A plant science knowledge graph construction method and a plant science knowledge question answering system

Abstract

The present application relates to the field of artificial intelligence and bioinformatics, and discloses a construction method of a plant science knowledge graph and a plant science knowledge question-answering system, wherein the construction method comprises: obtaining plant science data and preprocessing; automatically extracting biological entities and their relationships from the preprocessed data using a large language model; mapping the extraction results to a structured knowledge graph framework to form a plant science knowledge graph. The question-answering system is based on the knowledge graph, identifies key biological concepts in user queries, performs multi-stage subgraph retrieval in the knowledge graph to construct a knowledge chain with traceable information, and uses the "knowledge chain" to enhance the question-answering large language model to generate professional, accurate and verifiable answers. The present application can realize the full automation and structuring of field knowledge from construction to application, effectively solving the problems of inaccurate knowledge, lagging updates and untraceability of general large language models in professional fields.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and bioinformatics, and in particular to a method for constructing a plant science knowledge graph and a plant science knowledge question-and-answer system. Background Technology

[0002] In recent years, Large Language Models (LLMs) have performed well in general natural language processing tasks. However, their application in highly specialized fields such as plant science faces significant challenges. For example, LLMs may generate scientifically inaccurate or unverifiable information in specialized literature, lacking a deep understanding of complex biological entity relationships; the training data for these models has deadlines, failing to cover the latest scientific research findings and data; the models struggle to utilize structured domain knowledge and cannot provide the original literature or data sources upon which their arguments are based, resulting in low credibility; plant science data sources are diverse, including PDF documents and database records, with varied formats, making automated integration and knowledge extraction costly; and the reliance on environment-specific processing scripts and workflows makes one-click deployment and result reproduction difficult across different computing environments. Existing solutions largely rely on manually curated knowledge bases or single database retrieval, which are costly to build and maintain, slow to update, and difficult to achieve deep semantic association and reasoning.

[0003] Therefore, there is an urgent need for a professional domain-specific intelligent question-answering solution that can automatically integrate and update multi-source scientific knowledge and ensure information accuracy and traceability. Summary of the Invention

[0004] This invention provides a method for constructing a plant science knowledge graph and a plant science knowledge question-and-answer system to overcome the shortcomings of existing technologies.

[0005] This invention provides a method for constructing a plant science knowledge graph, comprising:

[0006] S1. Acquire plant science data, including unstructured data; S2. Preprocessing plant science data; S3. Extract biological entities and entity relationships from the preprocessed plant science data; S4. The extracted biological entities and entity relationships are mapped to a structured knowledge graph framework to obtain a plant science knowledge graph.

[0007] According to the method for constructing a plant science knowledge graph provided by the present invention, the preprocessing includes any one or any combination of the following: Optical character recognition (OCR) was used to parse non-text files in plant science data to obtain the original text corpus. Assign a unique identifier to each file in plant science data; The text data in plant science data is processed using a preprocessing workflow, which includes standardization encoding, correction of special characters, and filtering of line breaks. Convert English text in plant science data to lowercase.

[0008] According to the method for constructing a plant science knowledge graph provided by the present invention, step S3 includes: The first language model is used to extract text excerpts that meet the preset conditions from preprocessed plant science data, and named entity recognition task is performed on each text excerpt to identify biological entities and entity relationships in the text excerpts. In the process of identifying biological entities and entity relationships, contextual learning is applied, and expert-designed examples are added to the prompts to avoid illusions or entity omissions.

[0009] According to the method for constructing a plant science knowledge graph provided by the present invention, the preset conditions include setting the character length and / or the character overlap with continuous windows.

[0010] According to the method for constructing a plant science knowledge graph provided by the present invention, step S3 further includes: Using the second language model, combined with prompting, we examine the biological entities and entity relationships identified by the first language model.

[0011] According to the method for constructing a plant science knowledge graph provided by the present invention, step S4 includes: Biological entities are represented by nodes in a knowledge graph framework, and entity relationships are represented by edges in the knowledge graph framework. Standardize and merge synonyms and spelling variations; By summarizing all relevant text fragments describing biological entities and entity relationships collected during the extraction process, a final description is generated for each node; Assign a category to each node from a predefined list.

[0012] This invention provides a plant science knowledge question-and-answer system, comprising: The data receiving module is used to receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by the method for constructing a plant science knowledge graph according to any one of claims 1-6 based on the input data of the user query, and provide the response data to the question-answering big language model as data support for responding to the user query.

[0013] This invention provides a plant science knowledge question-and-answer system, the retrieval module of which includes: The entity recognition submodule is used to: perform named entity recognition on the user-queried input data through the first major language model in the plant science knowledge graph construction method described above, so as to identify key biological concepts in the input data; The node retrieval submodule is used to: search for the node most relevant to the key biological concept in the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods, based on the key biological concept in the input data. The response submodule is used to provide the final description of the node most relevant to the key biological concept to the question-answering language model as data support for responding to user queries.

[0014] According to the present invention, a plant science knowledge question-answering system is provided, wherein the step of searching for the most relevant node to the key biological concept in the plant science knowledge graph obtained by the above-mentioned plant science knowledge graph construction method, based on the key biological concept in the input data, includes: Biological entities identified through the named entity recognition task and their direct neighbors in the plant science knowledge graph are retrieved to form an initial query-related subgraph. By leveraging embedding-based semantic similarity, we can identify and retrieve semantically similar entities that are not directly connected but are conceptually related in a plant science knowledge graph. By traversing the k-hop neighborhoods of biological entities and semantically similar entities, the Dijkstra algorithm is applied to enumerate the shortest paths between related node pairs. The biological entities, entity relationships, paths, and associated text descriptions and metadata obtained from the above multi-stage retrieval are aggregated to form a "knowledge chain" with comprehensive traceability information. This "knowledge chain" is then used as rich contextual information and input into the large language model, enabling the large language model to generate user query responses that are evidence-based, verifiable, and fully traceable.

[0015] The present invention provides a plant science knowledge question-and-answer system, which further includes: The self-learning module is used to update the plant science knowledge graph based on the user's response to the question-and-answer language model and / or the information searched by the system in the public network based on the user's query input data.

[0016] The present invention also provides an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the method for constructing a plant science knowledge graph as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a plant science knowledge graph as described above.

[0018] The present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute any of the above-described methods for constructing a plant science knowledge graph.

[0019] The method for constructing a plant science knowledge graph and the plant science knowledge question-and-answer system provided by this invention can bring at least the following beneficial effects: By driving the AutoSKG process with a large language model, the end-to-end automated construction and updating of unstructured documents into structured knowledge graphs is achieved, which greatly improves knowledge processing efficiency and significantly reduces labor costs.

[0020] The professional knowledge graph built on domain data and its multi-stage retrieval strategy provide accurate and structured deep domain knowledge for the question answering system, ensuring the professionalism and scientific nature of the answers.

[0021] Through an incremental update mechanism based on user queries and uploads, the plant science knowledge graph can continuously absorb the latest scientific research results, becoming a "living" knowledge system that can keep pace with the times.

[0022] The built-in "knowledge chain" construction mechanism closely links the content generated by the system with the original evidence in the plant science knowledge graph, providing a clear traceability path and greatly enhancing the credibility and scientific rigor of the answers.

[0023] By encapsulating the system through modern software engineering practices such as containerization, the problems of environment dependency and deployment complexity are effectively solved, giving the system high portability, reproducibility and cloud-native deployment capabilities, making it easy to promote and apply on a large scale. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a method for constructing a plant science knowledge graph provided by the present invention.

[0026] Figure 2This is a schematic diagram of the structure of a plant science knowledge question-and-answer system provided by the present invention.

[0027] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] Figure 1 This is a flowchart illustrating a method for constructing a plant science knowledge graph provided by the present invention. The execution entity of this method can be any applicable terminal-side device or network-side device, such as a plant science knowledge graph construction apparatus.

[0030] See Figure 1 The present invention provides a method for constructing a plant science knowledge graph, which may include: S1. Obtain plant science data, including unstructured data.

[0031] S2. Preprocess plant science data.

[0032] In one embodiment, preprocessing includes any one or any combination of the following: Optical character recognition (OCR) is used to parse non-text files (such as PDFs) in plant science data to obtain the raw text corpus. Assign a unique identifier (e.g., hash ID) to each file in the plant science data for later backtracking; The text data in plant science data is processed using a preprocessing workflow, which includes standardization encoding, correction of special characters, and filtering of line breaks. Convert English text in plant science data to lowercase English to improve the processing efficiency of large language models.

[0033] S3. Extract biological entities and entity relationships from the preprocessed plant science data.

[0034] In one embodiment, step S3 includes: The first language model is used to obtain text extracts that meet preset conditions from preprocessed plant science data, and named entity recognition task is performed on each text extract to identify biological entities and entity relationships in the text extract. The preset conditions include the setting of character length and / or the degree of character overlap with continuous windows. In the process of identifying biological entities and entity relationships, contextual learning is applied, and expert-designed examples are added to the prompts to avoid illusions or entity omissions. Using the second language model and combining it with a prompting method, we examine the biological entities and entity relationships identified by the first language model. In order to extract biological entities and relationships, this embodiment prompts a large language model to read through the entire text corpus.

[0035] This embodiment uses GPT-4o-mini as the first large language model to maintain a good balance between performance and cost. Each time, the first large language model obtains a text extract of 1500 characters with 200 characters overlapping with consecutive windows. The first large language model is instructed to identify any biological entities and cite the context in which these entities appear, essentially performing a named entity recognition task. To improve extraction quality and avoid illusions or missed entities, this embodiment applies contextual learning, incorporating five expert-designed examples into the prompts. This embodiment also performs multiple rounds of careful checking, accompanied by subsequent critical analysis questions (e.g., "Does the current extraction cover all biological entities that appear in the original text?", "Does the current extraction contain any biological entities that are not actually present in the original text?") to ensure comprehensive coverage while avoiding illusions. After extracting all entities from the entire text corpus, this embodiment performs another round of large language model careful checking, using a similar prompting method to identify any explicitly stated biological relationships between previously extracted entities and recording the context involving such statements.

[0036] S4. The extracted biological entities and entity relationships are mapped to a structured knowledge graph framework to obtain a plant science knowledge graph.

[0037] In one embodiment, step S4 includes: Biological entities are represented by nodes in a knowledge graph framework, and entity relationships are represented by edges in the knowledge graph framework. Standardize and merge synonyms and spelling variations; By summarizing all relevant text fragments describing biological entities and entity relationships collected during the extraction process, a final description is generated for each node; Assign a category to each node from a predefined list.

[0038] After extracting all biological entities and entity relationships, this implementation maps all results to a structured knowledge graph, where nodes are the extracted biological entities and edges represent the stated relationships between them. During this process, synonyms and spelling variations are standardized and merged to avoid redundancy. This implementation also generates a final description for each node and entity by summarizing all relevant text fragments collected during the extraction process. A large language model, GPT-4o, is used to assign a category to each node from a predefined list. During the integration of the plant knowledge graph, new entities and relationships are directly incorporated into the knowledge graph, while pre-existing entities and relationships are coordinated using an automated structured knowledge graph process to integrate complementary information from both sources.

[0039] The Plant Science Knowledge Graph is constructed and continuously maintained through the automated AutoSKG workflow (steps S1-S4) of this embodiment. The initial version of the Plant Science Knowledge Graph focuses on model organisms such as Arabidopsis thaliana and key crops including wheat, rice, and maize, supplemented by structured, community-contributed data from researchers at the John Innes Centre and Sainsbury's Laboratory. These initial datasets were manually curated by plant experts before being processed by AutoSKG to ensure a high-quality foundation.

[0040] Since then, the Plant Science Knowledge Graph has been significantly expanded to include tens of thousands of full-text articles from various peer-reviewed journals in the field of plant science. This embodiment continuously expands its data sources to cover literature across a broader phylogenetic range. Furthermore, the Plant Science Knowledge Graph integrates the PlantConnectome database to enrich the knowledge base with additional structured biological relationships.

[0041] The Plant Science Knowledge Graph maintains its up-to-date state through a hybrid dual-mode update process, enabling continuous learning and knowledge expansion. First, when a user submits a query, the system performs a web search on the same topic to identify the latest publications not yet integrated into the Plant Science Knowledge Graph. These newly discovered studies are automatically processed and incorporated into the knowledge graph via the AutoSKG workflow. Second, users can directly upload their own research findings or domain-specific knowledge via (https: / / plantscience.ai / upload), which are also integrated into the Plant Science Knowledge Graph. To ensure computational efficiency, all newly acquired publications from web searches and user-submitted data are processed in a weekly collective batch update.

[0042] The plant science knowledge question-and-answer system provided by this invention will be described below. The plant science knowledge question-and-answer system described below can be referred to in correspondence with the plant science knowledge graph construction method described above.

[0043] The present invention provides a plant science knowledge question-and-answer system, which may include: The data receiving module is used to receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods based on the user's input data, and provide the response data to the question-answering big language model as data support for responding to the user's query.

[0044] Regarding the terminology recognition and disambiguation module, it can use a language model combined with a plant science terminology dictionary (initial database) to identify key biological concepts in queries and accurately resolve ambiguous abbreviations. 1. For single-ambiguous abbreviations: When a user queries "The role of ABA in drought resistance in Arabidopsis thaliana?", the system uses contextual information such as "Arabidopsis thaliana" and "drought resistance" plus a terminology database for double verification, eliminating irrelevant interpretations such as "application behavior analysis", and confirms that "ABA" = abscisic acid. 2. For combined abbreviation queries: When a user queries "Is FT the main target of CO?", the system automatically matches the terminology database entries and resolves it as "FT" = flowering gene (FLOWERING LOCUS T) and "CO" = CONSTANS transcription factor, rather than common definitions such as "foot (FT)" or "carbon monoxide (CO)". 3. Terminology Ambiguity Resolution: Based on a dual mechanism of "contextual semantics + terminology database priority", for example, when "FT" appears in the query but the field is not clearly defined, the system prioritizes matching high-frequency abbreviations in plant science (FLOWERING LOCUS T) and verifies it through knowledge graph retrieval (such as the regulatory relationship between FT and CO exists in the flowering pathway of Arabidopsis thaliana) to further confirm the accuracy of the interpretation.

[0045] The initial library contains 1200+ core plant science terms and abbreviations, with a focus on covering ambiguous abbreviations (such as "ABA = Abscisic Acid", "FT = Flowering Locus T", and "CO = CONSTANS Transcription Factor"). Ambiguity resolution rules: For cross-domain ambiguous abbreviations such as "ABA", the system determines the meaning through "contextual keyword matching (such as 'drought resistance' 'plant hormone') + domain probability model"; for intra-domain combination abbreviations such as "FT-CO", the system accurately resolves them through "abbreviation co-occurrence relation library" (such as FT and CO only appearing together in the flowering regulation pathway).

[0046] In one embodiment, the retrieval module includes: The entity recognition submodule is used to: perform named entity recognition on the user-queried input data through the first major language model in the plant science knowledge graph construction method described above, so as to identify key biological concepts in the input data; The node retrieval submodule is used to: search for the node most relevant to the key biological concept in the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods, based on the key biological concept in the input data. The response submodule is used to provide the final description of the node most relevant to the key biological concept to the question-answering language model as data support for responding to user queries.

[0047] In one embodiment, the step of searching for the node most relevant to the key biological concept in the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods, based on the key biological concept in the input data, includes: Biological entities identified through the named entity recognition task and their direct neighbors in the plant science knowledge graph are retrieved to form an initial query-related subgraph. By leveraging embedding-based semantic similarity, we can identify and retrieve semantically similar entities that are not directly connected but are conceptually related in a plant science knowledge graph. By traversing the k-hop neighborhoods of biological entities and semantically similar entities, the shortest path between related node pairs is enumerated. The biological entities, entity relationships, paths, and their associated text descriptions and metadata obtained from the multi-stage retrieval process are aggregated to form a knowledge chain, which is then used as contextual information to input into the question-answering big language model.

[0048] Regarding the search module, taking the user query "Is FT a major target of CO?" as an example, the search process is as follows: 1. Initial subgraph extraction: parse the key entities “FT (FLOWERING LOCUS T)” and “CO (CONSTANS)”, and retrieve their direct neighbors in the PSKG (such as the interacting gene SOC1 of FT and the downstream gene FT of CO). 2. Semantic similarity extension: Based on embedded vector retrieval, associate related entities such as "Arabidopsis flowering pathway" and "photoperiod regulation"; 3. Path enumeration: Traverse the 2-hop neighborhood and obtain the shortest regulatory path using Dijkstra's algorithm: "CO (transcription factor) → binds to FT gene promoter → regulates FT expression"; 4. Knowledge Chain Aggregation: Integrate path information and source evidence to form a structured knowledge chain.

[0049] In one embodiment, the plant science knowledge question-and-answer system further includes: The self-learning module is used to update the plant science knowledge graph based on the user's response to the question-and-answer language model and / or the information searched by the system in the public network based on the user's query input data.

[0050] Specifically, when a user submits a query, the system performs a web search on the same topic to identify the latest publications not yet integrated into the PSKG. These newly discovered studies are automatically processed through the AutoSKG pipeline and integrated into the knowledge graph. User contributions and batch updates: Users can directly upload their research findings or domain-specific knowledge through the PlantScience.ai platform. This user-submitted data is also integrated into the PSKG through the AutoSKG pipeline.

[0051] In one embodiment, the plant science knowledge question-and-answer system further includes: Data Reception and Privacy Protection Module: This module processes all user data only in a local environment to prevent data leakage. In one embodiment, for each user input, this embodiment first performs named entity recognition using the GPT-4o large language model to identify the key biological concepts involved. Then, it searches for relevant nodes in the plant science knowledge graph. To this end, this embodiment creates embedding vectors for all nodes in the knowledge graph and the identified biological concepts in a latent semantic space, where semantically similar concepts are close to each other. Here, this embodiment uses Voyage AI's voyage-3-large model, which demonstrates state-of-the-art performance on a wide range of text embedding benchmarks, outperforming similar models (such as OpenAI-v3-large) by an average of approximately 9.7% (Voyage AI, 2025). After constructing the embedding vectors, this embodiment searches the knowledge graph for nodes whose embedding vectors have the highest cosine similarity to the key concepts in the user input.

[0052] Then, this embodiment constructs a "knowledge chain" through a graph-based continuous incremental update module, which involves multi-stage subgraph retrieval and expansion. First, seed entities identified through NER and their direct neighbors are retrieved from the PSKG to form an initial query-related subgraph. Second, this embodiment utilizes embedding-based semantic similarity to identify and retrieve semantically similar entities, effectively capturing nodes that may not have direct connections in the graph topology but are conceptually related. Third, this embodiment performs neighborhood expansion by traversing the k-hop neighborhoods of seed entities and semantically similar entities, applying Dijkstra's algorithm to enumerate the shortest path between each pair of related nodes. This multi-stage graph RAG process ensures comprehensive retrieval of explicit structural connections and implicit semantic associations within the knowledge graph. Each generated knowledge chain contains a series of intermediate entities and typified semantic relations that establish logical connections between related biological concepts. Aggregated text descriptions and metadata associated with each entity and edge in the knowledge chain provide comprehensive source information. This embodiment then feeds this rich contextual knowledge into the LLM, enabling it to generate evidence-based, verifiable responses that directly address user queries while maintaining full traceability to the underlying data source.

[0053] This embodiment is based on the plant science knowledge graph and uses PlantScience.ai for full-stack development.

[0054] The PlantScience.ai web interface is developed as a comprehensive web-based application, leveraging modern front-end technologies to provide an intelligent research assistant for plant science. The front-end architecture utilizes Vue.js 3 and its composable API (vuejs.org). For front-end and back-end communication, this embodiment integrates Axios as an HTTP client (axios-http.com) to enable real-time streaming responses and continuous parsing of incoming data, crucial for the progressive delivery of answers generated by PlantScience.ai. The chat interface supports context-aware dialogue, with responses rendered via a Markdown rendering stack, allowing for rich formatting of scientific symbols, citations, and structured information. Interactive knowledge graph visualization is powered by ECharts (echarts.apache.org), enabling users to dynamically explore biological relationships. Prioritizing user privacy, this embodiment uses Pinia (pinia.vuejs.org) for state management, storing all conversation history and user data on the client side rather than transmitting it to the server. The application supports internationalization for multiple languages ​​and integrates a user authentication system with token-based secure authorization. The development methodology follows modern web standards and incorporates progressive web application features to ensure cross-platform compatibility across desktop and mobile devices.

[0055] The PlantScience.ai server is developed as a robust API server, leveraging modern backend technologies. The system architecture uses "FastAPI" (fastapi.tiangolo.com) as the primary framework, utilizing its asynchronous request processing to provide efficient real-time streaming responses. Knowledge extracted by AutoSKG is stored in the "Neo4j Graph Database," which supports comprehensive operations on nodes and relationships (create, read, update, delete), enabling automated knowledge updates and maintenance. To facilitate intelligent information retrieval, this embodiment implements a hybrid search system combining exact matching, fuzzy matching, and semantic matching. This embodiment also embeds an intent detection algorithm to determine whether a query requires access to the knowledge graph. Session management handles conversation data storage, automatic expiration, and temporary caching of graph data to optimize performance. The authentication system uses JWT tokens with bcrypt password hashes, supporting user registration with email verification, login authentication, password reset functionality, and usage restrictions (chat quotas and feedback counts). An integrated email service manages verification emails, password resets, and feedback notifications. Performance optimizations include batch queries, data compression (removing large fields, such as embedded vectors), timeout control, and entity indexes to ensure a responsive user experience even under complex queries.

[0056] The present invention provides a method for constructing a plant science knowledge graph and a plant science knowledge question-and-answer system, which have the following significant advantages: This invention achieves end-to-end automated construction of structured knowledge graphs from unstructured multi-source data (such as PDF documents) through the AutoSKG process based on a large language model. This completely changes the high-cost, low-efficiency model that relied on manual compilation of knowledge bases in the past, and greatly reduces the threshold and human cost of building and maintaining domain knowledge graphs.

[0057] By constructing a specialized plant science knowledge graph and providing a "knowledge chain" rich in entity relationships generated through multi-stage graph retrieval as context for the large language model, this invention ensures that the generated answers are rooted in structured domain knowledge. This effectively overcomes the problems of knowledge illusion and superficial understanding in specialized fields inherent in general-purpose large language models, significantly improving the scientific accuracy and professional depth of the answers.

[0058] This invention combines knowledge graphs with user queries and web searches to establish a continuous incremental update mechanism. This allows the knowledge graph to continuously absorb the latest scientific research results and user contributions, becoming a "living" knowledge system that can keep pace with the times, thus solving the core pain point of lagging knowledge updates in traditional models and static knowledge bases.

[0059] Through the "knowledge chain" mechanism, every key fact in the system's generated answer can be traced back to a specific node, relationship, and original textual evidence in the knowledge graph. This embedded tracing capability provides users with a clear path to verify information, greatly enhancing the credibility of the answer and the scientific rigor of the system—something that general-purpose large language models and traditional retrieval systems cannot match.

[0060] This invention systematically integrates and encapsulates the entire technical solution, including modules for data processing, map construction, retrieval, and generation. Through clearly defined methodological steps and a system architecture, it solves the deployment and reproducibility difficulties caused by reliance on specific environment scripts, enabling the system to possess high portability, reproducibility, and cloud-native deployment capabilities, laying the foundation for large-scale practical applications.

[0061] The Plant Science Knowledge Question-Answering System (PSKG) utilizes an automated incremental update mechanism to make PSKG a "living knowledge system," absorbing the latest research findings and user data in real time to address the problem of knowledge lag. Through local private deployment, encrypted storage, and access control, it ensures that sensitive information such as core breeding data and unpublished experimental results are not leaked. A triple mechanism of "terminology database + context + co-occurrence relationship" efficiently resolves ambiguous abbreviation issues (such as distinguishing between "ABA = abscisic acid" and "ABA = applied behavior analysis," and interpreting "FT-CO" as flowering regulation-related genes / transcription factors), completely eliminating semantic ambiguity in general LLMs and ensuring the accuracy of professional queries. It supports user-defined terminology databases and local hardware upgrades, making it compatible with various plant science research scenarios.

[0062] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the steps of any of the above-described methods for constructing a plant science knowledge graph, and / or, the following steps: Receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods based on the user's input data, and provide the response data to the question-answering big language model as data support for responding to the user's query.

[0063] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is capable of performing the steps of any of the above-described methods for constructing a plant science knowledge graph, and / or, the following steps: Receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods based on the user's input data, and provide the response data to the question-answering big language model as data support for responding to the user's query.

[0065] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for constructing a plant science knowledge graph, and / or the following steps: Receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by any of the above-described plant science knowledge graph construction methods based on the user's input data, and provide the response data to the question-answering big language model as data support for responding to the user's query.

[0066] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0067] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a plant science knowledge graph, characterized in that, include: S1. Acquire plant science data, including unstructured data; S2. Preprocessing plant science data; S3. Extract biological entities and entity relationships from the preprocessed plant science data; S4. The extracted biological entities and entity relationships are mapped to a structured knowledge graph framework to obtain a plant science knowledge graph.

2. The method for constructing a plant science knowledge graph according to claim 1, characterized in that, Preprocessing includes any one or any combination of the following: Optical character recognition (OCR) was used to parse non-text files in plant science data to obtain the original text corpus. Assign a unique identifier to each file in plant science data; The text data in plant science data is processed using a preprocessing workflow, which includes standardization encoding, correction of special characters, and filtering of line breaks. Convert English text in plant science data to lowercase.

3. The method for constructing a plant science knowledge graph according to claim 1, characterized in that, Step S3 includes: The first language model is used to extract text excerpts that meet the preset conditions from preprocessed plant science data, and named entity recognition task is performed on each text excerpt to identify biological entities and entity relationships in the text excerpts. In the process of identifying biological entities and entity relationships, contextual learning is applied, and expert-designed examples are added to the prompts to avoid illusions or entity omissions.

4. The method for constructing a plant science knowledge graph according to claim 3, characterized in that, Step S3 also includes: Using the second language model, combined with prompting, we examine the biological entities and entity relationships identified by the first language model.

5. The method for constructing a plant science knowledge graph according to claim 4, characterized in that, Step S4 includes: Biological entities are represented by nodes in a knowledge graph framework, and entity relationships are represented by edges in the knowledge graph framework. Standardize and merge synonyms and spelling variations; By summarizing all relevant text fragments describing biological entities and entity relationships collected during the extraction process, a final description is generated for each node; Assign a category to each node from a predefined list.

6. The method for constructing a plant science knowledge graph according to claim 3, characterized in that, Preset conditions include character length and / or character overlap with consecutive windows.

7. A plant science knowledge question-and-answer system, characterized in that, include: The data receiving module is used to receive user query input data; The terminology recognition and disambiguation module is used to: recognize and disambiguate terminology in user query input data; The retrieval module is used to: retrieve response data from the plant science knowledge graph obtained by the method for constructing a plant science knowledge graph according to any one of claims 1-6 based on the input data of the user query, and provide the response data to the question-answering big language model as data support for responding to the user query.

8. The plant science knowledge question-and-answer system according to claim 7, characterized in that, The search module includes: The entity recognition submodule is used to: perform named entity recognition on the input data queried by the user through the first large language model in the method for constructing a plant science knowledge graph according to any one of claims 1-6, so as to identify key biological concepts in the input data; The node retrieval submodule is used to: search for the node most relevant to the key biological concept in the plant science knowledge graph obtained by the method for constructing a plant science knowledge graph according to any one of claims 1-6, based on the key biological concept in the input data; The response submodule is used to provide the final description of the node most relevant to the key biological concept to the question-answering language model as data support for responding to user queries.

9. The plant science knowledge question-and-answer system according to claim 8, characterized in that, The step of searching for the most relevant nodes to the key biological concepts in the plant science knowledge graph obtained by the method for constructing a plant science knowledge graph according to any one of claims 1-6, based on the key biological concepts in the input data, includes: Biological entities identified through the named entity recognition task and their direct neighbors in the plant science knowledge graph are retrieved to form an initial query-related subgraph. By leveraging embedding-based semantic similarity, we can identify and retrieve semantically similar entities that are not directly connected but are conceptually related in a plant science knowledge graph. By traversing the k-hop neighborhoods of biological entities and semantically similar entities, the shortest path between related node pairs is enumerated. The biological entities, entity relationships, paths, and their associated text descriptions and metadata obtained from the multi-stage retrieval process are aggregated to form a knowledge chain, which is then used as contextual information to input into the question-answering big language model.

10. The plant science knowledge question-and-answer system according to claim 7, characterized in that, Also includes: The self-learning module is used to update the plant science knowledge graph based on the user's response to the question-and-answer language model and / or the information searched by the system in the public network based on the user's query input data.

Images