate results.

[0012] In a third aspect, the present application further provides a computer system. The computer system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the following steps are performed: Collecting low-altitude field data and cleaning the low-altitude field data, performing block processing on the cleaned low-altitude field data and generating a semantic vector index, and building a low-altitude field knowledge base based on the semantic vector index; Receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; Based on the keywords, extended list and number of selected words, fused semantic retrieval and keyword retrieval are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

[0013] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps: Collecting low-altitude field data and cleaning the low-altitude field data, performing block processing on the cleaned low-altitude field data and generating a semantic vector index, and building a low-altitude field knowledge base based on the semantic vector index; Receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; Based on the keywords, extended list and number of selected words, fused semantic retrieval and keyword retrieval are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

[0014] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps: Collecting low-altitude field data and cleaning the low-altitude field data, performing block processing on the cleaned low-altitude field data and generating a semantic vector index, and building a low-altitude field knowledge base based on the semantic vector index; Receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; Based on the keywords, extended list and number of selected words, fused semantic retrieval and keyword retrieval are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

[0015] 3. Beneficial effects This application adopts the above method. To address the problem of insufficient integration of specific domain knowledge in traditional retrieval-based question-answering systems, the present invention proposes a mechanism that combines domain retrieval to enhance generation, thereby improving the professionalism and accuracy of answers; the retrieved knowledge base content is re-screened to increase the probability of knowledge base hits. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Flowchart of a method for constructing low-altitude intelligent question-answering based on dynamic parameters in one embodiment; Figure 2 A flowchart of constructing a low-altitude domain knowledge base in one embodiment; Figure 3 This is a user interface diagram of a method for constructing low-altitude intelligent question-answering based on dynamic parameters in one embodiment; Figure 4 A diagram of a user interface for a method for constructing low-altitude intelligent question-answering based on dynamic parameters in another embodiment; Figure 5 A structural block diagram of a low-altitude intelligent question-answering system based on dynamic parameters in one embodiment; Figure 6 FIG. 1 is a diagram showing the internal structure of a computer system in one embodiment. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0018] The embodiment of the present application provides a method for constructing low-altitude intelligent question-answering based on dynamic parameters, which includes the following steps: Step 202 : collecting low-altitude domain data and cleaning the low-altitude domain data, dividing the cleaned low-altitude domain data into blocks and generating a semantic vector index, and building a low-altitude domain knowledge base based on the semantic vector index.

[0019] Among them, Figure 1 and Figure 2 As shown in the figure, the low-altitude field knowledge base data storage and cleaning process collects massive amounts of data related to the low-altitude field from multiple sources. After the data storage is completed, the raw data is cleaned, including removing redundant information, eliminating noise data, and supplementing missing information. Through a combination of automated annotation and manual verification, the accuracy and completeness of the knowledge base are ensured. It is worth mentioning that a knowledge base for the low-altitude domain was constructed. After data cleaning, the text data in the knowledge base was encoded and processed using vector representation technology to generate a semantic vector index. Based on this, a knowledge base system supporting multiple retrieval methods was constructed. Data cleaning: The collected data was preprocessed to remove redundant and noisy information. Deduplication involves identifying and removing duplicate data entries to ensure the uniqueness of each data item. Format unification involves converting data from different sources into a unified format to facilitate subsequent processing and analysis. Error correction involves checking for errors in the data and correcting them through manual verification or automatic algorithmic correction.

[0020] Step 204: Receive the user's natural language query and parse the query intent, extract keywords in the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, where the query type includes at least high-frequency phrase queries and low-frequency long-tail queries.

[0021] Among them, user input processing and demand analysis, keyword extraction and dynamic parameter generation, knowledge base retrieval, and result integration and generation.

[0022] Step 206, based on the keywords, the extended list and the number of selected words, a fusion semantic search and a keyword search are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

[0023] Among them, system design and platform construction: deploy the constructed Q&A robot workflow to the server, and complete the overall system architecture design and functional module division.

[0024] In the above-mentioned low-altitude intelligent question-answering construction method based on dynamic parameters, the present invention addresses the problem of insufficient integration of specific domain knowledge in traditional retrieval-based question-answering systems, and proposes a mechanism for enhancing generation by combining domain retrieval, thereby improving the professionalism and accuracy of answers; the retrieved knowledge base content is re-screened to increase the probability of hitting the knowledge base.

[0025] In one embodiment, the block processing of the cleaned low-altitude area data includes: Parent-child segmentation and block processing and sliding window block processing; the parent-child segmentation and block processing specifically includes: dividing the cleaned low-altitude field data into sub-text blocks of preset size; uniquely identifying each sub-text block and recording the position information in the original text; identifying the parent text unit containing the sub-text block and setting it as the parent module; establishing a mapping relationship between the sub-text block and the parent module.

[0026] Parent-child segmentation is a context-based segmentation method that aims to preserve the logical connections between text content. This method divides the entire text into multiple interconnected chunks by analyzing the dependencies or thematic consistency between sentences or paragraphs in the text.

[0027] Specifically, this process can be divided into two main stages: "submodule matching query" and "parent module providing context".

[0028] First, during the "submodule matching query" phase, the text is segmented into very small chunks, with each sentence having its own chunk. This fine-grained segmentation enables precise information location, quickly finding specific content highly relevant to the user's query. This fine-grained segmentation allows for rapid targeting of target information during retrieval.

[0029] Next, the "Parent Module Provides Context" phase focuses on expanding and enriching these matching results. Once a sub-block is matched, the system associates it with its corresponding parent module to provide additional context and relevant details. This hierarchical association mechanism ensures that the information returned to the user is not only accurate but also comprehensive.

[0030] It is worth mentioning that the sliding window block processing specifically includes: The method includes setting window parameters and performing a block operation on the low-altitude area data based on the window parameters, wherein the window parameters include at least the size and step size of the sliding window; intercepting the first text block according to the set window parameters and recording the starting and ending positions; sliding the window according to the set step size to intercept subsequent text blocks, where there is overlapping text between adjacent text blocks; and adjusting the window length to generate the final text block when the length of the remaining text block is less than the window length.

[0031] Sliding window is a location-based segmentation method that extracts content fragments by moving a fixed-size "window" through the text. This method does not rely on the content or logical relationships of the text, but instead segments it according to predefined rules.

[0032] Specifically, the process consists of the following steps: First, the window parameters are set, determining the sliding window size (containing 5 sentences or 200 characters) and step size (moving 1 sentence or 100 characters per step). Next, chunking is performed according to the set parameters. Starting from the beginning of the text, the content of the current window size is intercepted as the first chunk. The window is moved forward according to the set step size, and the content is repeatedly intercepted to form new chunks. If the remaining text is less than the length of a full window, the chunk size is adjusted to handle the border content.

[0033] In one embodiment, generating a semantic vector index includes: Each text block is standardized; the semantic vector of the text block is stored based on a preset vector database; metadata tags are added to the semantic vector, and the metadata tags include at least the block type, the original data source and the associated keywords; a vector index structure is established and the semantic vector index is output.

[0034] The text blocks are encoded using a semantic vector model, and the bge-m3 embedding model is loaded using Ollama. bge-m3 is an open-source, Transformer-based large-scale language model suitable for a variety of Chinese NLP tasks. By inputting text data into the model, it generates high-dimensional semantic vector representations. These vectors reflect the semantic characteristics of the text content and provide foundational support for subsequent analysis and processing. During the encoding process, text blocks are batch processed to ensure the model can efficiently process large amounts of text data.

[0035] After encoding, a set of low-dimensional vector representations is obtained, which can be used for further dimensionality reduction. Through dimensionality reduction techniques, these semantic vectors can be displayed in two-dimensional or three-dimensional space, allowing for more intuitive observation of the similarities and distribution relationships between texts.

[0036] In one embodiment, performing fusion semantic search and keyword search based on the keywords, the expansion list, and the number of selected words includes: Semantic search calculates the relevance between the text vector and the query vector based on cosine similarity; the keyword search locates the text containing the keyword based on the inverted index; and the relevance of the result is evaluated according to a preset weight.

[0037] Among them, the database retrieval methods include: semantic retrieval, keyword retrieval, hybrid retrieval, and two-stage retrieval.

[0038] Specifically, the content of each search method is as follows: Semantic retrieval is an information retrieval method based on the deep meaning of text. It converts text into vector representation through natural language processing technology and judges relevance based on the similarity between these vectors.

[0039] In semantic search, the text is first subjected to a series of preprocessing operations, such as word segmentation, part-of-speech tagging, and named entity recognition. Word segmentation breaks down continuous text into meaningful words, part-of-speech tagging assigns each word its part of speech, and named entity recognition identifies specific entities within the text, such as names of people, places, and organizations. Through these preprocessing operations, the text is converted into a form that is easier for computers to process.

[0040] Next, the processed text is converted into vectors using word embedding techniques. Word embedding is a technique that maps words into a low-dimensional vector space, such that words with similar semantics are closer in the vector space. Through these models, each word is represented as a vector, and the combination of all word vectors in the text forms the vector representation of the text. Finally, the relevance of the text is judged based on the similarity between these vectors. Available similarity calculation methods include cosine similarity, Euclidean distance, etc. Cosine similarity measures the similarity between two vectors by calculating the cosine value of the angle between them, and the closer the value is to 1, the higher the similarity.

[0041] Keyword search is a method that directly matches based on the keywords input by the user in the document. By preprocessing the text and building an index, it can quickly locate the content containing specific keywords.

[0042] Keyword search is a method that directly matches based on the keywords input by the user in the document. It is one of the most traditional and common retrieval methods. To quickly locate the content containing specific keywords, the text needs to be preprocessed and an index needs to be built. The preprocessing process is similar to semantic retrieval, including operations such as word segmentation and stop word removal. Stop words refer to words that frequently appear in the text but have little significance for retrieval, such as "de" (的), "shi" (是), "zai" (在), etc. Removing stop words can reduce the amount of data and improve the retrieval efficiency.

[0043] The next step is to build an index. An index is a data structure that records the location information of keywords in the document. An inverted index uses keywords as index entries and records all the documents containing that keyword and their locations in the document. When the user inputs keywords for retrieval, the system can directly locate the documents containing that keyword quickly through the index, without having to traverse the entire document collection. This method greatly improves the retrieval speed, especially suitable for retrieving large-scale document collections.

[0044] Hybrid retrieval combines the advantages of semantic retrieval and keyword retrieval. It searches using both keywords and natural language descriptions, and comprehensively evaluates the relevance of the results according to the set weights (keyword weight and semantic similarity weight), and returns the retrieved content based on the comprehensively evaluated results.

[0045] Two-stage retrieval is an optimized retrieval process. Its main idea is to process the data in stages to improve the retrieval effect while ensuring efficiency. In the first stage, a large number of candidate results are quickly screened through hybrid retrieval. Since hybrid retrieval combines the efficiency of keyword retrieval and the accuracy of semantic retrieval, it can screen out candidate documents related to the user's query from a large-scale dataset in a relatively short time.

[0046] In the second stage, a re-ranker is used to further rank and optimize these candidate documents. The re-ranker can re-rank candidate documents based on additional information, such as document quality and accurate calculation of relevance scores. This further improves the quality of search results, placing the most relevant documents at the top. Two-stage retrieval reduces the amount of data in the first stage, reducing the complexity of subsequent processing, while the second stage improves the quality of the results, making them more accurate and useful.

[0047] In one embodiment, performing secondary sorting on the candidate results based on a preset rearranger includes: After preprocessing based on the candidate structure, a set of core elements for relevance evaluation is preset, and the core element set includes at least entity elements, process elements and environment elements; element weights are dynamically allocated based on user query intentions; the frequency of core elements contained in the document is counted and the frequency matching score is calculated according to the frequency ratio; the semantic matching score between the document and the core elements is calculated using a pre-trained language model; the scenario matching score is calculated based on the user query scenario and the preset document scenario label; the comprehensive relevance score is calculated based on the frequency matching score, semantic matching score and scenario matching score, the candidate results are arranged in descending order based on the comprehensive relevance score and the sorted candidate results are output.

[0048] like Figure 3 and Figure 4 As shown in the figure, building a complete Q&A robot workflow includes the following steps: User input processing: Receive users' natural language queries and use large models to analyze their intentions and needs; Keyword extraction and dynamic parameter generation: A large model is used to break down user questions into several core keywords and TopK parameters. Keyword weights are automatically adjusted, and an expanded list of polysemous and synonymous words is generated. Differentiating user questions based on the number of keywords and their level of expertise. A compact candidate set (Top 100) is used for high-frequency, short queries, while the list is expanded to the Top 500 for low-frequency, long-tail queries. Knowledge base retrieval: Based on the extracted keywords and dynamic parameters, multi-dimensional information retrieval is performed through the low-altitude field knowledge base constructed above to obtain relevant professional content; result integration and generation: The retrieved professional content is subjected to semantic analysis, logical integration and natural language generation using a large model, and ultimately outputs answers that meet user needs.

[0049] It is worth mentioning that the front-end interface rendering: the entire system is based on the Streamlit framework for front-end rendering. Users can upload PDF or Word documents through the interface, and automatically parse and store the content in the database to ensure data security and persistence. System function module: The system function module includes options such as uploading documents, loading saved database content, selecting large language models and embedding models. Users can configure various parameters according to specific needs to optimize the performance of the retrieval algorithm. After completing all settings, the system can ask and answer questions for the selected knowledge base and return accurate results according to the preset process.

[0050] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0051] Based on the same inventive concept, the embodiment of the present application also provides a low-altitude intelligent question-answering construction system based on dynamic parameters for implementing the above-mentioned low-altitude intelligent question-answering construction method based on dynamic parameters. The implementation solution provided by this system is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations in the embodiments of one or more low-altitude intelligent question-answering construction systems based on dynamic parameters provided below can be found in the above-mentioned limitations on the low-altitude intelligent question-answering construction method based on dynamic parameters, and will not be repeated here.

[0052] In one embodiment, Figure 5 As shown, a low-altitude intelligent question-answering construction system based on dynamic parameters is provided, including: a low-altitude knowledge base construction module, a query intent parsing module and a query intent output module, wherein: A low-altitude knowledge base construction module is used to collect low-altitude field data and clean the low-altitude field data, perform block processing on the cleaned low-altitude field data and generate semantic vector indexes, and construct a low-altitude field knowledge base based on the semantic vector indexes; A query intent parsing module is configured to receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an expanded list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; The query intent output module is used to perform fusion semantic retrieval and keyword retrieval based on the keywords, extension list and number of selected words, quickly screen candidate results from the low-altitude domain knowledge base, perform secondary sorting on the candidate results based on a preset rearranger, prioritize candidate results related to the query intent and output the corresponding candidate results.

[0053] In one embodiment, the low-altitude knowledge base construction module is also used for: parent-child segmentation and block processing and sliding window block processing; the parent-child segmentation and block processing specifically includes: dividing the cleaned low-altitude field data into sub-text blocks of preset size; uniquely identifying each sub-text block and recording the position information in the original text; identifying the upper-level text unit containing the sub-text block and setting it as the parent module; establishing a mapping relationship between the sub-text block and the parent module.

[0054] In one embodiment, the low-altitude knowledge base construction module is also used for: the sliding window block processing specifically includes: setting window parameters and performing block operations on the low-altitude field data based on the window parameters, the window parameters at least include the size and step size of the sliding window; intercepting the first text block according to the set window parameters and recording the starting and ending positions; intercepting subsequent text blocks according to the set step size sliding window, and there is overlapping text between adjacent text blocks; when the length of the remaining text block is less than the window length, adjusting the window length to generate the final text block.

[0055] In one embodiment, the query intent parsing module is further used to: standardize each text block; store the semantic vector of the text block based on a preset vector database; add metadata tags to the semantic vector, wherein the metadata tags include at least the block type, the original data source and the associated keywords; establish a vector index structure and output the semantic vector index.

[0056] In one embodiment, the query intent parsing module is further used for: the semantic retrieval calculates the correlation between the text vector and the query vector based on cosine similarity; the keyword retrieval locates the text containing the keyword based on the inverted index; and the result correlation is evaluated according to a preset weight.

[0057] In one embodiment, the query intention output module is also used to: preset a set of core elements for relevance evaluation after preprocessing based on the candidate structure, the core element set including at least entity elements, process elements and environment elements; dynamically allocate element weights based on the user's query intention; count the frequency of the core elements in the document and calculate the frequency matching score according to the frequency ratio; use a pre-trained language model to calculate the semantic matching score between the document and the core elements; calculate the scene matching score based on the user query scene and the preset document scene label; calculate the comprehensive relevance score based on the frequency matching score, semantic matching score and scene matching score, arrange the candidate results in descending order based on the comprehensive relevance score and output the sorted candidate results.

[0058] Each module in the dynamic parameter-based low-altitude intelligent question-answering system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer system in hardware form, or stored in a computer system memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0059] In one embodiment, a computer system is provided. The computer system may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer system includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer system is used to provide computing and control capabilities. The memory of the computer system includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer system is used to store data. The network interface of the computer system is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a low-altitude intelligent question-answering construction method based on dynamic parameters is implemented.

[0060] In one embodiment, a computer system is provided. The computer system may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown. The computer system includes a processor, memory, communication interface, display screen and input system connected via a system bus. The processor of the computer system is used to provide computing and control capabilities. The memory of the computer system includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The communication interface of the computer system is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a low-altitude intelligent question-answering construction method based on dynamic parameters is implemented.

[0061] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer system to which the solution of the present application is applied. The specific computer system may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0062] In one embodiment, a computer system is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0063] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0064] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0065] It should be noted that the user information (including but not limited to user system information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0066] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0067] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for constructing low-altitude intelligent question answering based on dynamic parameters, characterized in that: The method comprises: Collecting low-altitude field data and cleaning the low-altitude field data, performing block processing on the cleaned low-altitude field data and generating a semantic vector index, and building a low-altitude field knowledge base based on the semantic vector index; Receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an extended list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; Based on the keywords, extended list and number of selected words, fused semantic retrieval and keyword retrieval are performed respectively, candidate results are quickly screened from the low-altitude domain knowledge base, and the candidate results are secondary sorted based on a preset rearranger, and candidate results related to the query intent are prioritized and the corresponding candidate results are output.

2. The method for constructing low-altitude intelligent question-answering based on dynamic parameters according to claim 1, characterized in that: The block processing of the cleaned low-altitude area data includes: Parent-child segmentation and block processing and sliding window block processing; The parent-child segmentation and block processing specifically includes: Cutting the cleaned low-altitude area data into sub-text blocks of preset sizes; Uniquely identify each sub-text block and record its location information in the original text; Identify the parent text unit containing the child text block and set it as the parent module; Establish a mapping relationship between the child text block and the parent module.

3. The method for constructing low-altitude intelligent question-answering based on dynamic parameters according to claim 2, characterized in that: The block processing of the cleaned low-altitude area data includes: The sliding window block processing specifically includes: Setting window parameters and performing a block operation on the low-altitude area data based on the window parameters, wherein the window parameters include at least the size and step size of the sliding window; Intercept the first text block according to the set window parameters and record the starting and ending positions; intercepting subsequent text blocks by sliding the window according to a set step size, where there are overlapping texts between adjacent text blocks; When the length of the remaining text block is less than the window length, the window length is adjusted to generate the final text block.

4. The method for constructing low-altitude intelligent question-answering based on dynamic parameters according to claim 3, characterized in that: Generating a semantic vector index includes: Normalize each text block; Storing semantic vectors of text blocks based on a preset vector database; Adding metadata tags to the semantic vector, wherein the metadata tags include at least the block type, the original data source, and the associated keywords; Establish a vector index structure and output the semantic vector index.

5. The method for constructing low-altitude intelligent question-answering based on dynamic parameters according to claim 1, characterized in that: The fusion semantic search and keyword search based on the keywords, the extended list and the number of selected words respectively include: The semantic retrieval calculates the correlation between the text vector and the query vector based on cosine similarity; The keyword search locates text containing keywords based on an inverted index; The relevance of the results is evaluated according to the preset weights.

6. The method for constructing low-altitude intelligent question-answering based on dynamic parameters according to claim 1, characterized in that: The secondary sorting of the candidate results based on the preset rearranger includes: After pre-processing the candidate structure, a core element set for correlation evaluation is preset, wherein the core element set includes at least entity elements, process elements, and environment elements; Dynamically assign factor weights based on user query intent; Count the frequency of core elements in the document and calculate the frequency matching score based on the frequency ratio; Use pre-trained language models to calculate the semantic matching scores between documents and core elements; Calculate the scene matching score based on the user query scene and the preset document scene label; The comprehensive relevance score is calculated based on the frequency matching score, the semantic matching score and the scene matching score, the candidate results are sorted in descending order based on the comprehensive relevance score and the sorted candidate results are output.

7. A low-altitude intelligent question-answering system based on dynamic parameters, characterized in that: The system comprises: A low-altitude knowledge base construction module is used to collect low-altitude field data and clean the low-altitude field data, perform block processing on the cleaned low-altitude field data and generate semantic vector indexes, and construct a low-altitude field knowledge base based on the semantic vector indexes; A query intent parsing module is configured to receive a user's natural language query and parse the query intent, extract keywords from the natural language query, generate an expanded list of polysemous words and synonyms based on the keywords, dynamically adjust the weights of different keywords, and match the corresponding number of candidate words based on the query type of the user's natural language query, which includes at least high-frequency phrase queries and low-frequency long-tail queries; The query intent output module is used to perform fusion semantic retrieval and keyword retrieval based on the keywords, extension list and number of selected words, quickly screen candidate results from the low-altitude domain knowledge base, perform secondary sorting on the candidate results based on a preset rearranger, prioritize candidate results related to the query intent and output the corresponding candidate results.

8. A computer system comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images