Intelligent retrieval method and system fusing knowledge graph and semantic vector
By integrating knowledge graphs and semantic vectors into an intelligent retrieval method, the limitations of existing RAG systems in complex concept associations and semantic understanding are overcome. This enables multi-dimensional, high-precision retrieval and refined sorting of text data, improving the comprehensiveness, accuracy, and relevance of retrieval results and adapting to ever-changing corpora.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FUJIAN STAR NET WISDOM TECH CO LTD
- Filing Date
- 2025-12-29
- Publication Date
- 2026-05-12
AI Technical Summary
Existing retrieval augmentation (RAG) systems have limitations in handling complex concept associations and multi-hop reasoning. They lack direct association and semantic understanding of the original text fragments, and the initial retrieval result set contains noise or inaccurate sorting, affecting the quality of the final generated answer.
An intelligent retrieval method integrating knowledge graphs and semantic vectors is proposed. By generating document summaries and constructing knowledge graphs through a large language model, and combining vector databases for multi-dimensional retrieval and fine-grained ranking of the reordering model, high-precision retrieval and fine-grained ranking of text data are achieved.
It significantly improves the comprehensiveness, accuracy, and relevance of search results, taking into account both the overall theme of the document and the detailed information of each fragment, and supports dynamic updates to adapt to constantly changing corpora.
Smart Images
Figure CN122019765A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of natural language processing and information retrieval technology, and in particular to an intelligent retrieval method and system that integrates knowledge graphs and semantic vectors. Background Technology
[0002] Existing Retrieval Augmentation (RAG) systems typically rely on vector similarity retrieval or keyword-based methods. While vector retrieval can capture semantic information, it has limitations in handling complex concept associations and multi-hop reasoning. On the other hand, traditional knowledge graphs, although capable of representing relationships between concepts, lack the ability to directly associate and semantically understand the original text fragments. Furthermore, the candidate result set obtained from the initial retrieval still has room for improvement in terms of relevance and accuracy, lacking a refined ranking mechanism to enhance the quality of the final returned results.
[0003] The shortcomings of existing technologies are at least as follows: Pure vector retrieval struggles to handle complex relationships between concepts; Keyword searches can easily miss semantically related content that uses different wording; Knowledge graph retrieval lacks direct links and semantic understanding of the original text fragments; The initial search results may contain noise or inaccurate sorting, affecting the quality of the final generated answer; the existing RAG system still has room for improvement in the accuracy and relevance ranking of search results.
[0004] Therefore, based on the above shortcomings, most existing searches are based on text fragments and lack an understanding of the overall theme and core ideas of the document. This may result in search results that are related in fragments but deviate from the main idea of the document. Summary of the Invention
[0005] The technical problem to be solved by this invention is to provide an intelligent retrieval method and system that integrates knowledge graphs and semantic vectors, so as to realize multi-dimensional and high-precision retrieval and fine sorting of text data, significantly improve the comprehensiveness, accuracy and relevance of retrieval results, and simultaneously take into account the overall theme of the document and the detailed information of the fragments.
[0006] In a first aspect, the present invention provides an intelligent retrieval method that integrates knowledge graphs and semantic vectors, comprising: Text preprocessing process: Import the document and segment the document text to obtain text segments; Document summary generation process: Generate an overall summary of the document text based on a large model; Knowledge graph construction process: Extract concept phrases from text segments and construct a knowledge graph to store the relationship between concept phrases and text segments; Vectorization process: The document summary and text segments are encoded into summary vectors and text segment vectors and stored in a vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs. The fusion retrieval process is used to retrieve text segments associated with the query statement through a knowledge graph, retrieve relevant summary vectors and text segment vectors through a vector database, and merge the retrieval results to form a preliminary candidate result set. Reordering process: Used to finely sort the initial candidate result set through a reordering model to obtain the final search results.
[0007] Furthermore, after the text preprocessing process, there is also an integrity judgment process, which is used to determine whether the text segment is complete based on the large model. If it is incomplete, the process returns to the text preprocessing process to re-segment the document text.
[0008] Furthermore, the knowledge graph construction process specifically includes: for the input text segment, using a large language model to extract key concept words, and constructing a knowledge link based on semantic association of the concept words, while assigning a unique link number to the knowledge link and establishing an association with the text segment; using the same concept nodes in multiple knowledge links as connection points, merging the knowledge links into a knowledge graph to form a semantic network.
[0009] Furthermore, in the fusion retrieval process, retrieving the text segments associated with the query statement through the knowledge graph specifically includes: extracting concept words from the query statement, matching the concept words in the knowledge graph to obtain relevant knowledge links, performing repetition rate statistics on the concept words contained in the matched knowledge links, sorting them by repetition rate, selecting the Top-N most relevant knowledge links, and then tracing back to the text segments associated with the knowledge links.
[0010] Secondly, the present invention provides an intelligent retrieval system that integrates knowledge graphs and semantic vectors, comprising: Text preprocessing module: used to import documents and segment document text to obtain text segments; Document summarization generation module: used to generate overall summaries of document text based on a large model; Knowledge graph construction module: used to extract concept phrases from text segments and construct a knowledge graph, storing the relationship between concept phrases and text segments; Vectorization module: used to encode document summaries and text segments into summary vectors and text segment vectors and store them in the vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs. The fusion retrieval module is used to retrieve text segments associated with query statements through knowledge graphs, retrieve relevant summary vectors and text segment vectors through vector databases, and merge the retrieval results to form a preliminary candidate result set. Reordering module: Used to finely sort the initial candidate result set through a reordering model to obtain the final search results.
[0011] Furthermore, following the text preprocessing module, there is also an integrity judgment module, which is used to determine whether the text segment is complete based on the large model. If it is incomplete, the text preprocessing module will re-segment the document text.
[0012] Furthermore, the knowledge graph construction module is specifically used for: extracting key concept words from the input text segment using a large language model, constructing a knowledge link based on semantic association of the concept words, assigning a unique link number to the knowledge link, and establishing an association with the text segment; using the same concept nodes in multiple knowledge links as connection points, merging the knowledge links into a knowledge graph to form a semantic network.
[0013] Furthermore, in the fusion retrieval module, retrieving the text segment associated with the query statement through the knowledge graph specifically includes: extracting concept words from the query statement, matching the concept words in the knowledge graph to obtain relevant knowledge links, performing repetition rate statistics on the concept words contained in the matched knowledge links, sorting them by repetition rate, selecting the Top-N most relevant knowledge links, and then tracing back to the text segment associated with the knowledge links.
[0014] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. By providing an enhanced RAG method and system that integrates knowledge graphs, vector retrieval, and reordering models, we can achieve multi-dimensional, high-precision retrieval and refined sorting of text data, significantly improving the comprehensiveness, accuracy, and relevance of retrieval results while taking into account both the overall theme of the document and the detailed information of each fragment.
[0015] 2. Enhanced search comprehensiveness: Combining conceptual association and semantic similarity avoids the limitations of a single search method; 3. High interpretability: Knowledge graphs provide visualization and explanation of retrieval paths; 4. More accurate ranking: The re-ranking model performs the final ranking of results based on deep semantic understanding, which significantly improves the quality of top results and is more suitable as the context for large models to generate answers; 5. Introduction of document-level semantic representation (summary): Effectively enhances the ability to retrieve the overall theme and core content of documents, avoids the search results from deviating from the main point, and is particularly suitable for query scenarios that require a macro-level understanding or filtering of documents, thereby improving the overall search depth and breadth of the system.
[0016] 6. Extremely high accuracy: Through completeness judgment, multi-dimensional retrieval and fine sorting of the reordering model, the relevance and accuracy of the results are guaranteed in three ways; 7. Good scalability: It supports dynamic updates of knowledge graphs and vector databases to adapt to constantly changing corpora.
[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0019] Figure 1 This is a flowchart illustrating the overall process of the method in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the fused knowledge graph structure in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the complete execution flow in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the system structure in Embodiment 2 of the present invention. Detailed Implementation
[0020] This application provides an intelligent retrieval method and system that integrates knowledge graphs and semantic vectors, enabling multi-dimensional, high-precision retrieval and refined sorting of text data. This significantly improves the comprehensiveness, accuracy, and relevance of retrieval results, while simultaneously taking into account both the overall theme of the document and the detailed information of individual segments.
[0021] The overall concept of the technical solution in this application is as follows: A large language model is used to generate a concise and accurate overall summary for each imported document, summarizing its core content. In addition to vectorizing the text segments, the generated document summary is also vectorized and stored in a vector database. The summary vector is stored in association with the text segment vectors of all the documents within it. When performing vector similarity retrieval, both the text segment vectors and the document summary vectors are queried simultaneously. This means that a document highly relevant to the query, even if its specific segments score poorly, may be included in the candidate set due to the relevance of its summary, thus better capturing document-level relevance. Through completeness judgment, multi-dimensional retrieval, and refined ranking by a reordering model, the relevance and accuracy of the results are triplely guaranteed. Example 1
[0022] This embodiment provides an intelligent retrieval method that integrates knowledge graphs and semantic vectors, such as... Figure 1 As shown, it includes: S1. Text preprocessing: Import the document and segment the document text to obtain text segments. The original document can be segmented into text segments of approximately 1024 characters each.
[0023] S2. Document Summary Generation Process: Generate a comprehensive summary of the document text based on a large model. A concise and accurate overall summary summarizing the core content of each imported document is generated using a large language model.
[0024] S3-1 Knowledge Graph Construction Process: Extract concept phrases from text segments and construct a knowledge graph, storing the relationship between concept phrases and text segments.
[0025] S3-2, Vectorization process: Encode the document summary and text segments into summary vectors and text segment vectors and store them in the vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs.
[0026] S4. Fusion Retrieval Process: This process retrieves text segments related to the query statement using a knowledge graph, and retrieves relevant summary vectors and text segment vectors from a vector database. The retrieval results (knowledge graph results, text segment vector results, and summary vector results) are then merged to form a preliminary candidate result set. When performing vector similarity retrieval, both text segment vectors and document summary vectors are queried simultaneously. This means that a document highly relevant to the query, even if its specific segments have low scores, may be included in the candidate set due to the relevance of its summary, thus better capturing document-level relevance.
[0027] S5, Reordering Process: This process is used to refine the initial candidate result set using a reordering model, and returns the top-K most relevant text fragments after reordering as the final search results.
[0028] In one specific embodiment, the knowledge graph construction process specifically includes: S31. Concept Extraction and Single Knowledge Link Construction: For the input text segment, the large language model is used to extract the key concept words, and the concept words are constructed into a knowledge link according to semantic association. At the same time, a unique link number is assigned to the knowledge link and an association is established with the text segment.
[0029] Example 1: Original text: "Apple peels contain a lot of nutrients, so it is recommended to wash them and eat them with the peel on." Extracted keywords: apple peel, nutrients, apple, fruit, water Constructing a knowledge chain: [Apple peel ↔ Nutrients ↔ Apple ↔ Fruit ↔ Water] id=1 Example 2: Original text: "Bananas are rich in potassium, which helps maintain heart health and regulate blood pressure." Extracted keywords: banana, fruit, potassium, heart health, blood pressure, human body Building a knowledge chain: [Banana ↔ Fruit ↔ Potassium ↔ Heart Health ↔ Blood Pressure ↔ Human Body] id=2 S32. Knowledge Link Fusion and Graph Construction: Using the same concept nodes in multiple knowledge links as connection points, the knowledge links are merged into a knowledge graph to form a semantic network.
[0030] The knowledge graph structure after merging Example 1 and Example 2 is as follows: Figure 2 As shown.
[0031] Continuous improvement mechanism: With the continuous input of new corpora, concepts will be continuously extracted, new links will be constructed and integrated into the existing knowledge graph, so that the knowledge network will be continuously expanded and optimized, and have the ability to evolve on its own.
[0032] By automating the construction of knowledge links, fusing links to form a graph, and supporting continuous updates, the transformation of knowledge representation from "manual construction" to "automatic evolution" has been realized, significantly improving the semantic understanding capability and scalability of the retrieval system.
[0033] In the fusion retrieval process, the text segments associated with the query statement are retrieved through knowledge graph retrieval (knowledge graph matching), and accurate original text backtracking is achieved through the following steps: S41. Concept Extraction: Use a large model to parse user query statements and extract concept words from the query statements.
[0034] S42. Link Matching: Match relevant knowledge links in the knowledge graph based on concept words, supporting direct association and multi-hop path retrieval.
[0035] S43. Repetition Rate Calculation: Perform repetition rate statistics on the concept words contained in the matched knowledge links.
[0036] S44. Sorting and Filtering: Sort by repetition rate and select the Top-N most relevant knowledge links.
[0037] S45. Then, trace back to the associated text segment based on the most relevant knowledge link.
[0038] Through the above process, the system can accurately locate and aggregate original texts that are highly relevant to the query, providing high-quality contextual support for subsequent mixed queries and deep reasoning in the large language model.
[0039] Preferably, after the text preprocessing process, an integrity check process is included to determine whether a text segment is complete based on a large model. If it is incomplete, the process returns to the text preprocessing process to re-segment the document text. The overall operation process is as follows: Figure 3 As shown, it may include the following steps: 1. Import the text corpus; 2. Use a large language model to generate a comprehensive summary for the document; 3. The generated document summaries are fed into an embedding model to obtain summary vectors, which are then stored in a vector database; 4. Divide the original document into text segments of approximately 1024 characters each; 5. Use a large model to determine if a text segment is complete; if incomplete, re-segment it. 6. Perform the following parallel processing on the complete text segment: 6-1. Extract concept phrases, construct a knowledge graph, and store the association between concepts and text IDs; 6-2. Generate text segment vectors and store them in a vector database; 7. During the retrieval phase, the user enters a query, and the system performs the following operations simultaneously through the fusion retrieval module: 7-1. Retrieve text IDs associated with related concepts using knowledge graphs; 7-2. Retrieve relevant text segments and document summaries simultaneously using vector similarity; 7-3. Merge the three parts of the search results (knowledge graph results, text segment vector results, and summary vector results) to form a preliminary candidate result set; 8. Reorder the preliminary candidate result set and the original query input using the reordering model; return the top-K most relevant text fragments after reordering as the final search results.
[0040] The enhanced RAG, which integrates knowledge graphs, vector retrieval, and reordering models, enables multi-dimensional, high-precision retrieval and refined sorting of text data. This significantly improves the comprehensiveness, accuracy, and relevance of retrieval results while simultaneously taking into account both the overall theme of the document and the detailed information of individual segments.
[0041] Based on the same inventive concept, this application also provides a system corresponding to the method in Embodiment 1, as detailed in Embodiment 2. Example 2
[0042] This embodiment provides an intelligent retrieval system that integrates knowledge graphs and semantic vectors, such as Figure 4 As shown, it includes: Text preprocessing module: Used to import documents and segment document text to obtain text segments.
[0043] Document summarization module: Used to generate overall summaries of document text based on a large model.
[0044] Knowledge graph construction module: used to extract concept phrases from text segments and construct a knowledge graph, storing the relationship between concept phrases and text segments.
[0045] Vectorization module: This module encodes document summaries and text segments into summary vectors and text segment vectors, and stores them in a vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs.
[0046] The fusion retrieval module is used to retrieve text segments associated with the query statement through a knowledge graph, retrieve relevant summary vectors and text segment vectors through a vector database, and merge the retrieval results to form a preliminary candidate result set.
[0047] Reordering module: Used to finely sort the initial candidate result set through a reordering model to obtain the final search results.
[0048] Preferably, after the text preprocessing module, there is also an integrity judgment module, which is used to determine whether the text segment is complete based on the large model. If it is not complete, the text preprocessing module re-segments the document text.
[0049] The connections between the components are as follows: After importing the document, the text preprocessing module sends the document to the document summarization generation module and performs text segmentation. The summarization generation module outputs a summary to the vectorization module. The segmented text fragments are then sent to the integrity judgment module, and the subsequent process remains unchanged. When querying the vector database, the fusion retrieval module includes both text segment vectors and summary vectors.
[0050] Preferably, the knowledge graph construction module is specifically used for: extracting key concept words from the input text segment using a large language model, constructing a knowledge link based on semantic association of the concept words, assigning a unique link number to the knowledge link, and establishing an association with the text segment; using the same concept nodes in multiple knowledge links as connection points, merging the knowledge links into a knowledge graph to form a semantic network.
[0051] Preferably, in the fusion retrieval module, retrieving the text segment associated with the query statement through the knowledge graph specifically includes: extracting concept words from the query statement, matching the concept words in the knowledge graph to obtain relevant knowledge links, performing repetition rate statistics on the concept words contained in the matched knowledge links, sorting them by repetition rate, selecting the Top-N most relevant knowledge links, and then tracing back to the text segment associated with the knowledge links.
[0052] Since the system described in Embodiment 2 of this invention is a system used to implement the method of Embodiment 1 of this invention, those skilled in the art can understand the specific structure and variations of this system based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All systems used in the method of Embodiment 1 of this invention fall within the scope of protection of this invention.
[0053] The technical solutions provided in the embodiments of the present invention have at least the following technical effects: 1. By providing an enhanced RAG method and system that integrates knowledge graphs, vector retrieval, and reordering models, we can achieve multi-dimensional, high-precision retrieval and refined sorting of text data, significantly improving the comprehensiveness, accuracy, and relevance of retrieval results while taking into account both the overall theme of the document and the detailed information of each fragment.
[0054] 2. Enhanced search comprehensiveness: Combining conceptual association and semantic similarity avoids the limitations of a single search method; 3. High interpretability: Knowledge graphs provide visualization and explanation of retrieval paths; 4. More accurate ranking: The re-ranking model performs the final ranking of results based on deep semantic understanding, which significantly improves the quality of top results and is more suitable as the context for large models to generate answers; 5. Introduction of document-level semantic representation (summary): Effectively enhances the ability to retrieve the overall theme and core content of documents, avoids the search results from deviating from the main point, and is particularly suitable for query scenarios that require a macro-level understanding or filtering of documents, thereby improving the overall search depth and breadth of the system.
[0055] 6. Extremely high accuracy: Through completeness judgment, multi-dimensional retrieval and fine sorting of the reordering model, the relevance and accuracy of the results are guaranteed in three ways; 7. Good scalability: It supports dynamic updates of knowledge graphs and vector databases to adapt to constantly changing corpora.
[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0060] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. An intelligent retrieval method integrating knowledge graphs and semantic vectors, characterized in that: include: Text preprocessing process: Import the document and segment the document text to obtain text segments; Document summary generation process: Generate an overall summary of the document text based on a large model; Knowledge graph construction process: Extract concept phrases from text segments and construct a knowledge graph to store the relationship between concept phrases and text segments; Vectorization process: The document summary and text segments are encoded into summary vectors and text segment vectors and stored in a vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs. The fusion retrieval process is used to retrieve text segments associated with the query statement through a knowledge graph, retrieve relevant summary vectors and text segment vectors through a vector database, and merge the retrieval results to form a preliminary candidate result set. Reordering process: Used to finely sort the initial candidate result set through a reordering model to obtain the final search results.
2. The system according to claim 1, characterized in that, Following the text preprocessing process, an integrity check process is also included, which is used to determine whether the text segment is complete based on the large model. If it is incomplete, the process returns to the text preprocessing process to re-segment the document text.
3. The system according to claim 1 or 2, characterized in that: The knowledge graph construction process specifically includes: for the input text segment, using a large language model to extract key concept words, constructing a knowledge link based on semantic association of the concept words, assigning a unique link number to the knowledge link, and establishing an association with the text segment; using the same concept nodes in multiple knowledge links as connection points, merging the knowledge links into a knowledge graph to form a semantic network.
4. The system according to claim 3, characterized in that, The fusion retrieval process specifically includes: extracting concept words from the query statement, matching relevant knowledge links in the knowledge graph based on the concept words, performing repetition rate statistics on the concept words contained in the matched knowledge links, sorting them by repetition rate, selecting the Top-N most relevant knowledge links, and then tracing back to the text segments associated with the knowledge links.
5. An intelligent retrieval system integrating knowledge graphs and semantic vectors, characterized in that, include: Text preprocessing module: used to import documents and segment document text to obtain text segments; Document summarization generation module: used to generate overall summaries of document text based on a large model; Knowledge graph construction module: used to extract concept phrases from text segments and construct a knowledge graph, storing the relationship between concept phrases and text segments; Vectorization module: used to encode document summaries and text segments into summary vectors and text segment vectors and store them in the vector database. The summary vector is stored in association with all text segment vectors of the document to which it belongs. The fusion retrieval module is used to retrieve text segments associated with query statements through knowledge graphs, retrieve relevant summary vectors and text segment vectors through vector databases, and merge the retrieval results to form a preliminary candidate result set. Reordering module: Used to finely sort the initial candidate result set through a reordering model to obtain the final search results.
6. The system according to claim 5, characterized in that, Following the text preprocessing module, there is also an integrity judgment module, which is used to determine whether the text segment is complete based on the large model. If it is incomplete, the text preprocessing module will re-segment the document text.
7. The system according to claim 5 or 6, characterized in that: The knowledge graph construction module is specifically used for: extracting key concept words from the input text segment using a large language model, constructing a knowledge link based on semantic association of the concept words, assigning a unique link number to the knowledge link, and establishing an association with the text segment; using the same concept nodes in multiple knowledge links as connection points, merging the knowledge links into a knowledge graph to form a semantic network.
8. The system according to claim 7, characterized in that: The fusion retrieval module is specifically used to extract concept words from the query statement, match the concept words in the knowledge graph to obtain relevant knowledge links, perform repetition rate statistics on the concept words contained in the matched knowledge links, sort them according to the repetition rate, select the Top-N most relevant knowledge links, and then trace back to the text segments associated with the knowledge links.