Grain depot intelligent question and answer method, equipment and medium

Abstract

The invention discloses a grain depot intelligent question and answer method and device and a medium. The method comprises the steps that a grain depot knowledge base is constructed; documents of the grain depot knowledge base comprise grain depot management files, historical operation records and grain depot monitoring data; training the general large model according to the special data of the grain depot field to obtain a grain depot special model; when a grain depot problem is received, loading a target document related to the grain depot problem from the knowledge base; performing text segmentation on the target document to obtain a plurality of text blocks, and determining a knowledge text vector of each text block; performing text vectorization on the grain depot problem to obtain a problem text vector; determining related text blocks of the grain depot problem according to the knowledge text vector and the problem text vector of each text block; and inputting the related text blocks and the grain depot questions into the grain depot special model to obtain answer data. Information acquisition and decision analysis can be efficiently and automatically performed on user questions.

Description

Technical Field

[0001] This application relates to the field of grain depot management technology, and in particular to a method, equipment and medium for intelligent question answering in grain depots. Background Technology

[0002] With the increasingly complex food security situation, the management needs of grain depots are rising. As the core institutions for grain storage and management, grain depots bear the important task of ensuring grain reserves and quality safety.

[0003] In existing technologies, most grain depot management systems are still limited to independent file storage and monitoring functions, forming information silos and making it difficult to achieve comprehensive data analysis and related applications. For example, although monitoring systems can provide monitoring of environmental data such as temperature and humidity, they cannot achieve real-time data sharing and comprehensive analysis.

[0004] Furthermore, knowledge management systems typically rely on simple keyword retrieval and cannot provide semantic-based intelligent search (e.g., searching for "temperature" can only match content containing the word "temperature"). They struggle to handle semantically related issues (e.g., "What to do if the warehouse is too hot" cannot be linked to the regulation "ventilation is required when the temperature exceeds 25°C"). Grain depot information retrieval is inefficient and lacks automatic decision-making and analysis capabilities.

[0005] In conclusion, the grain depot management system lacks intelligent, efficient, and comprehensive analysis and decision-making capabilities, and users cannot interact efficiently through natural language. Summary of the Invention

[0006] This application provides a method, device, and medium for intelligent question answering in grain depots, which addresses the problem that grain depot management systems lack intelligent, efficient, and comprehensive analysis and decision-making capabilities, and that users cannot interact efficiently through natural language.

[0007] The embodiments of this application adopt the following technical solutions: On one hand, this application provides an intelligent question-answering method for grain depots. The method includes: constructing a grain depot knowledge base; the documents in the knowledge base include grain depot management documents, historical operation records, and grain depot monitoring data; training a general-purpose model based on specialized data from the grain depot field to obtain a grain depot-specific model; upon receiving a grain depot question, loading a target document related to the question from the knowledge base; performing text segmentation on the target document to obtain multiple text blocks, and determining the knowledge text vector for each text block; vectorizing the grain depot question to obtain a question text vector; determining relevant text blocks for the grain depot question based on the knowledge text vector of each text block and the question text vector; and inputting the relevant text blocks and the grain depot question into the grain depot-specific model to obtain answer data.

[0008] In one example, the target document is segmented into multiple text blocks, specifically including: traversing the target document character by character; and dividing the target document into multiple text blocks with a fixed number of characters based on a fixed number of characters per block and an overlap parameter between blocks.

[0009] In one example, based on the fixed number of characters per block and the overlap between blocks, the target document is divided into multiple text blocks with a fixed number of characters. Specifically, this includes: generating the first text block when the cumulative number of characters reaches the fixed number; reusing the preset overlapping number of characters at the end of the first text block, and then continuing to extract the remaining characters from the end position of the first text block to obtain the second text block; the sum of the remaining characters and the preset overlapping number of characters is the fixed number of characters; until the entire document is divided, several continuous and partially overlapping text blocks with a fixed number of characters are obtained.

[0010] In one example, determining the relevant text blocks for the grain depot problem based on the knowledge text vector of each text block and the question text vector specifically includes: calculating the similarity between each knowledge text vector and the question text vector; and determining the text blocks corresponding to a preset number of knowledge text vectors with the highest similarity ranking as relevant text blocks.

[0011] In one example, calculating the similarity between each knowledge text vector and the question text vector specifically includes: calculating the Euclidean norm of each knowledge text vector and the Euclidean norm of the question text vector; calculating the dot product between each knowledge text vector and the question text vector; and obtaining the cosine similarity between a single knowledge text vector and the question text vector based on the Euclidean norm of the individual knowledge text vector, the Euclidean norm of the question text vector, and the dot product between the individual knowledge text vector and the question text vector.

[0012] In one example, determining the knowledge text vector for each text block involves converting each word in each text block into a high-dimensional vector representation using the bge-large-zh model, thus obtaining the knowledge text vector for each text block.

[0013] In one example, after inputting the relevant text block and the grain depot question into the grain depot-specific model to obtain the answer data, the method further includes: extracting unsatisfactory question-and-answer results from the user's feedback data; and updating and training the grain depot-specific model based on the unsatisfactory question-and-answer results to update the grain depot-specific model.

[0014] In one example, loading target documents related to the grain depot issue from the knowledge base specifically includes: performing semantic parsing on the user question to determine the knowledge category; matching the knowledge category with document tags in the knowledge base to obtain candidate documents; and loading the candidate documents through the unstructured library to determine the target document.

[0015] On the other hand, embodiments of this application provide a smart question-and-answer device for grain depots, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a smart question-and-answer method for grain depots as described above.

[0016] On the other hand, embodiments of this application provide a non-volatile computer storage medium for intelligent question-and-answer in grain depots, which stores computer-executable instructions capable of executing any of the above-described intelligent question-and-answer methods for grain depots.

[0017] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By constructing a grain depot knowledge base that integrates three types of core data, scattered information is uniformly classified and stored. Real-time integration with monitoring and storage management systems enables dynamic updates, allowing access to comprehensive information without switching between multiple systems. In particular, monitoring data (such as sudden temperature changes in warehouses) can be instantly incorporated into the knowledge base, ensuring that subsequent responses are based on the latest state of the grain depot, thus solving the problem of data lag in traditional systems.

[0018] By training a general-purpose large model with grain depot-specific data, the model's ability to understand professional scenarios is enhanced, thereby improving the accuracy of its responses.

[0019] By dividing the loaded document into smaller blocks for more efficient information processing and retrieval, and through a process of text segmentation → vectorization → vector matching, a rapid and accurate initial retrieval of semantic-level knowledge content related to the user's question is achieved.

[0020] By inputting relevant text blocks and grain depot questions into a dedicated grain depot model, in-depth semantic understanding and logical decision analysis can be performed on the relevant text blocks and grain depot questions through a dedicated large model. This supports multi-turn dialogues and complex decisions, and quickly outputs accurate question-and-answer data. Attached Figure Description

[0021] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which: Figure 1A flowchart illustrating an intelligent question-and-answer method for grain depots provided in this application embodiment; Figure 2 A schematic diagram of the architecture of a grain depot intelligent question-and-answer system provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a smart question-and-answer device for grain depots, provided as an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart illustrating an intelligent question-and-answer method for grain depots, provided as an embodiment of this application. This process can be executed by computing devices in the relevant field, and certain input parameters or intermediate results within the process can be manually adjusted to help improve accuracy.

[0025] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.

[0026] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.

[0027] Existing grain depot management models mostly rely on traditional knowledge management and monitoring systems. The lack of effective integration between these systems leads to information isolation and insufficient data sharing. Especially when facing increasingly complex grain depot management scenarios, traditional grain depot management systems have limited intelligent capabilities in knowledge management, grain depot status monitoring, and decision support. For example, they suffer from low knowledge management efficiency, untimely data acquisition, poor system integration, limited functionality, and inability to perform semantic analysis and decision-making, thus failing to meet the needs of modern management.

[0028] In some embodiments of this application, the system mainly consists of two core modules: a knowledge management module and an intelligent question-and-answer module. These two modules are integrated through a unified knowledge base. Simultaneously, the system seamlessly interfaces with grain depot monitoring systems and warehouse management systems, enabling real-time data acquisition and updates. The intelligent question-and-answer module facilitates efficient and accurate information interaction within grain depot management systems.

[0029] The knowledge management module categorizes, labels, and stores documents related to grain depot management. It can update the knowledge base via local uploads, collecting management regulations, operation manuals, and other documents from the grain depot, and can also dynamically update monitoring data and historical operation records from other management systems.

[0030] For documents related to grain depot management, users can quickly search and consult relevant management documents, operation manuals, emergency plans, etc. online according to their needs. They can also ask questions in natural language, and the system will provide specific answers or paragraphs related to their queries based on a large model. This integrated approach allows users to directly obtain key information within documents and locate relevant files, rather than simply searching for files using keywords or directories.

[0031] Intelligent Question Answering Module: Trained using a large language model, this module allows users to query grain depot management documents and dynamic data via natural language. It automatically retrieves answers from a knowledge base based on user input. Users can ask questions such as "What is the current temperature of the grain depot?" or "What are the recent outbound records?" The system provides accurate answers through semantic understanding and supports multi-turn dialogue based on context.

[0032] Therefore, some of the functions are described as follows: Deep integration of real-time data acquisition and knowledge management: The grain depot knowledge management module seamlessly interfaces with existing management systems (such as warehouse management systems, monitoring systems, and automated equipment management systems) to ensure real-time data acquisition and processing. Integration methods include API interfaces, data buses, or microservice architectures. The system obtains real-time monitoring data, inventory information, and operation logs from different modules through these methods. The integrated system can enhance the level of automated management in grain depots without changing existing management processes.

[0033] Regularly updated master model: To maintain the accuracy of question and answer responses and the efficient operation of the system, a mechanism for regularly updating the master model is designed. Based on the latest management data, operation records, and user feedback, the system can periodically fine-tune the master model to adapt to new business needs and changes.

[0034] Based on this Figure 1 The process includes the following steps: S101: Construct a grain depot knowledge base; the documents in the grain depot knowledge base include grain depot management documents, historical operation records, and grain depot monitoring data.

[0035] In some embodiments of this application, the key data collected includes grain depot management documents, monitoring data, and historical operation records. Through data cleaning, formatting, and classification, their accuracy and consistency are ensured.

[0036] For example, grain depots need to manage the storage process of wheat and corn. They have existing "Wheat Storage Temperature Control Specifications", "Corn Outbound Operation Manual", as well as sensors to monitor the temperature and humidity of the warehouse in real time, and nearly three months of grain entry and exit records.

[0037] Data Acquisition: The system automatically collects the above-mentioned specifications and manuals (management documents), and obtains the real-time temperature of 22℃ and humidity of 65% in Warehouse No. 1 (monitoring data) through the connection with the monitoring system, as well as the record of 500 tons of wheat entering the warehouse on March 1 (historical operation record).

[0038] Preprocessing: Remove duplicate content from the files, unify the temperature and humidity data format to "warehouse number + time + temperature + humidity", and organize the inbound records into "date + variety + quantity".

[0039] Building a knowledge base: The knowledge management module categorizes standards and manuals into the "Grain Storage Technology" and "Operating Procedures" directories, and stores real-time temperature and storage records in the "Dynamic Data" directory, forming a complete knowledge base.

[0040] It should be noted that documents can be further categorized with tags as a search index, such as document titles, to facilitate faster matching of user questions later.

[0041] S102: Train the general large model based on special data in the grain depot field to obtain a special model for grain depots.

[0042] In some embodiments of this application, the general large model refers to a pre-trained language model (e.g., chatglm3-6b) selected as the base model based on the specific needs of grain depot management. This model has undergone extensive pre-training on a general corpus and possesses strong language understanding and generation capabilities.

[0043] Based on this, and to meet the specific needs of grain depot management, a customized training program was developed using grain depot management-specific data to train a general-purpose model on top of the pre-trained model. During training, supervised learning was employed, with cross-entropy loss continuously optimizing model parameters to improve the model's understanding and response capabilities to grain depot-specific knowledge. This resulted in the successful construction of a large-scale language model specifically designed for grain depots.

[0044] It should be noted that the general-purpose large model may not understand the technical terms used in grain depots and needs to be trained to understand them.

[0045] S103: Upon receiving a grain depot problem, load the target document related to the grain depot problem from the knowledge base.

[0046] In some embodiments of this application, unstructured can be used to load collected data from a knowledge base. The unstructured library provides open-source components for extracting and preprocessing images and text documents (such as PDF, HTML, Word documents, etc.), which can decompose the original document into standard structured elements.

[0047] Based on this, semantic parsing is performed on the user's question to determine the knowledge category. The knowledge category is then matched with document tags in the knowledge base to obtain candidate documents. These candidate documents are then loaded using the unstructured library to determine the target document.

[0048] It should be noted that a document can be identified as a candidate document when the semantic similarity is greater than a preset threshold.

[0049] For example, if a user's question is: "What is the maximum storage temperature for wheat in Warehouse 1?", after analysis, it can be found that the knowledge category is "wheat storage temperature limit"; the relevant documents are documents such as "Wheat Storage Temperature Control Specification" and the real-time temperature records of Warehouse 1 in dynamic data.

[0050] It should be noted that when the question involves "wheat storage", documents in unrelated categories such as "corn fumigation operation" and "equipment maintenance records" can be excluded.

[0051] The process involves loading and processing data using the unstructured library, transforming it into structured elements that the system can parse (such as text paragraphs, table data, etc.), which are then identified as the target document for subsequent text segmentation, vectorization, and other steps.

[0052] For example, the chapters and clauses in the "Wheat Storage Temperature Control Specification" (PDF format) were broken down into structured text, and the temperature records (table format) of Warehouse No. 1 were converted into text descriptions.

[0053] S104: Perform text segmentation on the target document to obtain multiple text blocks, and determine the knowledge text vector for each text block.

[0054] In some embodiments of this application, the loaded document is divided into smaller blocks for more efficient information processing and retrieval. A fixed-size block method is used, directly setting the number of words in each block and selecting whether content is repeated between blocks. Compared to other forms of block segmentation (segmented by paragraph, chapter, punctuation, or semantic topic), fixed-size block segmentation is simple to use and requires less computational resources.

[0055] Based on this, firstly, the target document is traversed character by character. Then, according to the parameters of fixed number of characters per block and overlap between blocks, the target document is divided into multiple text blocks with a fixed number of characters.

[0056] The specific process of dividing the target document into multiple text blocks of a fixed number of characters is as follows: When the cumulative word count reaches a fixed number, the first text block is generated.

[0057] The second text block is obtained by reusing the preset number of overlapping characters at the end of the first text block, and then extracting the remaining characters from the end of the first text block. The sum of the remaining characters and the preset number of overlapping characters is a fixed number of characters.

[0058] This process continues until the entire document is segmented, resulting in several consecutive and partially overlapping text blocks of a fixed number of characters.

[0059] This application employs the bge-large-zh model for text vectorization, verifying that the bge-large-zh model performs better on blocks of size 256, ensuring that each block contains sufficient semantic information without causing semantic confusion due to excessive text length. The number of overlapping characters can be 50, meaning there is 50 characters of repeated content between two adjacent text blocks. This avoids semantic breaks caused by abrupt segmentation (e.g., a sentence split into two blocks, with the first half in the first block and the second half in the second block), maintaining contextual coherence through repetition.

[0060] Specifically, the bge-large-zh model is used to convert each word in each text block into a high-dimensional vector representation, thus obtaining the knowledge text vector for each text block.

[0061] It's important to note that text vectorization is the process of converting natural language text into vector representations that can be understood and processed by computers. The bge-large-zh model is a large-scale pre-trained language model based on the Transformer architecture, capable of transforming each word in a text into a high-dimensional vector representation. The core goal of this process is to map each word into a high-dimensional vector space so that machines can understand the semantic relationships between words.

[0062] Furthermore, after vectorizing the text, the generated knowledge text vectors are stored in an efficient vector database. Faiss, an open-source library developed by Facebook AIResearch, is used for efficient similarity search and vector storage to achieve efficient vector storage and retrieval.

[0063] S105: Vectorize the grain depot problem into a text vector to obtain the problem text vector.

[0064] In some embodiments of this application, the grain depot problem is still vectorized into text using the bge-large-zh model.

[0065] S106: Based on the knowledge text vector of each text block and the question text vector, determine the relevant text blocks of the grain depot problem.

[0066] In some embodiments of this application, the process of determining the relevant text blocks of the grain depot problem is as follows: First, calculate the similarity between each knowledge text vector and the question text vector. Then, identify the text blocks corresponding to the top-ranked knowledge text vectors based on their similarity as relevant text blocks.

[0067] One method to identify relevant text blocks is through cosine similarity. This involves calculating the cosine similarity between the question text vector and the knowledge text vectors in the vector database to find the text block that most closely matches the query content. Specifically: First, calculate the Euclidean norm of each knowledge text vector and the Euclidean norm of each question text vector.

[0068] Then, calculate the dot product between each knowledge text vector and the question text vector.

[0069] Then, based on the Euclidean norm of a single knowledge text vector, the Euclidean norm of a question text vector, and the dot product between a single knowledge text vector and a question text vector, the cosine similarity between a single knowledge text vector and a question text vector is obtained.

[0070] The formula for similarity measurement is:

[0071] in, It is the question text vector. It is a knowledge text vector. It is the dot product of vectors. and and are the Euclidean norms of the vectors, respectively. The cosine similarity ranges from [-1, 1], and the closer the value is to 1, the higher the semantic similarity between the two text blocks.

[0072] S107: Input the relevant text block and the grain depot question into the grain depot-specific model to obtain the answer data.

[0073] That is, by using cosine similarity to match the top K most relevant texts in the vector database, and using them as context along with the question as prompts, we input them into a large-scale model specific to grain depots for question answering.

[0074] In some embodiments of this application, a self-learning optimization mechanism is designed to ensure that the system can continuously adapt to changes in grain depot management needs and continuously improve the accuracy and intelligence of question answering. The self-learning optimization is dynamically updated through real-time data and user feedback.

[0075] User feedback-driven model updates: During actual use, the system automatically collects incorrect or unsatisfactory answers through user feedback mechanisms, and aggregates this feedback data for subsequent model training. In this way, the system can gradually improve the accuracy of its answers to specific domain-specific questions while maintaining efficient question answering.

[0076] Based on this, unsatisfactory Q&A results are extracted from user feedback data. These unsatisfactory results are then used to update the grain depot-specific model through retraining.

[0077] Automatic learning from new data: By collecting newly added management documents, monitoring data, and other relevant information from grain depots, the system adds them to the knowledge base. After adding new data, the system periodically triggers a fine-tuning process, integrating the knowledge from the new data into the existing question-and-answer model through incremental learning, enabling it to handle new scenarios and questions. For example, when a grain depot introduces new equipment or updates its operating procedures, the system can quickly learn and adjust its response strategy.

[0078] It should be noted that the automatic learning of new data is not just about updating the knowledge base, but rather a mechanism that combines knowledge base updates with incremental fine-tuning of the large model.

[0079] New data is first entered into the knowledge base for retrieval. At the same time, the system will periodically trigger fine-tuning to transform the new data into training samples and incrementally learn the large language model dedicated to grain depots. This allows the parameters of the large model itself to be updated, enabling it to directly cope with new scenarios and problems.

[0080] This ensures that the intelligent question-answering system can utilize external knowledge bases in real time and quickly absorb new knowledge within the model, achieving more efficient and accurate question-answering capabilities.

[0081] In summary, through a self-learning optimization mechanism, the intelligent question-answering system of this invention can continuously update and evolve to adapt to the diverse needs of grain depot management, and gradually improve the accuracy and intelligence of question-answering in actual operation.

[0082] In other words, through deep integration with warehouse management systems, monitoring systems, etc., this system can acquire dynamic data (monitoring data and management operation data) in the grain depot in real time, and ensure that the intelligent question-and-answer function is always in optimal condition by regularly updating the large model.

[0083] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S101 to S107 will be described sequentially, but this does not mean that steps S101 to S107 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S107 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S107 can be appropriately adjusted according to actual needs.

[0084] pass Figure 1 This method, through deep integration with warehouse management and monitoring systems, acquires real-time monitoring data such as grain depot temperature and humidity, as well as historical operation records. It utilizes vector semantic retrieval from a knowledge base, combined with a large language model, to achieve deep semantic understanding and intelligent question-answering capabilities. Simultaneously, the system has the ability to regularly update the large model, ensuring the accuracy and timeliness of intelligent question answering and analytical decision-making. This optimizes knowledge management processes, enhances system integration and intelligent analysis capabilities, and effectively improves the automation and intelligence of grain depot management. The specific concept is as follows: Deep integration and real-time data acquisition: Through seamless integration with warehouse management and monitoring systems, it can acquire dynamic data such as temperature and humidity from the monitoring system in real time, as well as historical operation records from the warehouse management system, to achieve synchronous updates of static management files and dynamic data, ensuring the accuracy of grain depot management.

[0085] Regularly updated large model: The system can regularly update the large model through incremental learning and data fine-tuning to ensure that it continuously adapts to the latest grain depot management requirements and data updates.

[0086] Intelligent question answering and precise retrieval: Based on vector matching technology, the system can use semantic understanding of natural language questions to filter relevant documents from the knowledge base, and further perform semantic understanding by combining a dedicated large language model to extract precise information and make accurate decisions on the answer. It also supports multi-turn dialogue and automated data analysis.

[0087] Based on this, a grain depot knowledge base integrating three types of core data is constructed to uniformly classify and store scattered information, and achieves dynamic updates through real-time integration with monitoring and storage management systems. Users can obtain comprehensive information without switching between multiple systems (e.g., when querying wheat storage standards, they can simultaneously access the corresponding operation manual, historical storage records, and real-time warehouse temperature and humidity). Furthermore, monitoring data (such as sudden changes in warehouse temperature) can be instantly incorporated into the knowledge base, ensuring that subsequent queries are based on the latest information, thus solving the problem of data lag in traditional systems.

[0088] By training a general-purpose large model with grain depot-specific data, the model's ability to understand professional scenarios is enhanced, thereby improving the accuracy of its responses.

[0089] By dividing the loaded document into smaller blocks for more efficient information processing and retrieval, and through a process of text segmentation → vectorization → vector matching, a rapid and accurate initial retrieval of semantic-level knowledge content related to the user's question is achieved.

[0090] By inputting relevant text blocks and grain depot questions into a dedicated grain depot model, in-depth semantic understanding and logical decision analysis can be performed on the relevant text blocks and grain depot questions through a dedicated large model, supporting multi-turn dialogue and complex decision-making, and outputting accurate question-and-answer data.

[0091] For example, real-time response to dynamic questions: If a user asks: Is the current humidity in Warehouse 3 too high? The system can load the latest monitoring data from the knowledge base (such as humidity 72%, standard ≤70%) and directly answer: It has exceeded the standard. It is recommended to start the dehumidification equipment.

[0092] Supports multi-round dialogue and complex decision-making: For example, if a user asks: How long after dehumidification should we check again? The system can combine the "Humidity Control Specification" and historical operation records to answer: Check once every 2 hours until it drops below 65%, forming a closed-loop decision support.

[0093] In summary, by optimizing the entire chain of knowledge integration, model specialization, semantic retrieval, and intelligent interaction, the problems of "dispersed knowledge, inefficient retrieval, inaccurate answers, and reliance on manual decision-making" in traditional grain depot management have been fundamentally solved. This has enabled the systematization of knowledge management, the vector semanticization of preliminary knowledge retrieval, the rapid semantic extraction of large language models, in-depth semantic understanding, professional answers, and intelligent decision-making, significantly improving the efficiency and accuracy of grain depot management.

[0094] More intuitively, Figure 2 This is a schematic diagram of the architecture of a grain depot intelligent question-and-answer system provided in an embodiment of this application.

[0095] exist Figure 2 The document demonstrates the user's intelligent question-and-answer processing flow, based on the knowledge management module and the intelligent question-and-answer module.

[0096] Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0097] Figure 3 A schematic diagram of the structure of a smart question-and-answer device for grain depots provided in this application embodiment includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the above-described intelligent question-and-answer method for grain depots.

[0098] Some embodiments of this application provide a non-volatile computer storage medium for intelligent question-and-answer in grain depots, which stores computer-executable instructions capable of executing any of the above-described intelligent question-and-answer methods for grain depots.

[0099] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0100] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0101] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0105] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0106] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0107] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0108] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0109] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the technical principles of this application should fall within the protection scope of this application.

Claims

1. A smart question-and-answer method for grain depots, characterized in that, The method includes: Construct a grain depot knowledge base; the documents in the grain depot knowledge base include grain depot management documents, historical operation records, and grain depot monitoring data; A grain depot-specific model is obtained by training a general large model using specialized data from the grain depot sector. Upon receiving a grain depot issue, load the target document related to the grain depot issue from the knowledge base; The target document is segmented into multiple text blocks, and the knowledge text vector of each text block is determined. The grain depot problem is vectorized into a text vector to obtain the problem text vector; Based on the knowledge text vector of each text block and the question text vector, the relevant text blocks for the grain depot problem are determined; The relevant text blocks and the grain depot question are input into the grain depot-specific model to obtain the answer data.

2. The method according to claim 1, characterized in that, The target document is segmented into multiple text blocks, specifically including: Perform a word-by-word traversal of the target document; Based on the parameters of fixed word count per block and overlap between blocks, the target document is divided into multiple text blocks with fixed word counts.

3. The method according to claim 2, characterized in that, Based on the parameters of fixed word count per block and overlap between blocks, the target document is divided into multiple text blocks with fixed word counts, specifically including: When the cumulative word count reaches a fixed number, the first text block is generated; The second text block is obtained by reusing the preset number of overlapping characters at the end of the first text block, and then extracting the remaining characters from the end of the first text block; the sum of the remaining characters and the preset number of overlapping characters is a fixed number of characters. This process continues until the entire document is segmented, resulting in several consecutive and partially overlapping text blocks of a fixed number of characters.

4. The method according to claim 1, characterized in that, The step of determining the relevant text blocks for the grain depot issue based on the knowledge text vector of each text block and the question text vector specifically includes: Calculate the similarity between each knowledge text vector and the question text vector; The text blocks corresponding to the top-ranked knowledge text vectors with a predetermined number of similarities are identified as relevant text blocks.

5. The method according to claim 4, characterized in that, The calculation of the similarity between each knowledge text vector and the question text vector specifically includes: Calculate the Euclidean norm of each knowledge text vector and the Euclidean norm of each question text vector; Calculate the dot product between each knowledge text vector and the question text vector; The cosine similarity between a single knowledge text vector and a question text vector is obtained by using the Euclidean norm of a single knowledge text vector, the Euclidean norm of a question text vector, and the dot product between a single knowledge text vector and a question text vector.

6. The method according to claim 1, characterized in that, Determine the knowledge text vector for each text block, specifically including: The bge-large-zh model converts each word in each text block into a high-dimensional vector representation, thus obtaining the knowledge text vector for each text block.

7. The method according to claim 1, characterized in that, After inputting the relevant text blocks and the grain depot question into the grain depot-specific model to obtain the response data, the method further includes: Extract unsatisfactory Q&A results from user feedback data; Based on the unsatisfactory question-and-answer results, the grain depot-specific model is updated and trained to update the grain depot-specific model.

8. The method according to claim 1, characterized in that, The loading of target documents related to the grain depot issue from the knowledge base specifically includes: Perform semantic analysis on user questions to determine knowledge categories; The knowledge categories are matched with document tags in the knowledge base to obtain candidate documents; The candidate documents are loaded using the unstructured library to determine the target document.

9. A smart question-and-answer device for grain depots, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the intelligent question-and-answer method for grain depots as described in any one of claims 1-8.

10. A non-volatile computer storage medium for intelligent question-and-answer systems in grain depots, storing computer-executable instructions, characterized in that... The computer-executable instructions are capable of executing the intelligent question-and-answer method for grain depots as described in any one of claims 1-8.

Images