Seed industry information question answering system and method

Abstract

The invention discloses a seed information question-answering system and method. The invention provides a variety information question-answering system. The system comprises a computing resource; the big language model calls computing resources to perform seed information question answering, the big language model is configured to learn knowledge related to the seed from a seed field knowledge source, and the big language model comprises an input acquisition module configured to be used for acquiring text information related to seed information questions from user input; the semantic understanding module is coupled to the input acquisition module and is configured to be used for converting the text information into vector information; the information reasoning module is coupled to the semantic understanding module and is configured to be used for generating an answer information vector based on knowledge related to the seed industry in combination with the vector information and the context, and the answer information vector comprises breeding decision information; and the result output module is coupled to the information reasoning module and is configured to be used for converting the answer information vector into answer text information.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and specifically to a seed industry information question-and-answer system, method, and device. Background Art

[0002] Breeding is one of the core contents in agricultural science. It usually requires a comprehensive analysis of core issues such as variety selection, agronomic traits, cultivation techniques and promotion areas. On this basis, variety selection programs are designed and new varieties are created to improve crop yield and quality. Traditional breeding decisions rely on the experience and knowledge accumulation of experts, and the process is cumbersome and time-consuming. In particular, in the face of increasingly complex agricultural environments and diversified breeding needs, it is difficult to quickly and comprehensively obtain and process a large amount of breeding information by manual labor alone.

[0003] With the rapid development of artificial intelligence (AI) technology, especially large language models, in recent years, traditional social production and lifestyles have also been greatly impacted. Large language models (LLMs) generally refer to models that use deep learning methods to process and generate natural language text by training huge neural network models. In large language models, knowledge bases are usually used as background knowledge, so that large language models can use the information in the knowledge base to perform natural language processing tasks such as text generation, machine translation, and intelligent question and answer.

[0004] The rise of large language models has brought hope to the processing of breeding information. By utilizing the powerful language understanding and generation capabilities of large language models, as well as further deep reasoning capabilities, intelligent processing of breeding information can be achieved, providing strong support for breeding decisions.

[0005] However, on the one hand, most of the current large language models are general artificial intelligence products that do not have the ability to answer questions related to the seed industry, and their help to my country's biological breeding industry is limited. At present, there are no academic articles or patent documents related to the realization of seed industry information question and answer based on large language models. In particular, the current question and answer system based on large language models usually uses a single corpus knowledge for training, which cannot adapt to the multimodal large language model training in biological breeding.

[0006] On the other hand, the relevant experience and knowledge of biological breeding are often scattered in a large number of scientific documents and books in text form. The amount of data is huge and complex, and information in different dimensions such as genome, agronomic shape, and environmental adaptability are isolated from each other, making it difficult to form a data set that can be applied to large language models. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a system and method for seed industry information question and answer using a large language model. In the present invention, the large language model learns the extensive multi-dimensional knowledge accumulated in the breeding field of my country, and provides real-time and accurate answers to seed industry information-related questions raised by users. This system and method for seed industry information question and answer using a large language model lowers the application threshold through a natural language dialogue interface. Users can directly ask questions and quickly obtain easy-to-understand information, ensuring the real-time and efficient acquisition of information, greatly improving the efficiency of information acquisition.

[0008] Furthermore, the seed industry information question and answer system and method provided by the present invention also integrates question and answer functions in multiple core areas such as variety selection, agronomic traits, cultivation techniques and promotion areas, and can provide comprehensive and accurate breeding decision support, helping breeding experts optimize breeding plans and improve breeding efficiency and results.

[0009] In the first aspect of the present invention, a seed industry information question and answer system is provided, comprising: computing resources; and a large language model, the large language model calls the computing resources to perform seed industry information question and answer, the large language model is configured to learn knowledge related to the seed industry from a seed industry domain knowledge source, the large language model comprises: an input acquisition module, configured to obtain text information related to seed industry information questions from user input; a semantic understanding module, coupled to the input acquisition module, the question parsing module is configured to convert the text information into vector information; an information reasoning module, coupled to the semantic understanding module, the information reasoning module is configured to generate an answer information vector based on knowledge related to the seed industry, in combination with the vector information and context, the answer information vector includes breeding decision information; a result output module, coupled to the information reasoning module, the result output module is configured to convert the answer information vector into answer text information.

[0010] In the seed industry information question-answering system as described above, the large language model is configured to obtain data related to the seed industry from a knowledge source in the seed industry field.

[0011] A seed industry information question and answer system as described in any of the above items, wherein the seed industry field knowledge source includes a professional database in the seed industry field, and / or the seed industry-related data includes at least one of genomic data, phenotypic group data, environmental group data and literature group data.

[0012] A seed industry information question and answer system as described in any of the above items, wherein the large language model is also configured to: correct seed industry related data to avoid omissions and conflicts in the data; and / or integrate seed industry related data to ensure consistency of the data in time and / or space.

[0013] A seed industry information question and answer system as described in any of the above items, wherein the seed industry field knowledge source includes at least one of academic literature, books, field logs, and government public information; and / or the seed industry information questions include questions related to at least one of the following fields: variety breeding, agronomic shape, cultivation technology, and promotion area.

[0014] A seed industry information question and answer system as described in any of the above items, wherein the information reasoning module is further configured to: extract data related to the seed industry and information associated with the data, wherein the information associated with the data includes at least one of the following: the time when the data was generated, the place where the data was generated, the background of the data, and the applicability of the data; and combine the data and the information associated with the data into the answer information vector.

[0015] The seed industry information question and answer system as described in any of the above items also includes a model training module, which includes: a base training module, which is used to train the large language model base on a general training set to obtain a first intermediate large language model; an incremental training module, which is used to perform multimodal incremental training on the first intermediate large language model on a seed industry professional training set to obtain a second intermediate large language model, wherein the seed industry professional training set includes at least one of genomic data, phenotypic group data, environmental group data and literature group data of crop varieties; and a model fine-tuning module, which is used to perform supervised learning on the second intermediate large language model based on the seed industry question and answer sample training set to obtain the large language model.

[0016] A seed industry information question and answer system as described in any of the above items, wherein the large language model is used together with at least one of the following: a knowledge graph, a rule engine and an expert system, and a data warehouse and online analytical processing technology.

[0017] In a second aspect of the present invention, a seed industry information question and answer method based on a large language model is provided, wherein the large language model is configured to learn knowledge related to the seed industry from a knowledge source in the seed industry field, and the method includes using the large language model to perform the following steps: obtaining text information related to seed industry information questions from user input; converting the text information into vector information; generating an answer information vector based on knowledge related to the seed industry and combining the vector information and context, the answer information vector including breeding decision information; and converting the answer information vector into answer text information.

[0018] In the seed industry information question-answering method as described above, the large language model is configured to obtain data related to the seed industry from a knowledge source in the seed industry field.

[0019] A seed industry information question and answer method as described in any of the above items, wherein the seed industry field knowledge source includes a professional database in the seed industry field, and / or the seed industry-related data includes at least one of genomic data, phenotypic group data, environmental group data and literature group data.

[0020] A seed industry information question-and-answer method as described in any of the above items, wherein the large language model is also configured to: correct seed industry-related data to avoid omissions and conflicts in the data; and / or integrate seed industry-related data to ensure consistency of the data in time and / or space.

[0021] A seed industry information question-and-answer method as described in any of the above items, wherein the seed industry field knowledge source includes at least one of academic literature, books, field logs, and government public information; and / or the seed industry information questions include questions related to at least one of the following fields: variety breeding, agronomic shape, cultivation technology, and promotion area.

[0022] A seed industry information question and answer method as described in any of the above items, wherein generating an answer information vector based on knowledge related to the seed industry and combining the vector information and contextual context includes: extracting data related to the seed industry and information associated with the data, wherein the information associated with the data includes at least one of the following: the time when the data was generated, the place where the data was generated, the background of the data, and the applicability of the data; and combining the data and the information associated with the data into the answer information vector.

[0023] A seed industry information question and answer method as described in any of the above items, wherein the large language model is trained by the following steps: training a large language model base on a general training set to obtain a first intermediate large language model; performing multimodal incremental training on the first intermediate large language model on a seed industry professional training set to obtain a second intermediate large language model, wherein the seed industry professional training set includes at least one of genomic data, phenotypic group data, environmental group data and literature group data of crop varieties; and performing supervised learning on the second intermediate large language model based on the seed industry question and answer sample training set to obtain the large language model.

[0024] A seed industry information question-and-answer method as described in any of the above items, wherein the large language model is used together with at least one of the following: a knowledge graph, a rule engine and an expert system, and a data warehouse and online analytical processing technology.

[0025] In a third aspect of the present invention, a machine-readable medium is provided, comprising machine-executable instructions stored thereon, wherein when the instructions are executed, the machine is caused to perform any of the methods described above.

[0026] In a fourth aspect of the present invention, a computer program product is provided, comprising computer instructions, which implement the method as described in any one of the above when executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Some implementations of the present invention will be described below by way of example only and with reference to the accompanying drawings. In the drawings, each identical or nearly identical component shown throughout the various drawings may be represented by the same reference numeral. For clarity, not every component is labeled in each figure. In the drawings:

[0028] Figure 1 An exemplary application scenario of the seed industry information question and answer system according to an embodiment of the present invention is shown;

[0029] Figure 2 A seed industry information question and answer system according to an embodiment of the present invention is shown;

[0030] Figure 3 A seed industry information question and answer system according to another embodiment of the present invention is shown;

[0031] Figure 4 The seed industry information question and answer method according to an embodiment of the present invention is shown;

[0032] Figure 5 An operation for generating a reply information vector according to an embodiment of the present invention is shown;

[0033] Figure 6 The operation of constructing a large language model suitable for seed industry information question and answering according to an embodiment of the present invention is shown;

[0034] Figure 7 The operation of training a large language model suitable for answering questions about seed industry information according to another embodiment of the present invention is shown. DETAILED DESCRIPTION

[0035] Next, exemplary embodiments of the present invention will be described in detail with reference to the attached drawings, wherein, unless otherwise expressly stated, the same or similar reference numerals in different drawings represent the same or similar elements. In addition, it should be noted that the exemplary embodiments of the present invention may perform the steps of the corresponding method in a different order than that shown and described in the specification. In some embodiments of the present invention, the method may include more or fewer steps than those described in the specification and shown in the attached drawings. In addition, a single step used in some embodiments of this specification may be decomposed into multiple steps in other embodiments, or multiple steps used in some embodiments of this specification may be combined into a single step in other embodiments.

[0036] In the following description, numerous specific details are set forth. However, it should be understood that embodiments of the present invention may be practiced without these specific details. In other instances, well-known circuits, structures, and techniques are not shown in detail to avoid obscuring the understanding of this description.

[0037] References in the specification to "one embodiment," "an embodiment," "an example embodiment," etc. indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment may include the particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Furthermore, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it is considered to be within the knowledge of those skilled in the art to affect such feature, structure, or characteristic in conjunction with other embodiments, whether or not explicitly described.

[0038] For purposes of this disclosure, the phrase "A and / or B" means (A), (B), or (A and B). For purposes of this disclosure, the phrase "A, B, and / or C" means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0039] In the following description and claims, the terms "coupled" and "connected" and their derivatives may be used. It should be understood that these terms are not intended to be synonymous with each other. On the contrary, in certain embodiments, "connected" may be used to indicate that two or more elements are in direct physical or electrical contact with each other. "Coupled" may mean that two or more elements are in direct physical or electrical contact. However, "coupled" may also mean that two or more elements are not in direct contact with each other, but still cooperate or interact with each other.

[0040] At least some embodiments of the present invention provide a seed industry information question-and-answer system that can be used to answer professional questions in the seed industry, and provide strong support for agricultural scientific research and decision-making such as seed industry research, seed selection and breeding. Specifically, the present invention involves pre-training and fine-tuning a large language model base to form a knowledge base and corpus adapted to the seed industry, so as to use the large language model to extract information and generate dialogues, so that it can be applied to answer professional questions in the seed industry.

[0041] Figure 1 An exemplary application scenario of the seed industry information question-and-answer system according to an exemplary embodiment of the present invention is shown. Figure 1 In the exemplary application scenario shown in , a user 12 inputs seed industry information-related questions to a seed industry information question and answer system 16 via a terminal device 14. The seed industry information question and answer system 16 generates answers to the user questions based on the received user questions using a trained and fine-tuned large language model, and outputs the answers to the user 12 in an appropriate form (e.g., text, audio, etc.) via the terminal device 14.

[0042] As an example, large language models generally use deep learning technology and a large amount of text data to run. Large language models are usually based on the Transformer architecture and consist of multiple layers of neural networks. The parameters of each layer of the neural network can be fine-tuned during the training process. These neural network layers use attention mechanisms, allowing the model to pay attention to the information of other positions when processing each input position, and can dynamically adjust the importance of different positions, further enhancing the function of the neural network. In the large language model, the encoder is used for position encoding to represent the position information of the words in the input sequence, and the position information is encoded into a vector form to capture the semantic information of the input sequence; the decoder is responsible for generating the target sequence based on the output of the encoder; thus, the large language model uses the decoder to generate the target output sequence based on the input feature vector generated by the encoder. Generally speaking, during the generation process, the decoder can use the attention mechanism to pay attention to different parts of the input sequence, thereby generating a more accurate output sequence.

[0043] As an example, for domain-optimized large language models such as BioBERT and SciBERT, domain-specific papers are used to construct training sets and train the BERT base model to make it suitable for downstream tasks in the corresponding field. For example, BioBERT is first pre-trained using the BERT base model and then fine-tuned on a large number of biomedical papers in the PubMed database, so that the model can understand and capture professional vocabulary and semantics in the biomedical field.

[0044] According to an embodiment of the present invention, the seed industry information question and answer system 16 has learned knowledge of seed industry related disciplines such as crop science, agronomy, genetics, and related fields such as biology, meteorology, and soil science from various information sources including academic literature, books, field logs, and government announcements, and has gathered this knowledge into the model weights through the technical means of the large language model, so that the large language model has the function of understanding seed industry related issues and providing intelligent answers in real time. According to an embodiment of the present invention, the seed industry information question and answer system 16 can be trained and fine-tuned based on various existing large language models to achieve professional, accurate, and fast seed industry information questions and answers.

[0045] like Figure 1 As shown, according to some embodiments of the present invention, the seed industry information question and answer system 16 and the terminal device 14 can be separate devices, and the seed industry information question and answer system 16 can be connected to the terminal device 14 by wire and / or wirelessly, and transmit interactive information in accordance with an agreed data format.

[0046] Alternatively, according to some other embodiments of the present invention, the seed industry information question and answer system 16 may be implemented in the terminal device 14 (not shown). As an example, the seed industry information question and answer system 16 may be implemented in the terminal device as a software module and / or a hardware module. For example, the seed industry information question and answer system 16 may be a software module in the operating system of the terminal device 14, an application developed for the terminal device 14, or a hardware module of the terminal device 14.

[0047] According to an embodiment of the present invention, the seed industry information question and answer system 16 can be implemented as an intelligent assistant terminal device. As an exemplary implementation, such an intelligent assistant terminal device can provide various user interfaces to receive user input, such as providing a microphone to receive audio information input by the user, providing a physical or virtual keyboard to receive text information input by the user, and providing a graphical user interface to receive tactile input. According to an embodiment of the present invention, speech recognition and natural language processing technology can be used to convert the received user input into text information, and provide it to a large language model to implement seed industry information question and answer, so as to improve the user experience of the seed industry information question and answer system.

[0048] According to an embodiment of the present invention, the seed industry information question and answer system 16 can be implemented as a computer program product, as a mobile application running on the user terminal 14 or other appropriate hardware devices, to facilitate users to query seed industry information.

[0049] Next reference Figure 2 The seed industry information question and answer system 20 according to an embodiment of the present invention is described below. Figure 2 As shown, the seed industry information question and answer system 20 includes a large language model 202 and computing resources 204, wherein the large language model 202 calls the computing resources 204 to perform seed industry information question and answer.

[0050] According to an embodiment of the present invention, the large language model 202 is configured to learn knowledge related to the seed industry from a seed industry domain knowledge source. According to an embodiment of the present invention, the large language model 202 learns knowledge related to the seed industry from a seed industry domain knowledge source, which may include obtaining data related to the seed industry from the seed industry domain knowledge source.

[0051] According to an embodiment of the present invention, the seed industry knowledge source includes a professional database in the seed industry, and / or the data related to the seed industry includes at least one of genomic data, phenotypic group data, environmental group data, and literature group data. As an exemplary implementation, the professional database in the seed industry is specially designed for the seed industry, and contains professional data on biological breeding research collected for major food species in my country. These professional data include detailed genomic data, agricultural traits and phenotypic data, environmental data, breeding test data in different eco-geographical regions, etc.

[0052] According to an embodiment of the present invention, knowledge sources in the seed industry may also include government public information, such as seed industry-related data released by agricultural management departments at all levels of government in my country, including policies, regulations, and statistical data; academic literature and books related to the seed industry (for example, academic literature published on my country’s major food species), which cover authoritative and latest research results and experimental data; and / or collected field logs, which cover the accumulated experience in agricultural production practice.

[0053] According to an embodiment of the present invention, the large language model 202 obtains data related to the seed industry from the seed industry domain knowledge source, which may further include correcting and / or integrating the obtained data related to the seed industry. According to an embodiment of the present invention, before the large language model 202 learns knowledge related to the seed industry from the seed industry domain knowledge source, the data related to the seed industry obtained from the seed industry domain knowledge source may be preprocessed, which may include: correcting the data to identify and process omissions and conflicting parts in the data; and / or integrating the data to integrate scattered data sources so that the data is consistent in time and / or space. As an exemplary implementation, the professional knowledge of biological breeding can be used to correct the data. For example, the rice genome data of the second generation sequencing and the third generation sequencing can be cross-fused to ensure the consistency and integrity of the data. As another exemplary implementation, the conflicting data can be integrated through domain knowledge to ensure the accuracy and reliability of the data. According to an embodiment of the present invention, preprocessing the data ensures the high quality and professionalism of the data, so that the large language model has higher accuracy when dealing with complex biological breeding problems.

[0054] Continue to refer Figure 2 According to an embodiment of the present invention, the large language model 202 may include an input acquisition module 2022, which may be configured to obtain text information related to seed industry information issues from user input. As an exemplary implementation, the input acquisition module 2022 may be coupled to a user's terminal device, and collect user input via the terminal device and obtain text information related to seed industry information issues from the user input. As an example, a user may input text, audio, etc. containing seed industry information issues via a user interface provided by the terminal device (e.g., a physical and / or virtual keyboard, microphone, graphical user interface, etc.), and the input acquisition module 2022 may obtain text information related to seed industry issues from these user inputs.

[0055] like Figure 2As shown, the large language model 202 may also include a semantic understanding module 2024, which is coupled to the input acquisition module 2022. According to an embodiment of the present invention, the semantic understanding module 2024 may be configured to convert text information related to seed industry information issues obtained by the input acquisition module 2022 into vector information. As an exemplary implementation, the semantic understanding module 2024 may perform word segmentation on the text information, and use a semantic encoder to encode the segmented text information, capture the semantic relationship and context information between words, so as to extract text feature word sequences related to seed industry information issues, generate word feature vectors of seed industry information issues, and thus convert text information related to seed industry information issues into vector information. Generally, the text information input by the user may contain non-standard expressions, typos or grammatical errors, etc., and / or the text information may also include non-professional descriptions that lack seed industry professional terms and expressions. Therefore, by encoding and understanding the text information based on the corpus learned by the large language model through the semantic understanding module 2024, the text information can be optimized to generate a more accurate and meaningful feature vector representation that is easy for the large language model to understand.

[0056] According to an embodiment of the present invention, the semantic understanding module 2024 can convert text information into vector information of seed industry information problems associated with a specific seed industry field. According to an embodiment of the present invention, seed industry information problems can be associated with at least one of the following fields: variety breeding, agronomic shape, cultivation technology, and promotion area. According to an embodiment of the present invention, seed industry information problems related to the field of variety breeding are intended to obtain historical breeding records of existing varieties, including successful cases of variety breeding and problems and challenges of variety breeding, so as to learn from these successful experiences to optimize the current breeding plan and avoid repeating previous mistakes in new breeding projects, thereby improving the success rate of breeding. According to an embodiment of the present invention, seed industry information problems related to the field of agronomic traits are intended to obtain crop growth characteristics and performance (such as plant morphology, maturity days, grain quality, etc.) to understand and select the most suitable germplasm resources and optimize breeding plans. According to an embodiment of the present invention, seed industry information problems related to the field of cultivation technology are intended to obtain the best cultivation plan (such as irrigation method, fertilization plan, etc.) suitable for the environmental conditions and crop characteristics provided by the user, so as to seek the most suitable cultivation technology adapted to specific conditions to optimize crop growth conditions and improve yield and quality. According to an embodiment of the present invention, seed industry information issues related to promotion areas are intended to obtain historical promotion areas, so as to learn the adaptability and performance of crop varieties under different environmental conditions and evaluate the adaptability and promotion value of crops in different regions, so as to seek scientific promotion area recommendations and improve the adaptability and market competitiveness of crops.

[0057] like Figure 2As shown, the large language model 202 may also include an information reasoning module 2026, which is coupled to the semantic understanding module 2024. According to an embodiment of the present invention, the information reasoning module 2026 may be configured to generate an answer information vector based on knowledge related to the seed industry, in combination with the vector information generated by the semantic understanding module 2024, and the context context. According to an embodiment of the present invention, the answer information vector may include breeding decision information. As an exemplary implementation, the information reasoning module 2026 may query the seed industry knowledge base of the large language model based on the vector information generated by the semantic understanding module 2024, and generate an answer information vector in combination with the context context. According to an embodiment of the present invention, the seed industry knowledge base stores the seed industry expertise of crop varieties learned by the large language model from the seed industry knowledge source, and multi-dimensionally associates the seed industry expertise with respect to the following fields: variety breeding, agronomic traits, cultivation techniques and promotion areas. According to an embodiment of the present invention, corresponding to the vector information of the seed industry information question, the answer information vector may also be associated with at least one of variety breeding, agronomic traits, cultivation techniques and promotion areas. As an exemplary implementation, the answer information vector can be associated with the field of variety breeding to provide historical breeding records of existing varieties; the answer information vector can be associated with the field of agronomic traits to provide crop growth characteristics and performance; the answer information vector can be associated with the field of cultivation technology to provide a cultivation plan that is compatible with the environmental conditions and crop characteristics provided by the user; and / or the answer information vector can be associated with the promotion area to provide the historical promotion area of ​​the crop variety, as well as an assessment of the adaptability and promotion value of the crop in different regions. According to an embodiment of the present invention, the answer information vector may include breeding decision information associated with the above-mentioned various fields. In addition, according to an embodiment of the present invention, the information reasoning module 2026 may be configured to extract data related to the seed industry and information associated with the data, and combine the data and the information associated with the data into the answer information vector. According to an embodiment of the present invention, the information associated with the data may include at least one of the following: the time when the data was generated, the place where the data was generated, the background of the data, and the applicability of the data.

[0058] like Figure 2 As shown, the large language model 202 may further include a result output module 2028, which is coupled to the information reasoning module 2026. According to an embodiment of the present invention, the result output module 2028 may be configured to convert the answer information vector into answer text information. As an exemplary implementation, the result output module 2028 may generate answer text information based on the answer information vector based on the corpus learned by the large language model and the seed industry professional knowledge base.

[0059] According to the embodiment of the present invention, the seed industry information question and answer system 20 can understand user questions through large language model technology, and give corresponding professional answers and intelligent suggestions, thereby assisting seed industry workers to make breeding decisions accurately and efficiently.

[0060] Figure 3 Another industry information question and answer system 30 according to an embodiment of the present invention is shown. Figure 3 As shown, in addition to the large language model 302 and the computing resources 304, the seed industry information question answering system 30 also includes a model training module 306, which is coupled to the large language model 302. According to an embodiment of the present invention, the model training module 306 can be configured to pre-train and fine-tune the large language model base to obtain a large language model 302 suitable for seed industry information question answering.

[0061] Figure 3 The large language model 302 and computing resources 304 shown in FIG. Figure 2 The described large language model 202 and computing resources 204 are similar and are not described in detail herein to avoid obscuring the understanding of the present invention.

[0062] like Figure 3 As shown, the model training module 306 may include a base training module 3062. According to an embodiment of the present invention, the base training module 3062 may be configured to train the large language model base on a universal training set to obtain a first intermediate large language model. As an exemplary implementation, the base training module 3062 may be configured to train an appropriate large language model base on a universal training set so that the large language model base learns general knowledge in an unsupervised learning manner to obtain a first intermediate large language model that has learned the general knowledge.

[0063] like Figure 3As shown, the model training module 306 may also include an incremental training module 3064, which is coupled to the base training module 3062. According to an embodiment of the present invention, the incremental training module 3064 may be configured to perform multimodal incremental training on the first intermediate large language model trained by the base training module 3062 on the seed industry professional training set to obtain a second intermediate large language model. According to an embodiment of the present invention, the seed industry professional training set may include at least one of the genomic data, phenotypic group data, environmental group data and literature group data of the crop variety. According to an embodiment of the present invention, the incremental training module 3064 may align its genomic data, phenotypic group data, environmental group data and literature group data based on the crop variety, so that incremental training can be performed in multiple modes such as genome, phenotypic group, environmental group, and literature group. As an exemplary implementation, the incremental training module 3064 may be configured to perform multimodal incremental training on the first intermediate large language model that has learned general knowledge on the aligned seed industry professional training set, so that the first intermediate large language model learns the professional knowledge of the seed industry field in an unsupervised learning manner, thereby obtaining a second intermediate large language model.

[0064] like Figure 3 As shown, the model training module 306 may also include a model fine-tuning module 3066, which is coupled to the incremental training module 3064. According to an embodiment of the present invention, the model fine-tuning module 3066 may be configured to conduct supervised learning on the second intermediate large language model based on the seed industry question and answer sample training set to obtain a large language model suitable for seed industry information question and answer. As an exemplary implementation, the model fine-tuning module 3066 may conduct supervised learning on the second intermediate large language model through artificially prepared seed industry-related questions and corresponding answers, so that the large language model obtains question and answer capabilities that meet human preferences. For example, in view of the question and answer habits of breeders who are more concerned about the background and applicability of data, a seed industry question and answer sample training set that provides detailed background information and data interpretation of breeding can be designed to make the answers more in line with the breeders' usage habits.

[0065] According to an embodiment of the present invention, the seed industry information question and answer system 30 can train the existing general large language model, and use multimodal incremental training to enable the large language model to provide more comprehensive and logical answers, and through model fine-tuning, the answers of the large language model are made consistent with the question and answer habits in the breeding field.

[0066] According to at least some embodiments of the present invention, the large language model can be used in combination with at least one of the following: a knowledge graph, a rule engine and an expert system, as well as a data warehouse and an online analytical processing technology. According to an embodiment of the present invention, the large language model can be used together with the knowledge graph technology to implement a seed industry information question and answer system. As an exemplary implementation, knowledge entities and relationships can be extracted from the seed industry knowledge source to construct a comprehensive breeding knowledge graph, and semantic query and reasoning can be performed on the knowledge graph based on the vector information of the seed industry information questions generated by the large language model to implement seed industry information question and answer, thereby further improving the intelligence and accuracy of the seed industry information question and answer system. According to an embodiment of the present invention, the large language model can be used together with a rule engine and an expert system to implement a seed industry information question and answer system. As an exemplary implementation, the seed industry domain professional knowledge can be collected and sorted from the seed industry knowledge source and stored as a knowledge base of the expert system; matching rules between seed industry information questions and seed industry information answers can be designed to construct a rule engine; based on the rule engine and the knowledge base, the vector information of the seed industry information questions generated by the large language model can be reasoned, and the final seed industry information answers can be generated via the large language model, thereby improving the flexibility and practicality of the seed industry information question and answer system. According to an embodiment of the present invention, a large language model can be used together with a data warehouse and online analytical processing (OLAP) technology to implement a seed industry information question-and-answer system. As an exemplary implementation, a data warehouse can be used to implement the construction of a knowledge base of a large language model, and OLAP technology can be used to implement efficient query and analysis of the knowledge base based on vector information of seed industry information questions generated by a large language model, thereby improving the efficiency and responsiveness of the seed industry information question-and-answer system. According to an embodiment of the present invention, a large language model can also be used together with any two or more of a knowledge graph, a rule engine and an expert system, and a data warehouse and online analytical processing technology to implement a seed industry information question-and-answer system.

[0067] Figure 4 A flow chart of a seed industry information question answering method 40 according to an embodiment of the present invention is shown. According to an embodiment of the present invention, the seed industry information question answering method 40 performs seed industry information question answering based on a large language model, which is configured to learn knowledge related to the seed industry from a seed industry domain knowledge source.

[0068] like Figure 4 As shown, method 40 includes: at block 402, the large language model obtains text information related to the seed industry information question from user input. According to an embodiment of the present invention, the user can input text, audio, etc. containing the seed industry information question via a user interface (e.g., a physical and / or virtual keyboard, a microphone, a graphical user interface, etc.) provided by the terminal device, and the large language model can obtain text information related to the seed industry question from these user inputs.

[0069] like Figure 4 As shown, method 40 may also include: at box 404, the large language model converts the obtained text information into vector information. As an exemplary implementation, the large language model can perform word segmentation on the text information, and use a semantic encoder to encode the segmented text information, capture the semantic relationship and context information between words, so as to extract text feature word sequences related to seed industry information issues, generate word feature vectors of seed industry information issues, and thus convert the text information related to seed industry information issues into vector information. According to an embodiment of the present invention, seed industry information issues may be associated with at least one of the following fields: variety breeding, agronomic shape, cultivation technology, and promotion area, and accordingly, the vector information of the seed industry information issue may include various dimensions associated with these fields. According to an embodiment of the present invention, seed industry information issues related to the field of variety breeding are intended to obtain historical breeding records of existing varieties in order to understand successful cases of variety breeding as well as problems and challenges of variety breeding; seed industry information issues related to the field of agronomic traits are intended to obtain crop growth characteristics and performance (such as plant morphology, days to maturity, grain quality, etc.) in order to understand and select the most suitable germplasm resources; seed industry information issues related to the field of cultivation technology are intended to obtain the best cultivation schemes (such as irrigation methods, fertilization schemes, etc.) suitable for the environmental conditions and crop characteristics provided by the user, in order to seek the most suitable cultivation technology adapted to specific conditions to optimize crop growth conditions; seed industry information issues related to promotion areas are intended to obtain historical promotion areas, so as to learn the adaptability and performance of crop varieties under different environmental conditions and the assessment of the adaptability and promotion value of crops in different regions, in order to seek scientific promotion area recommendations.

[0070] like Figure 4 As shown, method 40 may also include: at box 406, the large language model generates an answer information vector based on the knowledge related to the seed industry, combined with the vector information and the context. According to an embodiment of the present invention, the answer information vector includes breeding decision information. As an exemplary implementation, the large language model can use the seed industry knowledge base based on the vector information of the seed industry information question, and generate an answer information vector in combination with the context. According to an embodiment of the present invention, the seed industry knowledge base stores the seed industry expertise of crop varieties learned by the large language model from the seed industry knowledge source, and performs multi-dimensional associative storage of the seed industry expertise in the following fields: variety selection, agronomic traits, cultivation techniques, and promotion areas. According to an embodiment of the present invention, corresponding to the vector information of the seed industry information question, the answer information vector may also be associated with at least one of variety selection, agronomic traits, cultivation techniques, and promotion areas.

[0071] like Figure 4As shown, method 40 may further include: at block 408, the large language model converts the answer information vector into answer text information. As an exemplary implementation, the large language model may generate answer text information according to the answer information vector based on its corpus and seed industry professional knowledge base.

[0072] Figure 5 A flowchart of operation 50 for generating a response information vector by a large language model based on knowledge related to the seed industry and combining the vector information with contextual context according to an embodiment of the present invention is shown.

[0073] like Figure 5 As shown, operation 50 may include: extracting data related to the seed industry and information associated with the data at block 502. According to an embodiment of the present invention, the information associated with the data may include at least one of the following: the time when the data was generated, the location where the data was generated, the background of the data, and the applicability of the data.

[0074] like Figure 5 As shown, operation 50 may also include, at block 504 , combining the data and information associated with the data into a reply information vector.

[0075] The seed industry information question-and-answer method according to an embodiment of the present invention can understand user questions through large language model technology, and give corresponding professional answers and intelligent suggestions, thereby assisting seed industry workers to make breeding decisions accurately and efficiently. In addition, according to an embodiment of the present invention, in addition to seed industry-related data, the answer information vector also includes information related to the data. This model answering method is adapted to the question-and-answer habits in the breeding field, and pays more attention to the logic of the answer. The seed industry information question-and-answer method according to an embodiment of the present invention not only provides specific data, but also emphasizes the time and place when the data is generated, thereby highlighting the logic of crops being affected by the environment, and provides the background and applicability of the data in the answer, so that the answer is more in line with the usage habits of breeders.

[0076] Figure 6 A flowchart of operation 60 of building a large language model suitable for answering questions about seed industry information according to an embodiment of the present invention is shown.

[0077] like Figure 6 As shown, operation 60 may include: at block 602 , obtaining data related to the seed industry from a seed industry domain knowledge source.

[0078] According to an embodiment of the present invention, the seed industry knowledge source includes a professional database in the seed industry, and / or the data related to the seed industry includes at least one of genomic data, phenotypic group data, environmental group data, and literature group data. As an exemplary implementation, the professional database in the seed industry is specially designed for the seed industry, and contains professional data on biological breeding research collected for major food species in my country. These professional data include detailed genomic data, agricultural traits and phenotypic data, environmental data, breeding test data in different eco-geographical regions, etc.

[0079] According to an embodiment of the present invention, knowledge sources in the seed industry may also include government public information, such as seed industry-related data released by agricultural management departments at all levels of government in my country, including policies, regulations, and statistical data; academic literature and books related to the seed industry (for example, academic literature published on my country’s major food species), which cover authoritative and latest research results and experimental data; and / or collected field logs, which cover the accumulated experience in agricultural production practice.

[0080] like Figure 6 As shown, operation 60 may further include: at box 604 , correcting and / or integrating the seed industry-related data obtained at box 602 .

[0081] According to an embodiment of the present invention, data related to the seed industry obtained from a seed industry domain knowledge source can be preprocessed, including: correcting the data to identify and process omissions and conflicting parts in the data; and / or integrating the data to integrate scattered data sources so that the data is consistent in time and space. As an exemplary implementation, the professional knowledge of biological breeding can be used to correct the data. For example, the rice genome data of the second generation sequencing and the third generation sequencing can be cross-fused to ensure the consistency and integrity of the data. As another exemplary implementation, the conflicting data can be integrated through domain knowledge to ensure the accuracy and reliability of the data.

[0082] According to an embodiment of the present invention, the seed industry information question and answer method can be used to adopt Figure 6 The large language model constructed by operation 60 shown above, which has learned knowledge related to the seed industry from knowledge sources in the seed industry field, is used to perform seed industry information question and answer.

[0083] According to an embodiment of the present invention, operation 60 obtains seed industry related data from a knowledge source in the seed industry field, and adopts a professional database designed for the seed industry field that provides detailed genomic data, agricultural traits and phenotypic data, environmental data, breeding test data in different eco-geographical regions, etc., to ensure the comprehensiveness, professionalism and authority of the corpus and knowledge base learned by the large language model. In addition, method 60 pre-processes the collected raw data, identifies and processes omissions and conflicting parts in the data, and ensures the accuracy and reliability of the data. Therefore, operation 60 ensures the high quality and professionalism of the data, so that the model has higher accuracy when dealing with complex biological breeding problems.

[0084] Figure 7 A flowchart of operation 70 for training a large language model suitable for seed industry information question answering according to an embodiment of the present invention is shown.

[0085] like Figure 7 As shown, operation 70 may include: at box 702, training a large language model base on a universal training set to obtain a first intermediate large language model. As an exemplary implementation, an appropriate large language model base may be trained on a universal training set so that the large language model base learns universal knowledge in an unsupervised learning manner to obtain a first intermediate large language model that has learned the universal knowledge.

[0086] like Figure 7 As shown, operation 70 may also include: at box 704, multimodal incremental training is performed on the first intermediate large language model trained at box 702 on the seed industry professional training set to obtain a second intermediate large language model. According to an embodiment of the present invention, the seed industry professional training set includes at least one of the genomic data, phenotypic group data, environmental group data, and literature group data of crop varieties. According to an embodiment of the present invention, the genomic data, phenotypic group data, environmental group data, and literature group data of crop varieties can be aligned, and multimodal incremental training is performed on the first intermediate large language model that has learned general knowledge on the aligned seed industry professional training set, so that the first intermediate large language model learns the professional knowledge in the seed industry field in an unsupervised learning manner, thereby obtaining a second intermediate large language model.

[0087] like Figure 7As shown, operation 70 may also include: at box 706, the second intermediate large language model trained at box 704 is subjected to supervised learning based on the seed industry question and answer sample training set to obtain a large language model suitable for seed industry information question and answer. According to an embodiment of the present invention, the second intermediate large language model can be subjected to supervised learning through artificially prepared seed industry-related questions and corresponding answers, so that the large language model obtains question and answer capabilities that meet human preferences. For example, in view of the question and answer habits of breeders who are more concerned about the background and applicability of data, a seed industry question and answer sample training set that provides detailed background information and data interpretation of breeding can be designed to make the answers more in line with the breeders' usage habits.

[0088] According to an embodiment of the present invention, the seed industry information question and answer method can be based on Figure 7 The large language model obtained by training in operation 70 shown in FIG. 1 is used to perform seed industry information question answering.

[0089] According to an embodiment of the present invention, through pre-training and fine-tuning of the large language model base, the large language model can be suitable for question and answering in the seed industry, the large language model can provide more comprehensive and logical answers, and the large language model conforms to the question and answer habits in the breeding field.

[0090] According to at least some embodiments of the present invention, the large language model can be used together with at least one of the following: a knowledge graph, a rule engine and an expert system, as well as a data warehouse and online analytical processing technology. As an exemplary implementation, knowledge entities and relationships can be extracted from the seed industry knowledge source to construct a comprehensive breeding knowledge graph, and semantic query and reasoning can be performed on the knowledge graph based on the vector information of the seed industry information questions generated by the large language model to realize seed industry information question and answer, thereby further improving the intelligence and accuracy of the seed industry information question and answer method. As another exemplary implementation, professional knowledge in the seed industry can be collected and sorted from the seed industry knowledge source and stored as a knowledge base of the expert system; matching rules between seed industry information questions and seed industry information answers can be designed to build a rule engine; based on the rule engine and the knowledge base, the vector information of the seed industry information questions generated by the large language model can be reasoned, and the final seed industry information answers can be generated via the large language model, thereby improving the flexibility and practicality of the seed industry information question and answer method. According to another embodiment of the present invention, a data warehouse can be used to implement the construction of a knowledge base of a large language model, and OLAP technology can be used to implement efficient query and analysis of the knowledge base based on the vector information of seed industry information questions generated by the large language model, thereby improving the efficiency and responsiveness of the seed industry information question-answering system. According to an embodiment of the present invention, the large language model can also be used together with any two or more of the knowledge graph, rule engine and expert system, and data warehouse and online analytical processing technology to implement the seed industry information question-answering method.

[0091] One or more aspects of at least one embodiment of the present invention may be implemented as a computer program product, wherein the computer program product includes computer instructions. Program code is applied to input instructions to perform the functions described herein and generate output information. Output information can be applied to one or more output devices in a known manner. For the purpose of this application, the present invention also provides a computing device, which includes any system with a processor, such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC) or a microprocessor. The computing device also includes a memory, which is coupled to the processor for storing program code. The program code can be implemented with a high-level process programming language or an object-oriented programming language.

[0092] One or more aspects of at least one embodiment of the invention may be implemented by representative instructions stored on a machine-readable medium that represent various logic within a processor, which when read by a machine causes the machine to fabricate logic for performing the techniques described herein.

[0093] The present invention provides a system and method for seed industry information question and answer using a large language model, wherein the large language model has learned the extensive multi-dimensional knowledge accumulated in the breeding field of my country, and can support real-time and accurate answers to seed industry information-related questions raised by users. This system and method for seed industry information question and answer using a large language model lowers the application threshold through a natural language dialogue interface, and users can directly ask questions and quickly obtain easy-to-understand information, ensuring the real-time and efficient acquisition of information, greatly improving the efficiency of information acquisition.

[0094] In addition, the seed industry information question-and-answer system and method based on a large language model proposed in the present invention utilizes a technology that combines artificial intelligence and biological breeding knowledge, and can also provide agricultural policy support, providing government agencies with comprehensive and accurate information consultation related to seed industry information and policy recommendations based on seed industry big data; providing business consulting service support, assisting breeding companies and other agricultural science and technology companies in providing professional consulting services, helping them optimize breeding plans and market strategies; providing agricultural education and training support, providing teaching auxiliary tools for relevant teachers and students of colleges and universities, assisting agricultural colleges and scientific research institutions in agricultural education and training, and helping students and researchers to quickly acquire and understand breeding knowledge; and providing international cooperation and exchange support, promoting international agricultural technology cooperation and exchanges with multi-language and multi-regional data analysis functions.

[0095] The preferred embodiments of the present invention are described in detail above. However, it should be understood that the present invention can be implemented in various ways and variations without departing from its broad spirit and scope. A person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A seed industry information question and answer system, comprising: Computing resources; as well as A large language model, wherein the large language model calls the computing resources to perform seed industry information question answering, wherein the large language model is configured to learn knowledge related to the seed industry from a seed industry domain knowledge source, and wherein the large language model comprises: An input acquisition module is configured to obtain text information related to seed industry information issues from user input; A semantic understanding module, coupled to the input acquisition module, wherein the semantic understanding module is configured to convert the text information into vector information; An information reasoning module, coupled to the semantic understanding module, wherein the information reasoning module is configured to generate an answer information vector based on knowledge related to the seed industry and in combination with the vector information and context, wherein the answer information vector includes breeding decision information; The result output module is coupled to the information reasoning module, and is configured to convert the answer information vector into answer text information.

2. The seed industry information question and answer system according to claim 1, in, The large language model is configured as follows: Obtain data related to the seed industry from seed industry domain knowledge sources.

3. The seed industry information question and answer system according to claim 2, wherein: The seed industry knowledge source includes a professional database in the seed industry, and / or The data related to the seed industry include at least one of genomic data, phenotypic group data, environmental group data and literature group data.

4. The seed industry information question and answer system according to claim 2, in, The large language model is also configured to: Correcting data related to the seed industry to avoid omissions and conflicts in the data; and / or Integrate seed industry-related data to ensure consistency in time and / or space.

5. The seed industry information question and answer system according to claim 1, wherein: The seed industry knowledge source includes at least one of academic literature, books, field diaries, and government public information; and / or The seed industry information issues include issues related to at least one of the following areas: variety breeding, agronomic shape, cultivation technology, and promotion area.

6. The seed industry information question and answer system according to claim 1, wherein: The information reasoning module is also configured to: Extracting data related to the seed industry and information associated with the data, wherein the information associated with the data includes at least one of the following: the time when the data was generated, the location where the data was generated, the background of the data, and the applicability of the data; and The data and information associated with the data are combined into the answer information vector.

7. The seed industry information question-and-answer system according to claim 1, further comprising a model training module, wherein the model training module comprises: A base training module is configured to train the large language model base on a universal training set to obtain a first intermediate large language model; An incremental training module is configured to perform multimodal incremental training on the first intermediate large language model on a seed industry professional training set to obtain a second intermediate large language model, wherein the seed industry professional training set includes at least one of genomic data, phenotypic group data, environmental group data, and literature group data of crop varieties; as well as The model fine-tuning module is configured to perform supervised learning on the second intermediate large language model based on the seed industry question and answer sample training set to obtain the large language model.

8. The seed industry information question and answer system according to claim 1, wherein: The large language model is used together with at least one of the following: a knowledge graph, a rule engine and an expert system, and a data warehouse and online analytical processing technology.

9. A seed industry information question-answering method based on a large language model, wherein the large language model is configured to learn knowledge related to the seed industry from a seed industry domain knowledge source, and the method comprises using the large language model to perform the following steps: Obtain text information related to seed industry information issues from user input; Convert the text information into vector information; Generate an answer information vector based on knowledge related to the seed industry and in combination with the vector information and context, wherein the answer information vector includes breeding decision information; and The answer information vector is converted into answer text information.

10. The seed industry information question and answer method according to claim 9, wherein: The large language model is configured as follows: Obtain data related to the seed industry from seed industry domain knowledge sources.

11. The seed industry information question-and-answer method according to claim 10, wherein: The seed industry knowledge source includes a professional database in the seed industry, and / or The data related to the seed industry include at least one of genomic data, phenotypic group data, environmental group data and literature group data.

12. The seed industry information question and answer method according to claim 10, wherein: The large language model is also configured to: Correcting data related to the seed industry to avoid omissions and conflicts in the data; and / or Integrate seed industry-related data to ensure consistency in time and / or space.

13. The seed industry information question and answer method according to claim 9, wherein: The seed industry knowledge source includes at least one of academic literature, books, field diaries, and government public information; and / or The seed industry information issues include issues related to at least one of the following areas: variety breeding, agronomic shape, cultivation technology, and promotion area.

14. The seed industry information question and answer method according to claim 9, wherein: Generating an answer information vector based on knowledge related to the seed industry and combining the vector information and context includes: Extracting data related to the seed industry and information associated with the data, wherein the information associated with the data includes at least one of the following: the time when the data was generated, the location where the data was generated, the background of the data, and the applicability of the data; and The data and information associated with the data are combined into the answer information vector.

15. The seed industry information question and answer method according to claim 9, wherein: The large language model is trained by the following steps: Training the large language model base on the universal training set to obtain a first intermediate large language model; Performing multimodal incremental training on the first intermediate large language model on a seed industry professional training set to obtain a second intermediate large language model, wherein the seed industry professional training set includes at least one of genomic data, phenotype group data, environmental group data, and literature group data of crop varieties; and The second intermediate large language model is subjected to supervised learning based on the seed industry question and answer sample training set to obtain the large language model.

16. The seed industry information question and answer method according to claim 9, wherein: The large language model is used together with at least one of the following: a knowledge graph, a rule engine and an expert system, and a data warehouse and online analytical processing technology.

17. A machine-readable medium comprising machine-executable instructions stored thereon, which instructions, when executed, cause a machine to perform the method of any one of claims 9 to 16.

18. A computer program product comprising computer instructions which, when executed, implement the method of any one of claims 9 to 16.

Images