Intelligent retrieval method and device, electronic equipment and computer storage medium

Abstract

The invention relates to an intelligent retrieval method and device, electronic equipment and a computer storage medium, and belongs to the technical field of intelligent retrieve.The intelligent retrieval method comprises the steps that multi-level title texts of structured documents in a preset document library and content texts under the title texts are obtained; constructing a knowledge graph based on the father-son affiliation relationship of the title corresponding to each title text in the structured text, wherein the knowledge graph is used for indicating the father-son affiliation relationship of each title; semantic information of a natural retrieval language is obtained, an initial title text matched with the semantic information is determined in the knowledge graph based on the semantic information, and an associated title text of the initial title text is obtained; and determining the title text and the content text under the associated title text as a retrieval result corresponding to the natural retrieval language. According to the method, the retrieval accuracy and integrity can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent retrieval technology, and in particular to an intelligent retrieval method, device, electronic device, and computer storage medium. Background Technology

[0002] When building a Retrieval Enhanced Generation (RAG) system, especially when processing technical documents, legal provisions, or reports with a clear hierarchical structure, a commonly used text chunking strategy is structured chunking. This strategy follows the original chapter and heading structure of the document, dividing it into segments and storing the final heading and its corresponding content as an independent semantic unit (i.e., a knowledge block) in a vector database. The initial intention of this method is to preserve the integrity of the local context to the greatest extent possible. When a user's query precisely hits a specific knowledge point, the system can return highly relevant text blocks with minimal interfering information, thereby generating an accurate answer.

[0003] However, this static and fine-grained chunking strategy exposes its inherent flaws when dealing with macro-level, general user queries. For example, when a user asks "Please summarize the core functions of XX service," the answer may be widely distributed across multiple subheadings under the main title of that service. Existing RAG systems, during the retrieval phase, typically only recall a few text chunks most similar to the query vector, resulting in severely incomplete contextual information provided to the subsequent Large Language Model (LLM). This "context fragmentation" problem directly leads to the final generated answer containing missing key information, logical jumps, and factual errors, failing to form a comprehensive and coherent summary.

[0004] This shows that there is an irreconcilable contradiction between existing technologies in ensuring the "accuracy" of queries and the "completeness" of answers. Summary of the Invention

[0005] In view of this, it is necessary to provide an intelligent retrieval method, device, electronic device, and computer storage medium to solve the problem of the irreconcilable contradiction between ensuring the "accuracy" of the query and the "completeness" of the answer in the existing technology.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides an intelligent retrieval method, comprising: Retrieve the multi-level heading text and the content text under each heading text from the predefined document library of structured documents; A knowledge graph is constructed based on the parent-child relationship of the titles corresponding to each title text in the structured document, and the knowledge graph is used to indicate the parent-child relationship of each title. Obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The title text and the content text under the associated title text are determined as the search results corresponding to the natural search language.

[0007] In one possible implementation, obtaining the multi-level heading text and the content text under each heading text of a structured document in a preset document library includes: The system uses a pre-defined automated script to read multiple structured documents from a pre-defined document library. The multiple structured documents are parsed to obtain the multi-level heading text and the content text under each heading text in each structured document.

[0008] In one possible implementation, the construction of a knowledge graph based on the parent-child relationship between the titles corresponding to each of the title texts in the structured document, wherein the knowledge graph is used to indicate the parent-child relationship between each of the titles, includes: Independent nodes are constructed based on each of the title texts, and directed edges are constructed between each of the independent nodes based on the parent-child relationship between them. The directed edges are used to point from the parent node to the child node. A knowledge graph is constructed based on each of the independent nodes and the directed edges between them.

[0009] In one possible implementation, determining the starting title text matching the semantic information in the knowledge graph based on the semantic information includes: The semantic information is matched with keywords in each of the title texts; If the keyword match is successful, the title text of the successful keyword match will be used as the starting title text; If keyword matching fails, the vector similarity between the semantic information and each of the title texts is calculated, and the title text with the highest vector similarity is determined as the starting title text.

[0010] In one possible implementation, obtaining the associated title text of the starting title text includes: Based on the starting title text, a graph traversal algorithm is used to traverse the knowledge graph to obtain the descendant nodes under the independent node corresponding to the starting title text; The title text corresponding to the descendant node is determined as the associated title text related to the starting title text.

[0011] In one possible implementation, determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The initial title text and the text content under each of the associated title texts are aggregated according to the order in which the corresponding title texts appear in the structured document to generate a search document.

[0012] In one possible implementation, determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The large language model is used to analyze the retrieved documents and the natural search language to determine the search results.

[0013] Secondly, the present invention also provides an intelligent retrieval device, comprising: The text acquisition module is used to acquire the multi-level heading text and the content text under each heading text of structured documents in the preset document library; A knowledge graph construction module is used to construct a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, wherein the knowledge graph is used to indicate the parent-child relationship of each title; The title determination module is used to obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The retrieval module is used to determine the title text and the content text under the associated title text as the retrieval results corresponding to the natural search language.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the intelligent retrieval method described in any of the above embodiments.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps of the intelligent retrieval method described in any of the above embodiments.

[0016] The beneficial effects of this invention are as follows: The intelligent retrieval method provided by this invention obtains multi-level title text and content text under each title text of a structured document in a preset document library; constructs a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, the knowledge graph being used to indicate the parent-child relationship of each title; obtains semantic information of natural search language, determines the starting title text matching the semantic information in the knowledge graph based on the semantic information, and obtains the associated title text of the starting title text; and determines the title text and the content text under the associated title text as the retrieval result corresponding to the natural search language. By constructing a knowledge graph of documents and combining it with a custom graph traversal retrieval algorithm, dynamic aggregation of context is achieved, enabling the question-answering system to accurately answer detailed questions by locating individual nodes, and to comprehensively and systematically answer general questions by traversing subgraphs, thereby simultaneously improving the accuracy and completeness of knowledge retrieval. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an intelligent retrieval method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a text content acquisition method provided in an embodiment of the present invention; Figure 3 A flowchart illustrating a knowledge graph construction method provided in an embodiment of the present invention; Figure 4 A flowchart illustrating a title matching method provided in an embodiment of the present invention; Figure 5 A flowchart illustrating an associated title matching method provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an intelligent retrieval device provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] A specific embodiment of the present invention, such as Figure 1 As shown, an intelligent retrieval method is disclosed, including: S101, retrieve the multi-level heading text and the content text under each heading text of the structured document in the preset document library; S102, construct a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, the knowledge graph being used to indicate the parent-child relationship of each title; S103, obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; S104, the title text and the content text under the associated title text are determined as the search results corresponding to the natural search language.

[0022] In this embodiment of the invention, the provided intelligent retrieval method is used to perform document retrieval using natural language search, generating accurate and complete retrieval results. The preset document library refers to a document database targeted by natural language search, used to store pre-set documents. A structured document refers to a document system with logical structures such as titles, chapters, and paragraphs. Content is organized through framework construction, similar to the process of writing an outline, making each part of the content form a closely related whole. Structured documents follow specific construction principles, emphasizing hierarchical logical relationships. For each structured document, the content will contain multi-level headings, and each heading will have corresponding content. Multi-level heading text refers to the text content of each level of heading, and the content text under each heading text refers to the specific text content recorded under each heading.

[0023] In this embodiment of the invention, the knowledge graph can represent the parent-child relationship between each title using nodes and the parent-child relationship between nodes. By generating corresponding independent nodes based on the title text at each level, and determining the parent-child relationship between the independent nodes corresponding to each title text based on the parent-child relationship between the titles corresponding to each level of title text, a knowledge graph is constructed.

[0024] In this embodiment of the invention, after constructing the knowledge graph, semantic information of natural search language for a preset document library is obtained. This semantic information can be determined using existing semantic recognition models. Then, based on this semantic information, the corresponding title text and associated title text are searched in the knowledge graph. The content text under the title text and the associated title text are determined as the search results corresponding to the natural search language.

[0025] The intelligent retrieval method provided by this invention obtains multi-level title text and content text under each title text from a pre-set document library; constructs a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, the knowledge graph indicating the parent-child relationship of each title; obtains semantic information of natural search language, determines the starting title text matching the semantic information in the knowledge graph based on the semantic information, and obtains the associated title text of the starting title text; and determines the title text and the content text under the associated title text as the retrieval result corresponding to the natural search language. By constructing a knowledge graph of documents and combining it with a custom graph traversal retrieval algorithm, dynamic aggregation of context is achieved, enabling the question-answering system to accurately answer detailed questions by locating individual nodes and comprehensively and systematically answer general questions by traversing subgraphs, thereby simultaneously improving the accuracy and completeness of knowledge retrieval.

[0026] In some possible embodiments of the present invention, such as Figure 2 As shown, the step of obtaining the multi-level heading text and the content text under each heading text of a structured document in a preset document library includes: S201, uses a preset automated script program to read multiple structured documents from a preset document library; S202, parse the multiple structured documents to obtain the multi-level heading text and the content text under each heading text in each structured document.

[0027] In this embodiment of the invention, an automated script program can be used to read the original structured document (such as a Word or Markdown file), parse out its complete multi-level heading system and the body text content corresponding to each heading.

[0028] Furthermore, such as Figure 3 As shown, the knowledge graph is constructed based on the parent-child relationship between the titles corresponding to each title text in the structured document. The knowledge graph is used to indicate the parent-child relationship between each title, including: S301, construct independent nodes based on each of the title texts, and construct directed edges between each of the independent nodes based on the parent-child relationship between each of the independent nodes, wherein the directed edges are used to point from the parent node to the child node; S302, Construct a knowledge graph based on each of the independent nodes and the directed edges between each of the independent nodes.

[0029] In this embodiment of the invention, each parsed title (regardless of level) is created as an independent node in the graph. Each node contains at least two core attributes: title (title text) and content (body content under that title). Directed edges are established between nodes based on the parent-child relationships between titles. For example, if "1.1 Cockpit Factory" is a subtitle of "1. Component Engine," a directed edge is created from the node representing "Component Engine" pointing to the "Cockpit Factory" node. This edge can be defined as HAS_SUBSECTION (having a subsection). The complete graph data structure containing all nodes and edges is persistently stored in a graph database (such as Neo4j or NebulaGraph) for efficient retrieval during the online search phase.

[0030] Furthermore, such as Figure 4 As shown, determining the starting title text that matches the semantic information in the knowledge graph based on the semantic information includes: S401, perform keyword matching between the semantic information and each of the title texts; S402, If the keyword match is successful, the title text of the successful keyword match shall be used as the starting title text; S403, if keyword matching fails, calculate the vector similarity between the semantic information and each of the title texts, and determine the title text with the highest vector similarity as the starting title text.

[0031] In this embodiment of the invention, the question-answering system front-end receives a user's natural language query, which includes natural language retrieval language. A semantic recognition model identifies the semantic information of this natural language retrieval language. Based on this semantic information, the system locates the "starting node" in the knowledge graph that best matches the semantic information, thereby determining the starting title text. Specifically, the location method can employ a hybrid strategy: firstly, it performs precise or fuzzy keyword matching on the title attribute of all nodes. If a matching result exists, the title text matching the semantic information is determined as the starting title text; if no matching result exists, it performs vector similarity search on the content attribute of all nodes to find the most relevant starting node and determine the starting title text.

[0032] Furthermore, such as Figure 5 As shown, obtaining the associated title text of the starting title text includes: S501, based on the starting title text, a graph traversal algorithm is used to traverse the knowledge graph to obtain the descendant nodes under the independent node corresponding to the starting title text; S502, the title text corresponding to the descendant node is determined as the associated title text associated with the starting title text.

[0033] In this embodiment of the invention, after determining the starting title text, the knowledge graph can be traversed based on the starting title text. A graph traversal algorithm is used to recursively search and visit all direct and indirect descendant nodes under the starting node along the edge of type HAS_SUBSECTION, thereby forming a complete "subgraph".

[0034] Furthermore, determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The initial title text and the text content under each of the associated title texts are aggregated according to the order in which the corresponding title texts appear in the structured document to generate a search document.

[0035] In this embodiment of the invention, the contents (content attribute) of the "starting node" and all the nodes traversed in the subgraph are concatenated according to their order in the original document and dynamically aggregated into a logically coherent and information-complete long text string, which serves as the final context.

[0036] Furthermore, determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The large language model is used to analyze the retrieved documents and the natural search language to determine the search results.

[0037] In this embodiment of the invention, the complete context dynamically aggregated in the foregoing embodiments, along with the user's original query, is submitted to a large language model (LLM). The LLM performs understanding and reasoning based on this comprehensive and uninterrupted background knowledge, generates the final answer, and presents it to the user through the question-answering system front-end.

[0038] This invention achieves a leap from "retrieving isolated text blocks" to "retrieving a complete logical subtree," fundamentally solving the problem of incomplete context under macro-level questions. 1. For detailed questions, precise answers can be achieved by locating individual nodes; for macro-level questions, complete context can be dynamically aggregated by traversing the subgraph to generate comprehensive answers, resolving the contradiction between the two. 3. The retrieval logic is controlled by a custom service, which can easily add more complex traversal rules (such as handling "see also," "reference," etc.) or integrate multi-document knowledge, without being constrained by specific platform functions. Transforming documents into a knowledge graph not only preserves the hierarchical structure but also explicitly defines the complex relationships between different knowledge units, laying a solid foundation for future higher-level knowledge reasoning and applications.

[0039] To better implement the intelligent retrieval method in the embodiments of the present invention, based on the intelligent retrieval method, correspondingly, as follows: Figure 6 As shown, this embodiment of the invention also provides an intelligent retrieval device, the intelligent retrieval device 600 including: The text acquisition module 601 is used to acquire the multi-level heading text and the content text under each heading text of the structured document in the preset document library; The knowledge graph construction module 602 is used to construct a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, wherein the knowledge graph is used to indicate the parent-child relationship of each title. The title determination module 603 is used to obtain semantic information of natural retrieval language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The retrieval module 604 is used to determine the title text and the content text under the associated title text as the retrieval results corresponding to the natural retrieval language.

[0040] The intelligent retrieval device 600 provided in the above embodiments can implement the technical solutions described in the above intelligent retrieval method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above intelligent retrieval method embodiments, and will not be repeated here.

[0041] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702, and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0042] In some embodiments, processor 701 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 702 or process data, such as the intelligent retrieval method of the present invention.

[0043] In some embodiments, processor 701 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 701 may be local or remote. In some embodiments, processor 701 may be implemented on a cloud platform. In some embodiments, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, or any combination thereof.

[0044] In some embodiments, memory 702 may be an internal storage unit of electronic device 700, such as a hard disk or memory of electronic device 700. In other embodiments, memory 702 may also be an external storage device of electronic device 700, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 700.

[0045] Furthermore, the memory 702 may include both internal storage units of the electronic device 700 and external storage devices. The memory 702 is used to store application software and various types of data installed on the electronic device 700.

[0046] In some embodiments, display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 703 is used to display information from electronic device 700 and to display a visual user interface. Components 701-703 of electronic device 700 communicate with each other via a system bus.

[0047] In some embodiments, when the processor 701 executes the intelligent retrieval program in the memory 702, the following steps may be performed: Retrieve the multi-level heading text and the content text under each heading text from the predefined document library of structured documents; A knowledge graph is constructed based on the parent-child relationship of the titles corresponding to each title text in the structured document, and the knowledge graph is used to indicate the parent-child relationship of each title. Obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The title text and the content text under the associated title text are determined as the search results corresponding to the natural search language.

[0048] It should be understood that when the processor 701 executes the intelligent retrieval program in the memory 702, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0049] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 700 mentioned. Electronic device 700 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 700 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0050] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the intelligent retrieval methods provided in the above-described method embodiments.

[0051] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0052] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent retrieval method, characterized in that, include: Retrieve the multi-level heading text and the content text under each heading text from the predefined document library of structured documents; A knowledge graph is constructed based on the parent-child relationship of the titles corresponding to each title text in the structured document, and the knowledge graph is used to indicate the parent-child relationship of each title. Obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The title text and the content text under the associated title text are determined as the search results corresponding to the natural search language.

2. The intelligent retrieval method according to claim 1, characterized in that, The step of obtaining the multi-level heading text and the content text under each heading text of a structured document in a preset document library includes: The system uses a pre-defined automated script to read multiple structured documents from a pre-defined document library. The multiple structured documents are parsed to obtain the multi-level heading text and the content text under each heading text in each structured document.

3. The intelligent retrieval method according to claim 1, characterized in that, The knowledge graph is constructed based on the parent-child relationship between the titles corresponding to each title text in the structured document. The knowledge graph is used to indicate the parent-child relationship between each title, including: Independent nodes are constructed based on each of the title texts, and directed edges are constructed between each of the independent nodes based on the parent-child relationship between them. The directed edges are used to point from the parent node to the child node. A knowledge graph is constructed based on each of the independent nodes and the directed edges between them.

4. The intelligent retrieval method according to claim 3, characterized in that, The step of determining the starting title text that matches the semantic information in the knowledge graph based on the semantic information includes: The semantic information is matched with keywords in each of the title texts; If the keyword match is successful, the title text of the successful keyword match will be used as the starting title text; If keyword matching fails, the vector similarity between the semantic information and each of the title texts is calculated, and the title text with the highest vector similarity is determined as the starting title text.

5. The intelligent retrieval method according to claim 4, characterized in that, The step of obtaining the associated title text of the starting title text includes: Based on the starting title text, a graph traversal algorithm is used to traverse the knowledge graph to obtain the descendant nodes under the independent node corresponding to the starting title text; The title text corresponding to the descendant node is determined as the associated title text related to the starting title text.

6. The intelligent retrieval method according to claim 5, characterized in that, The step of determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The initial title text and the text content under each of the associated title texts are aggregated according to the order in which the corresponding title texts appear in the structured document to generate a search document.

7. The intelligent retrieval method according to claim 6, characterized in that, The step of determining the title text and the content text under the associated title text as the search results corresponding to the natural search language includes: The large language model is used to analyze the retrieved documents and the natural search language to determine the search results.

8. An intelligent retrieval device, characterized in that, include: The text acquisition module is used to acquire the multi-level heading text and the content text under each heading text of structured documents in the preset document library; A knowledge graph construction module is used to construct a knowledge graph based on the parent-child relationship of the titles corresponding to each title text in the structured document, wherein the knowledge graph is used to indicate the parent-child relationship of each title; The title determination module is used to obtain semantic information of natural search language, determine the starting title text that matches the semantic information in the knowledge graph based on the semantic information, and obtain the associated title text of the starting title text; The retrieval module is used to determine the title text and the content text under the associated title text as the retrieval results corresponding to the natural search language.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps of the intelligent retrieval method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps of the intelligent retrieval method according to any one of claims 1 to 7.

Images