Plant knowledge retrieval enhancement system and method based on data fusion

Abstract

The invention relates to the technical field of plant knowledge retrieval, in particular to a plant knowledge retrieval enhancement system and method based on data fusion. Comprising a plant text processing module, a user input module, an input preprocessing module, a model input processing module, a model output processing module, a parallel retrieval module, a data fusion module and an output result module. Constructing the relational database, the graph database and the vectorization knowledge base; receiving a natural language query input by a user, and generating a structured query statement and question classification information through a large language model; retrieving a relational database, a graph database and a vectorization knowledge base in parallel based on the structured query statement to obtain an intermediate retrieval result; and carrying out data fusion processing on the intermediate retrieval result, sorting according to the occurrence frequency of the record identifier, and retrieving from the original database and outputting a final result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flora knowledge retrieval, and more particularly to a flora knowledge retrieval enhancement system and method based on data fusion. BACKGROUND

[0002] The flora mainly records the classification information, description information, distribution information and use information of plants. Since the description information, distribution information and use information of plants are mainly in the form of unstructured text content, the traditional flora knowledge retrieval method mainly retrieves according to the plant classification information with high structured degree, such as inputting the Chinese name or Latin name of a plant to retrieve the information of a certain type or certain plant.

[0003] The large language model has been very mature in natural language processing ability after years of rapid development, and has high accuracy in natural language understanding, structured data extraction and formatted result output. Based on the large language model, database, knowledge graph and knowledge base technologies, the present application establishes a work flow for structuring, organizing, processing, storing and knowledge retrieval of flora data, constructs a structured relational database, a graph database based on a knowledge graph and a vectorized knowledge base, thereby developing and implementing a flora knowledge retrieval enhancement system based on data fusion. Compared with the traditional flora knowledge retrieval method, the retrieval enhancement system has a more flexible retrieval method, such as inputting any natural language for retrieval, and the retrieval result has high recall rate and accuracy. SUMMARY

[0004] Therefore, the present application provides a flora knowledge retrieval enhancement system and method based on data fusion, aiming to solve the above technical problems.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0006] A flora knowledge retrieval enhancement system based on data fusion, comprising: a flora processing module, a user input module, an input preprocessing module, a model input processing module, a model output processing module, a parallel retrieval module, a data fusion module and an output result module;

[0007] The flora processing module is used to structure the data in the flora original database and to finely segment and reorganize the data, so as to construct a relational database, a graph database based on a knowledge graph and a vectorized knowledge base for data parallel retrieval by the parallel retrieval module;

[0008] The user input module is used to receive any natural language input by a user;

[0009] The input preprocessing module is configured to preprocess the user input, and the preprocessing step includes annotation and field / attribute similarity matching.

[0010] The model input processing module is configured to output corresponding SQL query statements, Cypher query statements, and question classification information based on the user input annotated by the input preprocessing module, the field / attribute similarity matching result, and the prompt word prepared and debugged by the model input processing module.

[0011] The model output processing module is configured to process the SQL query statements, Cypher query statements, and question classification information output by the model input processing module.

[0012] The parallel retrieval module is configured to retrieve the relational database constructed by the flora processing module based on the SQL query statement output by the model output processing module, retrieve the graph database constructed by the flora processing module based on the Cypher query statement output by the model output processing module, and retrieve the target vectorized knowledge base output by the model output processing module based on the user input of the user input module. After parallel retrieval, the retrieval results from the above three different data sources are output.

[0013] The data fusion module is configured to fuse the results of the relational database, the graph database, and the vectorized knowledge base to generate a plant record list.

[0014] The output result module is configured to return the plant record list output by the data fusion module as the final query result to the user.

[0015] A flora knowledge retrieval enhancement method based on data fusion, comprising the following steps:

[0016] S1, structuring the unstructured text of the original database by a large language model to construct the relational database, the graph database, and the vectorized knowledge base;

[0017] S2, receiving the natural language query input by the user, and generating a structured query statement and question classification information by a large language model;

[0018] S3, retrieving the relational database, the graph database, and the vectorized knowledge base in parallel based on the structured query statement to obtain intermediate retrieval results;

[0019] S4, performing data fusion processing on the intermediate retrieval results, sorting according to the number of record identifiers, retrieving and outputting the final results from the original database.

[0020] Further, the structuring of the unstructured text of the original database by the large language model comprises:

[0021] S11, the large language model extracts plant description information, distribution information and use information according to attribute value rules to generate structured fields;

[0022] S12, the system unifies the units of numerical data and determines the range to form structured data of the relational database;

[0023] S13, the large language model stores plant classification information in the graph database node according to the genus and species hierarchy, and extracts attribute value relationships and stores them in the corresponding node;

[0024] S14, the system vectorizes the description information, distribution information and use information according to multi-granularity segments to generate a vectorized knowledge base containing and not containing classification information.

[0025] Further, the vectorization of the description information, distribution information and use information according to multi-granularity segments includes:

[0026] S141, the system vectorizes the description information, distribution information and use information according to fine-grained segments to generate a fine-grained knowledge base covering whole plant morphology, root morphology, stem morphology, leaf morphology, flower morphology, fruit morphology, seed morphology, domestic distribution, foreign distribution, growth altitude, living environment, type specimen origin, food use, medicinal use, industrial use, ornamental use, ecological protection use;

[0027] S142, the system vectorizes the description information, distribution information and use information according to coarse-grained segments to generate a coarse-grained knowledge base of plant description, distribution and use;

[0028] S143, the system vectorizes the description information, distribution information and use information according to combined segments to generate a combined knowledge base of description distribution, description use and distribution use;

[0029] S144, the system adds classification information to the fine-grained, coarse-grained and combined knowledge bases to generate a knowledge base containing classification information.

[0030] Further, the generation of structured query statements and problem classification information by the large language model includes:

[0031] S21, the system performs annotation processing on the natural language query input by the user to identify plant classification information and geographical distribution information;

[0032] S22, the system performs similarity matching between the annotated input and the relational database fields and the graph database attributes to generate a matching result;

[0033] S23, the large language model generates SQL query statements, Cypher query statements and question classification information according to the matching result and the preset prompt word;

[0034] S24, the system normalizes and checks the security of the SQL query statement and the Cypher query statement.

[0035] Further, the utilization of the structured query statement to retrieve the relational database, the graph database and the vectorized knowledge base in parallel comprises:

[0036] S31, the system retrieves the relational database using the processed SQL query statement to obtain structured data results;

[0037] S32, the system retrieves the graph database using the processed Cypher query statement to obtain knowledge graph node and relationship results;

[0038] S33, the system determines the target vectorized knowledge base according to the question classification information, retrieves the target vectorized knowledge base using user input, and obtains semantic retrieval results.

[0039] Further, the data fusion processing of the intermediate retrieval result according to the occurrence frequency of the record identifier comprises:

[0040] S41, the system counts the occurrence frequency of the record identifier in the retrieval results of the relational database, the graph database and the vectorized knowledge base;

[0041] S42, the system sorts the record identifier list in descending order according to the occurrence frequency to generate a sorted record identifier list;

[0042] S43, the system extracts the corresponding plant record from the original database as the final query result according to the record identifier list.

[0043] Further, the extraction of the corresponding plant record from the original database according to the record identifier list comprises:

[0044] S431, the system queries the plant record in the original database according to the sorted record identifier;

[0045] S432, the system formats the queried plant record to generate a plant record list containing classification information, description information, distribution information and use information;

[0046] S433, the system returns the plant record list as the final query result to the user.

[0047] Compared with the prior art, the present application extracts the text content of the original database of the flora by using a large language model according to different categories and attributes, and statistically analyzes the extraction results, and determines the table structure of the relational database in combination with the artificial review of a botany expert; constructs a knowledge graph according to the classification information and attribute extraction results of the plant and stores it in a graph database; and quantifies the knowledge base according to the extracted category content in different granularity or fragment combinations. When the system receives user input, the input is first labeled, and then the labeled user input is similarity matched with the fields of the relational database and the attributes of the graph database, and then the user input and the field / attribute with higher matching degree are input into the large language model, and the large language model generates and outputs SQL query statements, Cypher query statements and question classification information through prompt word engineering, processes the above outputs, and then retrieves the relational database, the graph database and the knowledge base to obtain multiple intermediate result data, and then retrieves the content of the original database again after fusion analysis of the multiple intermediate results and outputs the retrieval result to the user, so as to realize the retrieval of flora knowledge in a more flexible natural language manner, and ensure that the retrieval result has high recall rate and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only a part of the embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0049] Figure 1 The structure diagram of the vectorized knowledge base hierarchical architecture of the present application.

[0050] Figure 2 The flora knowledge retrieval enhancement method of the present application.

[0051] Figure 3 The flora knowledge retrieval enhancement system of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0053] Referring to the accompanying Figures 1-3 The flora knowledge retrieval enhancement system and method based on data fusion are disclosed in the embodiments of the present application, as shown inFigure 3 As shown, the retrieval enhancement system includes a flora processing module, a user input module, an input preprocessing module, a model input processing module, a model output processing module, a parallel retrieval module, a data fusion module, and an output result module, etc.

[0054] The main process is as shown in Figure 2 As shown, it includes:

[0055] Step 1: Based on the large language model, the text content of the original database of the flora is extracted and analyzed, and a relational database, a graph database based on knowledge graph, and a vectorized knowledge base are constructed according to the prescribed process and organizational structure;

[0056] Step 2: The retrieval enhancement system receives the user's input, and labels the user input, and the similarity matching between the labeled input and the fields / attributes of the database;

[0057] Step 3: The user input with labels, field / attribute similarity matching results, and prompt words are input into the large language model, and the output of the large language model is standardized and safety checked, etc.

[0058] Step 4: Based on the results of the previous processing, the relational database, graph database and knowledge base are retrieved in parallel, and the retrieval results (intermediate results) are returned;

[0059] Step 5: The intermediate result data returned in the previous step is data fused (merged, counted and sorted), and based on the data fusion result, the flora original database is retrieved again, and finally the final retrieval result is returned.

[0060] The retrieval enhancement system can preprocess the natural language input by the user, and generate data query statements and problem classification information based on the large language model, so as to realize the retrieval of the relational database, graph database and vectorized knowledge base, and after the fusion processing of the retrieval result, the plant information is retrieved from the flora original database and returned to the user, so as to not only realize the retrieval of flora knowledge through more flexible natural language, but also ensure that the retrieval enhancement system has higher recall rate and accuracy.

[0061] The method mainly includes the following data processing modules:

[0062] Flora data processing module. The organization and processing of flora data is the basis for ensuring the high accuracy of the retrieval enhancement system. The flora book mainly records the classification information, description information, distribution information and use information of plants, etc. The structure of the original flora database can be generally described as: id, volume, Chinese name, Latin name, Chinese alias, species classification, line drawing picture address, plant description information, plant distribution information, plant use information, etc. Among them, the classification information of plants (such as: id, volume, Chinese name, Latin name, Chinese alias, species classification, line drawing picture address) is structured data, while the description information, distribution information and use information of plants are stored in unstructured text content. Therefore, in this module, the data in the original flora database needs to be further structured and more granularly segmented and reorganized to build a relational database for parallel retrieval module, a graph database based on knowledge graph and a vectorized knowledge base for data parallel retrieval.

[0063] (1) Construction of relational database: The construction of relational database aims to make the structure granularity of original flora database finer, so as to realize more accurate data retrieval. In this module, the description information, distribution information and use information of plants are mainly processed:

[0064] For the description information of plants, the description information of all plants is structured and extracted according to the "attribute:value" rule by using a large language model, such as the extraction result may output "plant height: 50-80 cm" and the like. The number of occurrences of each attribute is counted according to the seven categories of whole plant, root, stem, leaf, flower, fruit and seed, and the attributes with more occurrences in each category are manually reviewed by botany experts to determine whether the attributes belong to the structured field list of plant description information. Finally, the structured fields of plant description information are determined, and the values corresponding to the determined fields are extracted from the plant description information by using a large language model, so as to process the description information into more granular plant description structured data.

[0065] For the distribution information of plants, the domestic distribution, foreign distribution, growth altitude, living environment and type specimen origin of plants are extracted from the distribution information of plants by using a large language model.

[0066] For the use information of plants, the food use, medicinal use, industrial use, ornamental use, ecological protection use and other use information of plants are extracted from the use information of plants by using a large language model.

[0067] Through the above processing, all structured fields of the relational database can be determined (including classification information subordinate fields, description information subordinate fields, distribution information subordinate fields, and use information subordinate fields). In order to achieve more accurate and quantitative retrieval requirements, numerical data also needs to be further processed, and numerical data is processed to have a unified unit and to determine its minimum value and maximum value (for example, plant A "plant height: 1-2 meters" and plant B "plant height: 50-80 centimeters" are processed: the unit is unified to centimeters, the "plant height minimum value" of plant A is 100, the "plant height maximum value" of plant A is 200, the "plant height minimum value" of plant B is 50, and the "plant height maximum value" of plant B is 80), thereby completing the construction of the flora relational database.

[0068] (2) Construction of a graph database based on a knowledge graph: The construction of a graph database based on a knowledge graph aims to improve the ease of understanding of the flora hierarchy and the efficiency of data queries. The flora records the classification levels of plants such as families, genera, and species (including sub-levels). The flora includes x families, each family includes y genera, and each genus includes z species. Therefore, the flora is essentially a tree structure with the plant kingdom as the root directory, branches representing the relationships between families, genera, and species, and leaves representing certain plants. This is consistent with the knowledge graph in essence, which represents relationships and nodes. Therefore, plant names are stored in nodes in the graph database according to three categories of tags: families, genera, and species. Relationships between nodes with superior-inferior relationships are established. Large language models are used to structure the extraction of plant description information, distribution information, and use information according to the "attribute: value" rule. The extracted attributes and values are stored in the corresponding graph database nodes, thereby constructing a flora graph database based on a knowledge graph.

[0069] (3) Construction of a vectorized knowledge base: The vectorized knowledge base mainly implements semantic retrieval functions. The size of the knowledge fragments in the knowledge base directly affects the accuracy of the retrieval results. Therefore, in order to achieve more accurate vector retrieval, the method of the present patent application uses multiple knowledge fragments and fragment combinations with different granularities. In order to minimize the impact of classification information on the accuracy of vector retrieval, this knowledge base is divided into two categories: "classification information non-vectorized knowledge base" and "classification information vectorized knowledge base", totaling 50 vectorized knowledge bases.

[0070] Classification information not vectorization knowledge base: the classification information not vectorization knowledge base only vectorizes the description information, distribution information and use information of plants, and is divided into multiple different granularity and segment combinations for vectorization, wherein the fine-grained knowledge base includes 18 knowledge bases of whole plant morphology, root morphology, stem morphology, leaf morphology, flower morphology, fruit morphology, seed morphology, domestic distribution, foreign distribution, growth altitude, living environment, type specimen origin, food use, medicinal use, industrial use, ornamental use, ecological protection use and other uses; the coarse-grained knowledge base includes 3 knowledge bases of plant description, plant distribution and plant use; and the combined knowledge base includes 4 knowledge bases of description distribution, description use, distribution use and description distribution use.

[0071] Classification information vectorization knowledge base: the classification information vectorization knowledge base not only vectorizes the description information, distribution information and use information of plants, but also vectorizes the classification information, so on the basis of the above-mentioned "classification information not vectorization knowledge base", the classification information (such as id, volume, Chinese name, Latin name, Chinese alias, species classification, Chinese name of family, Latin name of family, Chinese name of genus, Latin name of genus, Chinese name of species, Latin name of species and line drawing picture address) is added to the above-mentioned each fine-grained knowledge base, coarse-grained knowledge base and combined knowledge base and vectorized to construct 25 new vectorization knowledge bases.

[0072] User input module. The user input mainly receives any natural language input by the user, and is the query entrance of the flora knowledge retrieval enhancement system.

[0073] Input preprocessing module. The input preprocessing mainly pre-processes the input of the user, and the pre-processing steps include annotation and field / attribute similarity matching, wherein the annotation processing includes: if the plant name such as Chinese name of family, Latin name of family, Chinese name of genus, Latin name of genus, Chinese name of species and Latin name of species appears in the input of the user, the corresponding attribute is annotated after the plant name; if the name of China and the province, city, district and county in China appears in the input of the user, "domestic distribution" is annotated after the geographical name; and if the name of a country other than China appears in the input of the user, "foreign distribution" is annotated after the name of the country. The field / attribute similarity matching mainly performs similarity matching between the annotated input of the user and the fields of the relational database and the attributes of the graph database, and temporarily stores the fields and attributes with similarity in the Top 20, which are used for model input processing module processing.

[0074] Model input processing module. Based on the large language model to further understand the intention of the user input, therefore here is mainly to write a satisfactory prompt word, the input pre-processing module labeled user input, field / attribute similarity matching results and write and debug the prompt word input to the large language model, let the large language model output corresponding SQL query statement, Cypher query statement and question classification information, question classification information is in the corresponding category (whole plant morphology, root morphology, stem morphology, leaf morphology, flower morphology, fruit morphology, seed morphology, domestic distribution, foreign distribution, growth altitude, living environment, type specimen origin, food use, medicinal use, industrial use, ornamental use, ecological protection use, other use, plant description, plant distribution, plant use, description distribution, description use, distribution use, description distribution use) involved in the user input is marked. In order to avoid a single large language model to understand the intention is not sufficient to lead to empty output, model input processing module simultaneously call two different large language model synchronous processing.

[0075] Model output processing module. The SQL query statement, Cypher query statement and question classification information output by the model input processing module large language model are processed, wherein the processing of SQL query statement and Cypher query statement is mainly to standardize the query statement and check and process the safety, in order to avoid the exposure of real database information to the outside, the custom name / field of the query statement needs to be replaced, in order to avoid the database being maliciously deleted and tampered due to user input, etc. The processing of question classification information is mainly to determine the vectorized knowledge base to be queried. The steps are as follows: first, determine whether the classification information appears in the user input, so as to determine whether to use the "classification information non-vectorized knowledge base" of the input pre-processing module or the "classification information vectorized knowledge base"; second, select the appropriate granularity knowledge base or combined knowledge base according to the question classification information returned by the large language model. The selection principle is: if the question classification only involves a single category, the category is used as the target vectorized knowledge base for query; if the question classification involves multiple categories, their nearest common superior is used as the target vectorized knowledge base (for example, if the question classification involves "leaf morphology" and "growth altitude", the target vectorized knowledge base for query is "description distribution", as shown in Figure 1

[0076] ​Parallel retrieval module. In order to save data retrieval time, the following parallel processing is adopted: using the SQL query statement processed by the model output processing module to retrieve the relational database constructed by the flora processing module, using the Cypher query statement processed by the model output processing module to retrieve the graph database constructed by the flora processing module, and using the user input of the user input module to retrieve the target vectorized knowledge base determined by the model output processing module, and then outputting the retrieval results from the above three different data sources in parallel.

[0077] Data fusion module. Based on the retrieval results (marked as Result_SQL, Result_Cypher and Result_VDB respectively) from the relational database, the graph database and the knowledge base output by the parallel retrieval module, since the data of the relational database, the graph database and the knowledge base all come from the flora original database Flora_DB, the same records have the same id. First, count the number of occurrences of each id in Result_SQL, Result_Cypher and Result_VDB, and sort them in descending order of occurrence frequency, and then retrieve the corresponding plant record list from the flora original database Flora_DB according to the id and output.

[0078] Output result module. The plant record list output by the data fusion module is returned to the user as the final query result.

[0079] The flora knowledge retrieval enhancement system and method based on data fusion of the present patent application is different from the traditional retrieval flora method, which can realize the retrieval of flora through natural language, and improve the recall rate and accuracy of the retrieval result through various data fusion technologies.

[0080] By using a large language model to extract the text content of the original database of the flora according to different categories and attributes, and statistically analyzing the extraction results, the table structure of the relational database is determined in combination with the artificial review of botany experts; the knowledge graph is constructed according to the classification information and attribute extraction results of the plants and stored in the graph database; and the category content is combined and quantified into a knowledge base according to different granularities or fragments. When the system receives user input, the input is first labeled, then the labeled user input is matched with the fields of the relational database and the attributes of the graph database, and then the user input and the field / attribute with higher matching degree are input into the large language model to generate and output SQL query statements, Cypher query statements and question classification information through prompt word engineering, and the above outputs are processed and then used to retrieve the relational database, the graph database and the knowledge base to obtain multiple intermediate result data, the multiple intermediate results are fused and analyzed, and then the content of the original database is retrieved again to output the retrieval results to the user, so as to realize the retrieval of flora knowledge in a more flexible natural language manner and ensure that the retrieval results have high recall rate and accuracy.

[0081] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0082] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A flora knowledge retrieval enhancement system based on data fusion, characterized in that, include: The system includes a flora processing module, a user input module, an input preprocessing module, a model input processing module, a model output processing module, a parallel retrieval module, a data fusion module, and an output results module. The flora processing module is used to structure, finely segment, and reorganize the data in the original flora database, thereby constructing a relational database, a knowledge graph-based graph database, and a vectorized knowledge base for parallel data retrieval by the parallel retrieval module. The user input module is used to receive user input in any natural language. The input preprocessing module is used to preprocess user input. The preprocessing steps include annotation and field / attribute similarity matching. The model input processing module is used to output corresponding SQL query statements, Cypher query statements, and question classification information based on user input labeled by the input preprocessing module, field / attribute similarity matching results, and written and debugged prompt words, through a large language model. The model output processing module is used to process the SQL query statements, Cypher query statements, and problem classification information output by the model input processing module; The parallel retrieval module is used to retrieve the relational database constructed by the Flora processing module based on the SQL query statement of the model output processing module, retrieve the graph database constructed by the Flora processing module based on the Cypher query statement of the model output processing module, and retrieve the target vectorized knowledge base of the model output processing module using the user input of the user input module. After the parallel retrieval is completed, the retrieval results from the above three different data sources are output. The data fusion module is used to merge the results from relational databases, graph databases, and vectorized knowledge bases to generate a list of plant records; The output results module is used to return the list of plant records output by the data fusion module as the final query result to the user.

2. A method for enhancing flora knowledge retrieval based on data fusion, applicable to the flora knowledge retrieval enhancement system based on data fusion as described in claim 1, characterized in that, Includes the following steps: S1. Using a large language model, unstructured text from the original database is extracted into a structured form, and the relational database, graph database, and vectorized knowledge base are constructed. S2. Receive natural language queries input by the user and generate structured query statements and question classification information through a large language model; S3. Based on the structured query statement, perform parallel retrieval of relational databases, graph databases, and vectorized knowledge bases to obtain intermediate retrieval results; S4. Perform data fusion processing on the intermediate search results, sort them according to the frequency of occurrence of the record identifier, retrieve from the original database and output the final result.

3. The method for enhancing flora knowledge retrieval based on data fusion according to claim 1, characterized in that, The method of extracting structured text from the original database using a large language model includes: S11. The large language model extracts plant description information, distribution information and use information according to attribute value rules to generate structured fields; S12. The system unifies the units and determines the range of numerical data to form structured data in a relational database; S13. The large language model stores plant classification information into graph database nodes according to family, genus and species hierarchy, and extracts attribute value relationships and stores them into corresponding nodes; S14. The system vectorizes the description information, distribution information, and usage information into multi-granularity segments to generate a vectorized knowledge base that includes and does not include classification information.

4. The method for enhancing flora knowledge retrieval based on data fusion according to claim 3, characterized in that, The vectorization of descriptive information, distribution information, and usage information into multi-granularity fragments includes: S141. The system vectorizes the description information, distribution information and use information into fine-grained segments to generate a fine-grained knowledge base covering whole plant morphology, root morphology, stem morphology, leaf morphology, flower morphology, fruit morphology, seed morphology, domestic distribution, foreign distribution, growth altitude, living environment, type specimen origin, food use, drug use, industrial use, ornamental use and ecological protection use. S142. The system vectorizes description information, distribution information and use information into coarse-grained segments to generate a coarse-grained knowledge base of plant description, distribution and use; S143. The system vectorizes the description information, distribution information and usage information into combined fragments to generate a combined knowledge base of description distribution, description usage and distribution usage. S144. The system adds classification information to the fine-grained, coarse-grained, and combined knowledge bases, and vectorizes them to generate a knowledge base containing classification information.

5. The method for enhancing flora knowledge retrieval based on data fusion according to claim 2, characterized in that, The generation of structured query statements and question classification information through a large language model includes: S21. The system annotates the natural language query input by the user and identifies plant classification information and geographical distribution information; S22. The system performs similarity matching between the labeled input and the fields of the relational database and the attributes of the graph database to generate matching results; S23. The large language model generates SQL query statements, Cypher query statements, and question classification information based on the matching results and preset prompt words; S24. The system performs normalization and security checks on the SQL query statement and Cypher query statement.

6. The method for enhancing flora knowledge retrieval based on data fusion according to claim 2, characterized in that, The parallel retrieval of relational databases, graph databases, and vectorized knowledge bases using the structured query statement includes: S31. The system uses the processed SQL query statement to retrieve the relational database and obtain structured data results; The system described in S32 uses the processed Cypher query statement to retrieve the graph database and obtain knowledge graph nodes and relationship results; S33. The system determines the target vectorized knowledge base based on the question classification information, retrieves the target vectorized knowledge base using user input, and obtains semantic retrieval results.

7. The method for enhancing flora knowledge retrieval based on data fusion according to claim 2, characterized in that, The data fusion processing of intermediate search results, and sorting them according to the frequency of occurrence of record identifiers, includes: S41. The system counts the number of times record identifiers appear in the retrieval results of the relational database, graph database, and vectorized knowledge base; S42. The system sorts the records in descending order according to their frequency of occurrence and generates a sorted list of record identifiers. S43. The system extracts the corresponding plant records from the original database as the final query result based on the record identifier list.

8. The method for enhancing flora knowledge retrieval based on data fusion according to claim 7, characterized in that, The step of retrieving the corresponding plant records from the original database based on the record identifier list includes: S431. The system queries plant records in the original database based on the sorted record identifiers; S432. The system formats the retrieved plant records to generate a list of plant records containing classification information, description information, distribution information and use information. S433, The system returns the plant record list to the user as the final query result.

Images