A deep learning-based knowledge retrieval method and system in the field of traditional Chinese medicine acupuncture

Abstract

The application relates to the technical field of medical knowledge retrieval, in particular to a traditional Chinese medicine acupuncture field knowledge retrieval method and system based on deep learning, which comprises the following steps: acquiring acupoint sequence position, combining semantic construction vector segments, extracting symptom-related acupoints, analyzing responses, generating path sorting, extracting root combination vector matrices, matching roots to generate coverage structures, screening matched acupoint combinations, and outputting path sequences.In the application, the sequence position information of acupoints in meridian paths is extracted, structured expression is constructed by combining semantic description and treatment content, path organization and recognition ability are enhanced, sorting is formed by fusing curative effect grades and treatment cycles, acupoint response pertinence is improved, semantic comparison and standardization are strengthened by three-dimensional combination vectorization expression of treatment, direction and positioning roots, semantic association between input and knowledge is enhanced by combining keyword matching frequency and distribution density, acupoint combinations are screened based on semantic matching and path continuity, and the result is ensured to have relevance and practicability.

Description

Technical Field

[0001] This invention relates to the field of medical knowledge retrieval technology, and in particular to a knowledge retrieval method and system for the field of traditional Chinese medicine acupuncture based on deep learning. Background Technology

[0002] The field of medical knowledge retrieval technology involves the collection, organization, analysis, and querying of medical information. Core aspects include the construction of medical literature databases, standardization of medical terminology, construction of medical knowledge graphs and establishment of reasoning mechanisms, and medical semantic understanding and natural language processing. The overall technological foundation is based on artificial intelligence, natural language processing, and structured data modeling to achieve semantic-level association and precise retrieval of medical information. Traditional Chinese medicine (TCM) acupuncture knowledge retrieval refers to the process of querying and retrieving information from classical texts, treatment cases, meridians, and acupoints within the TCM acupuncture field. The technical challenges addressed include the diverse terminology, complex information structure, and loose knowledge connections in TCM acupuncture knowledge. Traditional methods construct keyword databases and use Boolean logic matching rules or statistical probability-based word frequency co-occurrence methods for retrieval. However, these methods rely on manually set rules or dictionaries and cannot effectively handle ambiguous expressions and implicit semantic relationships in TCM language. This paper proposes a knowledge retrieval system for the acupuncture field by constructing an entity relationship-annotated dataset, training a deep neural network model to extract acupuncture knowledge entities and their relationships, and combining this with semantic vector matching.

[0003] Traditional methods for retrieving knowledge about Traditional Chinese Medicine (TCM) acupuncture rely on keyword databases and Boolean logic matching rules, or on statistical models of word frequency co-occurrence for reasoning. This approach is highly dependent on manually defined dictionaries and rule systems, lacking dynamic adaptability. When faced with the ambiguous expressions, semantic ambiguity, and polysemy prevalent in TCM language, it often fails to accurately reconstruct the inherent relationships between terms, leading to distorted search results. For example, the same acupoint may use multiple synonyms in different documents; traditional search methods struggle to identify equivalence, resulting in relevant content not being retrieved. Furthermore, the upstream and downstream relationships between meridian pathways lack the expressive power of Boolean logic, making it impossible to include the sequential structure of acupoints in the analysis. These limitations lead to incomplete search results, omission of important information, and a lack of semantic hierarchy differentiation in practical applications. This results in weak decision support in clinical applications and fails to meet the demands for high-dimensional semantic integration and intelligent reasoning in precision medicine environments. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a knowledge retrieval method in the field of traditional Chinese medicine acupuncture based on deep learning, comprising the following steps:

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a knowledge retrieval method in the field of traditional Chinese medicine acupuncture based on deep learning, comprising the following steps:

[0006] S1: Obtain the ordinal position of acupoint nodes in the meridian path, combine the semantic description of adjacent acupoints with the indications of the meridian, perform word classification and root mapping, construct semantic structure fragments and mark the transmission direction, and generate acupoint path vector semantic fragments.

[0007] S2: Based on the set of acupoint nodes in the semantic fragment of the acupoint path vector, extract the acupoint records corresponding to the user's symptom keywords, read the efficacy level and treatment cycle, analyze the response characteristics of the symptoms, and generate the symptom-corresponding path ranking results.

[0008] S3: Based on the acupoint nodes in the path sorting results corresponding to the symptoms, extract the main treatment word roots, direction word roots and location word roots from the semantic structure fragments, combine them into a standard format, aggregate them into a vector set, and generate an acupoint combination semantic vector matrix;

[0009] S4: Based on the root word dimension in the semantic vector matrix of the acupoint combination, retrieve the location of user symptom keywords, summarize the root word matching frequency and distribution density, and generate the semantic coverage structure of input symptoms;

[0010] S5: Based on the semantic coverage structure of the input symptoms and the sorting results of the corresponding paths of the symptoms, filter the acupoint combinations that meet the conditions of semantic matching and path continuity, arrange them according to the frequency of path occurrence, and generate an acupoint retrieval path output sequence.

[0011] As a further aspect of the present invention, the acupoint path vector semantic fragment includes acupoint node sequence information, semantic description of adjacent acupoints, meridian indications, vocabulary classification, root word mapping, and semantic transmission direction; the symptom-corresponding path ranking result includes acupoint response features, efficacy level, and treatment cycle description; the acupoint combination semantic vector matrix includes indication roots, directional roots, and location roots; the input symptom semantic coverage structure includes root word matching frequency, root word distribution density, and semantic association degree; and the acupoint retrieval path output sequence includes semantic matching conditions, path continuity conditions, and path occurrence frequency.

[0012] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0013] S101: Obtain the ordinal information of acupoint nodes in the meridian path, call the correspondence between node number and meridian number, identify the sequential position of the preceding and following nodes in the path, extract structurally stable acupoint pairs according to the continuous distribution of index values, mark the connection direction, and generate acupoint connection ordinal information.

[0014] S102: Based on the information content of the acupoint connection sequence pairs, call the semantic description and meridian indication of adjacent acupoints, filter co-occurring terms and perform root mapping, generate a set of root words between adjacent acupoints according to the term frequency and root word fitting degree, and obtain the number of semantic root word pairs between adjacent acupoints.

[0015] S103: Based on the number of adjacent acupoint semantic root pairs, combined with connection order and direction information, set root weights and generate root index sequences, establish semantic guidance for segments in the path vector according to the weight value distribution, and generate acupoint path vector semantic segments.

[0016] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0017] S201: Based on the set of acupoint nodes included in the semantic fragment of the acupoint path vector, retrieve the symptom keywords input by the user, and compare them with the acupoint names and indications in the acupoint node set. Filter acupoint records with a fitting degree higher than the symptom matching threshold, extract the encoding and semantic identification information of the matched acupoints, and generate a set of symptom-related acupoint records.

[0018] S202: Call the acupoint encoding information in the symptom-associated acupoint record set, locate the corresponding data frame in the search sample, extract the efficacy level and treatment cycle parameters included in each record, construct the acupoint response feature vector group, organize it into a matrix structure, and generate the acupoint response performance matrix.

[0019] S203: Based on the efficacy level and treatment cycle parameters in the acupoint response performance matrix, construct a score value sequence and complete the standardization process, extract the score result of each acupoint, sort the paths according to the score value order, and generate the symptom-corresponding path sorting result.

[0020] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0021] S301: Based on the acupoint nodes ranked first in the symptom-corresponding path ranking results, retrieve the main treatment root, direction root, and location root associated with the corresponding acupoint in the semantic structure segment, extract the term codes of the three types of roots and complete the deduplication process to generate a set of acupoint associated root words;

[0022] S302: Based on the word content in the acupoint-related word root set, combine them according to the word class classification standards of the main treatment word root, direction word root and location word root, establish a coding sequence after unifying the combination format, and generate a three-class word root combination coding sequence;

[0023] S303: Call all the combination units in the three types of word root combination encoding sequences, extract the corresponding semantic vector values, and arrange all the combination vectors in sequence to construct a data frame structure and establish an acupoint combination semantic vector matrix.

[0024] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0025] S401: Based on the root dimension in the semantic vector matrix of the acupoint combination, retrieve the corresponding occurrence position of the symptom keywords input by the user in the vector structure, extract the keyword index information that overlaps with the root dimension, and summarize the records according to the index order to obtain the keyword matching position set.

[0026] S402: Call the index data in the keyword matching position set, calculate the word root matching frequency index based on the number of word roots in the vector structure and the frequency of keyword repetition, and calculate the position spacing distribution value by combining the position sequence of the matching position in the vector, and establish a set of word root distribution statistical parameters;

[0027] S403: Based on the frequency index and positional spacing distribution value in the set of root word distribution statistical parameters, determine the coverage of the corresponding keywords in each vector group, combine the mapping relationship between the keyword index and the original input content, integrate the matching structure in all vectors and encode it to generate the semantic coverage structure of the input symptoms.

[0028] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0029] S501: Based on the semantic coverage structure of the input symptoms and the sorting results of the corresponding paths of the symptoms, extract the acupoint node index that matches the semantic content, filter the sequences in the path that meet the node continuity condition, and obtain the set of semantically continuous matching paths.

[0030] S502: Call the path sequence in the semantic continuous matching path set, calculate the frequency of occurrence of acupoint node combinations in the sequence, summarize the correspondence between node combinations and frequencies, and obtain the acupoint combination frequency statistics matrix;

[0031] S503: Based on the frequency values ​​in the frequency statistics matrix of acupoint combinations, the acupoint node combinations are arranged in order, the sorted path index content is extracted and a corresponding number sequence is established, and an acupoint retrieval path output sequence is generated.

[0032] As a further aspect of the present invention, the acupoint node refers to the acupoint name or code clearly identified in the human body standard meridian system, which is derived from the standard documents and data resources of "The Classic of Acupuncture and Moxibustion" and "Location of Meridians and Acupoints";

[0033] The meridian pathways refer to the orderly arrangement of acupoints constructed based on the meridian theory of traditional Chinese medicine;

[0034] The semantic description refers to natural language text content derived from TCM acupuncture classics, acupuncture teaching materials, or acupoint databases;

[0035] The term "vocabulary classification" refers to the process of dividing the words extracted from the semantic description of acupoints according to their functions and uses, into three categories: main treatment, location, and direction.

[0036] The term "root mapping" refers to the process of reducing vocabulary to basic semantic units, and unifying and merging different forms of words with similar semantics through a root dictionary or language rules.

[0037] The structural segment refers to a structural semantic record composed of elements such as acupoint nodes, semantic roots of adjacent acupoints, and direction of transmission;

[0038] The direction of conduction refers to the sequence of therapeutic effects or the direction of physiological information flow between acupoint nodes in the meridian pathway, as reflected in their functional descriptions.

[0039] As a further aspect of the present invention, the therapeutic efficacy level refers to the qualitative evaluation of the acupoint treatment results in the sample, and is set into four levels: cured, significantly effective, effective, and ineffective, according to the "Standards for Evaluation of Therapeutic Effects of Traditional Chinese Medicine Diseases".

[0040] The treatment cycle refers to the length of time from the start of acupoint intervention to the first relief of symptoms, recorded in days;

[0041] The response characteristics refer to the comprehensive performance of acupoints in different symptom retrieval samples, including intervention time, relief time, and efficacy level.

[0042] The term "main treatment root" refers to the set of root words that indicate the target of acupoint action;

[0043] The directional roots refer to the set of roots that describe the direction of treatment or the trend of qi and blood circulation;

[0044] The locating roots are a set of roots that indicate the location of acupoints on the human body, corresponding to anatomical regions or clinical acupoint selection areas;

[0045] The matching frequency refers to the statistical count of the number of times the user-input keywords appear in the semantic vector matrix;

[0046] The distribution density refers to the degree of concentration of keywords in the vector dimension;

[0047] The input symptom semantic coverage structure refers to the set of matching relationships formed by the keywords input by the user in the acupoint semantic vector matrix;

[0048] The path continuity condition refers to the continuity rule of the sequential arrangement of acupoint nodes in the meridian path;

[0049] The frequency of occurrence of the path refers to the number of times the target acupoint combination path is called or appears in the search samples.

[0050] A knowledge retrieval system for the field of Traditional Chinese Medicine acupuncture based on deep learning, comprising:

[0051] The acupoint structure construction module obtains the ordinal information of acupoint nodes in the meridian path, extracts the semantic description content between adjacent acupoints and the meridian indications, identifies and classifies indication roots, location roots and direction roots, combines roots and acupoint ordinal information to construct structured fragments, and marks the semantic transmission direction between acupoints to generate acupoint path vector semantic fragments.

[0052] The symptom path ranking module extracts acupoint records that match the user-input symptom keywords based on the set of acupoint nodes in the semantic fragment of the acupoint path vector, reads the efficacy level information and treatment cycle description of the associated acupoints, and sorts the acupoint nodes according to the semantic correspondence between keywords and indications to generate symptom path ranking results.

[0053] The combined vector generation module extracts the main treatment root, direction root, and location root corresponding to the acupoint path vector semantic fragment based on the acupoint nodes in the symptom-corresponding path sorting results. It organizes the vector representation structure according to word class, integrates them into a combined representation set after unifying the format, and generates an acupoint combined semantic vector matrix.

[0054] The semantic structure induction module retrieves the standard word positions of the symptom keywords input by the user based on the word root dimension information in the semantic vector matrix of the acupoint combination, performs induction and organization on the occurrence frequency and distribution density of matching word roots in the vector, records the correspondence between the input content and the word root vector, and generates the semantic coverage structure of the input symptom.

[0055] The retrieval path output module selects acupoint combinations that simultaneously meet the semantic matching conditions and path continuity features based on the acupoint node information in the input symptom semantic coverage structure and the path ranking results corresponding to the symptoms. The acupoints are then arranged according to their frequency of occurrence in the structural segments to generate an acupoint retrieval path output sequence.

[0056] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0057] In this invention, by extracting the ordinal information of acupoints in the meridian pathway and combining it with the semantic description and indications of adjacent acupoints, a structured expression is constructed to enhance the path organization and recognition capabilities. By combining clinical dimensions such as efficacy level and treatment cycle, a multi-angle ranking is formed to improve the pertinence of acupoint response. Through the three-dimensional combination of indications, direction, and location roots, a unified vector representation is used to enhance semantic comparison and standardization capabilities. By integrating the matching frequency and distribution density of keywords, the semantic association between input and knowledge is strengthened. Acupoint combinations are selected based on the dual conditions of semantic matching and path continuity, resulting in output results that are more relevant, structurally reasonable, and clinically practical. Attached Figure Description

[0058] 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.

[0059] Figure 1 This is a schematic diagram of the steps of the present invention;

[0060] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0061] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0062] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0063] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0064] Figure 6 This is a detailed schematic diagram of S5 of the present invention;

[0065] Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0066] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0067] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0068] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0069] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0070] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0071] Please see Figure 1 This invention provides a knowledge retrieval method in the field of traditional Chinese medicine acupuncture based on deep learning, comprising the following steps:

[0072] S1: Obtain the ordinal information of acupoint nodes in the meridian path, combine the semantic description of adjacent acupoints with the indications of the meridian, perform word classification and root mapping, construct structured fragments and mark the semantic transmission direction, and generate acupoint path vector semantic fragments.

[0073] Acupoint nodes refer to the basic units that are clearly identified acupoint names or codes in the standard human meridian system, used to mark the therapeutic points in semantic structure construction. They are derived from standard documents and data resources such as "The Classic of Acupuncture and Moxibustion" and "Location of Meridians and Acupoints".

[0074] Meridian pathways refer to the orderly arrangement of acupoints based on the theory of meridians in Traditional Chinese Medicine. They are used to express the natural connection relationships and conduction directions of acupoints within a meridian, and are represented in a topological structure or list sequence manner.

[0075] Semantic description refers to natural language text content derived from TCM acupuncture classics, acupuncture teaching materials, or acupoint databases, used to express the location, action site, and indications of acupoints;

[0076] Lexical classification refers to the process of dividing the words extracted from the semantic description of acupoints according to their functions and uses, into three categories: main treatment, location, and direction.

[0077] Root word mapping refers to the process of reducing vocabulary to basic semantic units. It involves unifying and merging words with similar semantics in different forms through a root word dictionary or language rules to enhance the consistency and comparability of semantic structures.

[0078] Structured fragments refer to structured semantic records composed of elements such as acupoint nodes, semantic roots of adjacent acupoints, and transmission direction. They are used to construct semantic transmission relationships between acupoint paths and are expressed in a field-based or nested data structure.

[0079] Semantic transmission direction refers to the order of therapeutic effects or the direction of physiological information flow reflected in the functional descriptions of acupoint nodes in the meridian pathway, which is used to identify the rationality of semantic connection in the treatment pathway.

[0080] S2: Based on the set of acupoint nodes included in the semantic fragment of acupoint path vector, extract the acupoint records corresponding to the user's input symptom keywords, read the efficacy level and treatment cycle description attached to the set in the search sample, and generate the symptom-corresponding path ranking results by sorting and analyzing the acupoint response features under different symptoms.

[0081] The efficacy rating refers to the qualitative evaluation of the treatment results of acupoints in the sample. According to the "Evaluation Standards for Efficacy of Diseases and Syndromes in Traditional Chinese Medicine", it is set into four levels: cured, significantly effective, effective and ineffective, which are used to rank and analyze the treatment results.

[0082] The treatment cycle refers to the length of time from the start of acupoint intervention to the first relief of symptoms, recorded in days, and used to measure the response speed after acupoint intervention;

[0083] Response characteristics refer to the comprehensive performance of acupoints in different symptom retrieval samples, including intervention time, relief time, and efficacy level, which are used to establish the corresponding path between acupoints and symptoms.

[0084] S3: Based on the acupoint nodes ranked first in the symptom-corresponding path ranking results, extract the main treatment root, direction root, and location root in the semantic structure fragment, establish a three-dimensional combination representation according to word class, and after unifying the format, aggregate them into a vector group to generate an acupoint combination semantic vector matrix.

[0085] The term "therapeutic root" refers to the set of root words that indicate the target of acupoints, and is used to reflect the pathological site or system targeted by the treatment.

[0086] Directional roots refer to a set of roots that describe the direction of treatment or the trend of Qi and blood circulation, used to indicate the path of acupoint action or the direction of meridian flow;

[0087] Locating roots are a set of roots that indicate the location of acupoints on the human body, corresponding to anatomical regions or clinical acupoint selection areas;

[0088] Three-dimensional combination representation refers to a vector structure formed by combining the three types of root words—therapeutic root words, directional root words, and location root words—in the target order, which is used to construct the input features in the acupoint semantic model;

[0089] A vector group is a set of vectors composed of three-dimensional combinations of multiple acupoints, which express the relationship between multiple acupoints in semantic space;

[0090] A vector matrix is ​​a two-dimensional semantic representation matrix constructed with acupoints as indices and three types of word roots as dimensions. It is used for matching and analysis between user input and the semantic structure of acupoints.

[0091] S4: Based on the root word dimension in the semantic vector matrix of acupoint combination, retrieve the occurrence position of the symptom keywords input by the user, perform structural induction on the root word matching frequency and distribution density in each vector, record the degree of correlation between the input content and the semantic vector, and generate the semantic coverage structure of the input symptom.

[0092] Matching frequency refers to the number of times user-input keywords appear in the semantic vector matrix, which is used to measure the coverage of keywords in the acupoint semantic space;

[0093] Distribution density refers to the degree of concentration of keywords in the vector dimension, and is used to evaluate the aggregation of information and user input in the semantic vector;

[0094] The semantic coverage structure of input symptoms refers to the set of matching relationships formed by the keywords input by the user in the semantic vector matrix of acupoints, which is used to guide the semantic relevance analysis of path selection;

[0095] S5: Based on the semantic coverage structure of the input symptoms and the sorting results of the corresponding paths, filter the acupoint node combinations that simultaneously meet the conditions of semantic matching and path continuity, arrange them according to the frequency of path occurrence, and generate the acupoint retrieval path output sequence.

[0096] The path continuity condition refers to the continuity rules of the sequential arrangement of acupoint nodes in the meridian path, which is used to screen path structures that meet the requirements of treatment logic or meridian direction.

[0097] The frequency of path occurrence refers to the number of times the target acupoint combination path is called or appears in the search sample, which is used to reflect the intensity of path use in real treatment records.

[0098] The semantic fragment of the acupoint path vector includes acupoint node sequence information, semantic description of adjacent acupoints, meridian indications, vocabulary classification, root word mapping, and semantic transmission direction. The symptom-corresponding path ranking results include acupoint response features, efficacy level, and treatment cycle description. The acupoint combination semantic vector matrix includes indication root words, direction root words, and location root words. The input symptom semantic coverage structure includes root word matching frequency, root word distribution density, and semantic association degree. The acupoint retrieval path output sequence includes semantic matching conditions, path continuity conditions, and path occurrence frequency.

[0099] Please see Figure 2 The specific steps of S1 are as follows:

[0100] S101: Obtain the ordinal information of acupoint nodes in the meridian path, call the correspondence between node number and meridian number, identify the sequential position of the preceding and following nodes in the path, extract structurally stable acupoint pairs according to the continuous distribution of index values, mark the connection direction, and generate acupoint connection ordinal information.

[0101] When obtaining the sequence information of acupoint nodes in the meridian path, the system first needs to extract the correspondence between each node and the meridian number from the acupoint database and mark its sequence position in the path. For example, if the acupoint numbered N001 belongs to the meridian numbered M01 and is the first in the meridian, it is recorded as M01-N001-1. By sequentially reading the list of acupoint nodes in each meridian, the system extracts the index value of each acupoint in the path and compares adjacent nodes to determine whether their index values ​​are continuous. If the continuous interval is 1, it is considered to form a structurally stable acupoint pair. The numbers of the two nodes and their connection direction are recorded. The direction is determined by the increasing sequence number. For example, an index value from 4 to 5 is considered forward, and from 5 to 4 is considered reverse. This determination is applicable to path direction recognition. The system can traverse all acupoints. The meridian paths are processed in batches. Taking the Lung Meridian as an example, it contains 11 nodes. The index relationship from N001 to N011 is compared sequentially, and 10 stable connection sequence pairs are extracted. Each pair is recorded in the form of (previous node number, next node number, connection direction). To avoid misidentification, the continuous index judgment threshold is set to ±1. Connection pairs exceeding this range will not be counted. During processing, a sequential structure is traversed and judged in real time. The uniqueness of the path is ensured by combining the meridian to which the node belongs. For example, when N008 and N009 are "Jingqu" and "Taiyuan" respectively, and their sequence positions in the Lung Meridian are 5 and 6 respectively, they are determined to be a valid stable acupoint pair with a positive direction. The information item is recorded as (N008, N009, →). The sequence connection information of all meridian paths can be output in batches in this way.

[0102] S102: Based on the information content of acupoint connection sequence pairs, call the semantic description and meridian indication of adjacent acupoints, filter co-occurring terms and perform root mapping, generate a set of root words between adjacent acupoints according to term frequency and root word fitting degree, and obtain the number of semantic root word pairs between adjacent acupoints.

[0103] When invoking the semantic descriptions and meridian indications of adjacent acupoints, a semantic content extraction system needs to be constructed to extract the functional descriptions, location descriptions, and indications of acupoint nodes. It also needs to retrieve the pathological and indication information of the meridian to which the acupoint is attached, forming a term pool. For each pair of connected acupoints, a term set is established, and the co-occurrence of terms is statistically analyzed. For example, if two acupoint descriptions both contain terms like "headache" and "nasal congestion," a co-occurrence relationship is considered. After extracting the co-occurring terms, a root word extraction operation is performed on the terms. Natural language processing tools are used to decompose the terms into core roots, unifying the semantic expression form and forming a standard root word set. The root word sets of each pair of acupoints are then used to perform set intersection operations to obtain two sections. The system retains only acupoint pairs with at least one co-occurring root word between points. In practical applications, if the descriptions of "Yingxiang" and "Juliao" contain "facial paralysis," "nasal congestion," and "mouth deviation," and "facial paralysis," "eyelid ptosis," and "nasal congestion," respectively, after root word extraction, the roots "face" and "nose" are obtained. If the two roots appear simultaneously in two acupoints, it indicates that the semantic root word pair of the acupoint pair has a value of 2. All root word statistics are achieved through co-occurrence frequency. Root word generation is achieved by mapping entries to a standard root word library. In this process, there is no need for manual screening of word meanings. It is only necessary to construct rule mapping relationships and iteratively process all connection order pairs to obtain the root word quantity information between each pair of acupoints.

[0104] S103: Based on the quantity of adjacent acupoint semantic root pairs, combined with connection order and direction information, set root weights and generate root index sequences. Based on the weight value distribution, establish semantic guidance for segments in the path vector and generate acupoint path vector semantic segments.

[0105] Based on the number of semantic root pairs between adjacent acupoints, combined with their connection order and direction, weights can be assigned to each root. The weight depends on the frequency of the root's occurrence across all paths and its semantic importance. Frequency is obtained through co-occurrence counts, and semantic importance is obtained through corpus comparison. The two indicators are multiplied to form the weight value. To improve semantic indexing accuracy, a minimum retention weight of 0.05 can be set. Roots with weights below this value will not be included in the path vector calculation. All retained roots are sorted by weight to form an index sequence. In the path, roots are arranged according to the acupoint connection order to construct a semantic-guided sequence, forming semantic segments of the acupoint path. For example, in the Lung Meridian acupoints, the path from "Lieque" to "Taiyuan" is a continuous connection path. Root indexes are extracted from all segment connection pairs. If “qi”, “cough”, and “pant” appear consecutively between multiple acupoints, a root sequence “qi→cough→pant” is generated according to the connection order, representing the semantic direction of the path. Taking the complete path as the unit, all segments are combined to form a path vector. For example, a path can be composed of 5 sets of connection pairs, each of which generates a root sequence, together forming an acupoint path vector with semantic representation. During the execution process, all root extraction, matching, sorting, filtering, and weight assignment are processed by automated algorithms. The required data comes from a standardized semantic description library and a thesaurus of treatments. Root weight calculation and sorting do not depend on a specific platform, only data processing tools with semantic statistical capabilities are required.

[0106] Please see Figure 3 The specific steps of S2 are as follows:

[0107] S201: Based on the set of acupoint nodes included in the semantic fragment of acupoint path vector, retrieve the symptom keywords entered by the user, and compare them with the acupoint names and indications in the acupoint node set. Filter acupoint records with a fitting degree higher than the symptom matching threshold, extract the encoding and semantic identification information of the matched acupoints, and generate a set of symptom-related acupoint records.

[0108] Based on the set of acupoint nodes included in the semantic fragments of acupoint path vectors, an index dataset containing acupoint names, numbers, associated paths, and indications needs to be constructed first. For user-inputted symptom keywords such as "headache" and "menstrual irregularities," term standardization is performed, breaking down compound words into basic keywords. For example, "headache" can be decomposed into "head" and "pain," and a standardized word root table is used for mapping. After term cleaning, the content of each acupoint name and indication description field in the acupoint node set is compared. Each keyword is compared with the corresponding acupoint text content, and a term similarity mechanism is used to evaluate the fit between keywords and text content. During the processing, a similarity model based on word frequency vectors is used to construct the similarity index. A similarity scoring system can be used, for example, to set the word frequency vector value as the number of times a word appears in the description, and then calculate the cosine value between the two vectors as the basic similarity score. The symptom matching threshold is set to 60%. When the comparison score is higher than this threshold, the acupoint is determined to be the symptom matching target. The acupoint number, name, fitted keywords and comparison score are extracted and saved as a valid matching record. The record format is composed of code, name, word and score. For example, after inputting "sore throat", if the description of the acupoint "Tian Tu" contains the keywords "pharynx", "throat" and "pain", the matching rate is 66.7%, and it is recorded as a result. All records that meet the matching conditions will be summarized into a set of symptom-related acupoint records for subsequent processing.

[0109] S202: Call the acupoint encoding information in the symptom-associated acupoint record set, locate the corresponding data frame in the search sample, extract the efficacy level and treatment cycle parameters included in each record, construct the acupoint response feature vector group, organize it into a matrix structure, and generate the acupoint response performance matrix.

[0110] The system retrieves acupoint codes from the symptom-associated acupoint record set. It then searches the acupoint database for the corresponding data unit for each acupoint, locates its data frame position, and extracts the efficacy level and treatment cycle fields. The efficacy level is typically an integer between 1 and 5, representing an effect from weak to strong. The treatment cycle is in days or weeks, and the cycle must be formatted as days in the data structure. All extracted levels and cycles form acupoint feature records, and a data list is constructed record by record for each acupoint. For example, acupoint A103 has an efficacy level of 4 and a treatment cycle of 7 days. Therefore, the efficacy level of this acupoint is... The eigenvector is set to level 4 and period 7. After extracting all acupoints that meet the criteria, a response feature matrix containing several acupoints is formed. These are arranged in order, such as A101 level 3 period 7 days, A102 level 4 period 5 days, A103 level 5 period 6 days, A104 level 3 period 9 days, and A105 level 2 period 8 days. This structure is used in subsequent calculation stages. If some acupoints are missing field information, the default level is set to 3 and the period to 7 days to ensure the consistency of the feature structure. During the processing, all data structures are integrated using encoding as an index to support index matching and data traceability in the subsequent scoring process.

[0111] S203: Based on the efficacy level and treatment cycle parameters in the acupoint response performance matrix, construct a score value sequence and complete the standardization process, extract the score results for each acupoint, sort the paths according to the score values, and generate the path sorting results corresponding to the symptoms.

[0112] Based on the efficacy level and treatment cycle parameters included in the acupoint response performance matrix, a scoring mechanism is established to construct an acupoint priority score sequence. The scoring rule directly assigns integer scores based on the efficacy level, with the treatment cycle serving as a negative correction factor. Higher efficacy levels result in higher scores, while longer treatment cycles result in lower scores. The score is obtained by subtracting the cycle correction value from the efficacy level score. For example, an acupoint with a level of 4 and a treatment cycle of 5 days scores 3.5, while an acupoint with a level of 3 and a treatment cycle of 9 days scores 2.1. After all acupoints are scored according to the same scoring rule, the original score sequence is standardized to ensure that all... The scores are distributed between 0 and 1, and are converted using a linear distribution between the maximum and minimum values. For example, if the score range is between 2.0 and 4.2, and a certain acupoint scores 3.5, its standardized result is approximately 0.68. After processing all acupoint scores, the standardized results are sorted in descending order to obtain the corresponding acupoint path ranking sequence. For example, the standardized scores from high to low are A103, A102, A105, A104, and A101, which represent the priority order of the acupoint paths associated with the currently input symptom information. The system outputs the ranking results for structural calls and efficacy comparison analysis.

[0113] Please see Figure 4 The specific steps of S3 are as follows:

[0114] S301: Based on the acupoint nodes ranked first in the symptom-corresponding path ranking results, retrieve the main treatment root, direction root, and location root associated with the corresponding acupoint in the semantic structure segment, extract the term codes of the three types of roots and complete the deduplication process to generate a set of acupoint associated root words;

[0115] Based on the top-ranked acupoint nodes in the symptom-corresponding path ranking results, the descriptive information of each acupoint in the semantic structure fragment is read sequentially. Three types of word roots related to each acupoint are extracted: indication roots, directional roots, and location roots. Indication roots express the name of the disease symptom associated with the acupoint, such as "pharynx," "pain," and "cough." Directional roots express the directional description of energy or regulatory effects during treatment, such as "ascending" and "internal penetration." Location roots characterize the body part where the acupoint is located, such as "neck," "chest," and "face." Each acupoint corresponds to multiple word roots. Semantic items are identified one by one using structured parsing, converting the text information into standard word root codes. These word root codes are mapped according to a semantic standard library; for example, "pharynx" is mapped to D101. "Internal penetration" is mapped to F202, and "neck" is mapped to L305. After mapping, all encoded information is added to three independent lists: main treatment, direction, and location. The encoded content in the lists is deduplicated by setting a unique set structure to automatically exclude duplicates. For example, if multiple acupoints involve the main treatment root "pharynx", only one record, D101, is retained. After processing, the encoding sets for main treatment roots are D101, D105, and D108; for direction roots, F201 and F203; and for location roots, L301 and L305. The three sets are merged to form a set of acupoint-related roots, ensuring that there is no redundancy, no missing information, and unique encoding within the set, providing a semantic support foundation for subsequent combination encoding work.

[0116] S302: Based on the word content in the acupoint-related word root set, combine them according to the word class classification standards of the main treatment word root, the direction word root and the location word root, and establish a coding sequence after unifying the combination format to generate a three-class word root combination coding sequence;

[0117] Based on the extracted set of acupoint-related word roots, the word roots are categorized into three groups according to their semantic categories: indication, direction, and location. These three categories must be strictly distinguished and cannot be cross-classified. The word root codes within each list are arranged in a standard order, such as D101, D102, and D108 for indication, F201 and F202 for direction, and L301 and L305 for location. Then, word roots are combined in a sequence with indication first, direction second, and location last. All possible word root combination units are constructed by traversing the three categories. If there are 3 word roots in the indication category, 2 in the direction category, and 2 in the location category, a maximum of 3 × 2 × 2, or 12, word root combinations can be formed. Each combination constitutes a ternary word root code. For example, the first group is D101, F201, L301, which are concatenated into the combination structure D101F201L301. To ensure standardized format, a unified prefix identification method is used for the combination codes. For example, Z represents the main treatment, F represents the direction, and L represents the location, so the combination code is Z101F201L301. All combinations use this coding standard to form a complete coding sequence. If a certain type of word root is missing, it is filled with NULL. For example, if there is no combination content for the direction category, the combination is Z101NULLL301. After all combinations are completed, they are uniformly deduplicated and sorted lexicographically to obtain a complete and hierarchical three-type word root combination coding sequence, which provides a standard index for the semantic vector construction process.

[0118] S303: Call all combination units in the three types of word root combination encoding sequences, extract the corresponding semantic vector values, and arrange all combination vectors in sequence to construct a data frame structure and establish an acupoint combination semantic vector matrix;

[0119] The algorithm retrieves each combination unit recorded in the three types of root word combination encoding sequences, and sequentially searches the root word semantic vector database for the vector value corresponding to each encoding group. Each semantic vector consists of multiple dimensions, with a fixed number of dimensions (e.g., 6 dimensions), corresponding to semantic category, functional tendency, symptom fit, directional strength, spatial distribution range, and location precision. Each dimension's value is a decimal between 0 and 1, representing the semantic weight distribution under different dimensions. For example, the vector values ​​corresponding to the combination encoding Z101F201L301 in the database are 0.82, 0.45, 0.67, and 0.54. 0.28, 0.91, all codes are matched with their vector values ​​in sequence. If a code is not recorded in the database, a default value of 0.5 is assigned as the filling content for each dimension of the vector. After the vector retrieval is completed, the code and its semantic vector are combined to construct a structured data frame. The first column is the combined code, and the subsequent columns are the vector dimension values. All data frame structure records are added one by one according to the order of the code list to form an acupoint combined semantic vector matrix. Each row represents the semantic features of a combined unit, and each column is the basis for comparison under the same dimension. This matrix will serve as the basic data input for subsequent semantic evaluation and path modeling.

[0120] Please see Figure 5 The specific steps of S4 are as follows:

[0121] S401: Based on the root dimension in the semantic vector matrix of acupoint combination, retrieve the corresponding occurrence positions of the symptom keywords entered by the user in the vector structure, extract the keyword index information that overlaps with the root dimension, summarize the records in index order, and obtain the keyword matching position set.

[0122] Based on the root word dimension in the semantic vector matrix of acupoint combinations, the symptom keywords provided by the user need to be structured first. Effective semantic units in the input content are identified through word segmentation technology. For example, "cough, chest tightness, shortness of breath" is identified as the keyword set "cough," "chest tightness," and "shortness of breath." Then, by referring to the preset root word mapping relationship, each keyword is converted into a corresponding root word code. For example, "cough" corresponds to the main treatment root word D101, "chest tightness" corresponds to D112, and "shortness of breath" corresponds to D110, resulting in the code set D101, D112, and D110. After entering the semantic vector matrix, it needs to be checked row by row. The vector dimension content of each acupoint combination code is checked, and it is determined whether the three types of root codes of indication, direction and location contain any of the above codes. If there is at least one matching root code in a row, it is considered a matching event, and its row index number in the matrix is ​​recorded. For example, if rows 3, 7 and 12 match D101, D110 and D112 respectively, the index is recorded as 3, 7 and 12. These matching row numbers are arranged in ascending order of value to construct a keyword matching position set. This set reflects the position distribution of the input symptoms mapped and identified in the semantic matrix, providing a position information basis for subsequent frequency and structural coverage analysis.

[0123] S402: The index data in the keyword matching position set is retrieved, and the specific formula for calculating the root word matching frequency index is as follows, based on the number of root words and the frequency of keyword repetition in the vector structure:

[0124] ;

[0125] Calculate the word root matching frequency index value, combine it with the position sequence of the matching position in the vector, calculate the position spacing distribution value, and establish a set of word root distribution statistical parameters;

[0126] in, Representative word root The matching frequency index value, This represents the presence of word roots in the vector structure. The total number of keywords Represents the keyword matching location set with the root word The total number of all matched keywords Representing the The root of each keyword The frequency of repetition, Representing the The semantic weight value of each keyword in the vector structure. Representing the The index difference between the keyword matching position and the vector center position. This represents the distance unit (such as character, byte, or vector unit interval) between two adjacent positions in a vector. Represents a very small positive real number, used to avoid division by zero. Representatives and word roots The average semantic weight of all matched keywords in the vector structure. Indicates all related to the root word The matched keywords are summed. Represents the absolute value of the overall calculation result. The square root of the number of keywords corresponding to a word root is used to normalize frequency metrics.

[0127] The keyword vector structure has a length of 200, a center index position of 100, and the detected matching root word "act" in the vector. Five keywords containing this root word were detected through word segmentation and stemming, numbered k1 to k5. Let their position indices in the vector be 82, 95, 104, 110, and 121. The position difference parameters calculated based on the index offset and the center position are as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] ;

[0132] ;

[0133] Vector unit spacing value The unit of measurement is character granularity. According to the character vector encoding structure, the distance length per character unit is 1, and the unit is L (here, the logical length unit). Monitoring has confirmed that this is set to... ;

[0134] Minimal smoothing value The range of values ​​is set based on literature review and numerical stability requirements. to To ensure effective avoidance of division-by-zero errors while preventing disturbance effects, it is set as follows: ;

[0135] The frequency of repetition of the root word 'r' in each keyword, as determined by term frequency statistics, is as follows:

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] Keyword semantic weight TF-IDF weights are calculated based on context and normalized to a range between 0 and 1. Specific values ​​are as follows:

[0142] ;

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] Average semantic weight The arithmetic mean of the weights of the five keywords mentioned above is calculated as follows:

[0148] ;

[0149] Substitute into the main formula:

[0150] The total number of keywords contained in the root word ;

[0151] Matching location set keyword count ;

[0152] Calculate the main term separately:

[0153] Item 1:

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] Item 2:

[0159] ;

[0160] ;

[0161] ;

[0162] Combine the two items to calculate the difference:

[0163] ;

[0164] Normalized multiplication terms:

[0165] ;

[0166] The results show that, in the current matching vector, the semantic density of the root word "act" after positional weighting and frequency fusion processing has a frequency index of 0.7614. This value reflects that the root word appears frequently and is concentrated in the nearest neighbor position in the current semantic vector structure, and has a high degree of weight aggregation. It can be used as the input feature parameter of the root word density distribution model, and can be used in subsequent root word distribution trend modeling or classification feature selection scenarios.

[0167] This formula integrates three factors—word root frequency, semantic weight, and positional distribution—to construct a composite index reflecting the density and concentration of word roots within a vector structure. The frequency parameter reflects the intensity of word root occurrence within local keywords, while the semantic weight measures the semantic importance of the corresponding keyword. The product of these two factors represents the contribution of each matching position to the overall word root expression. Positional distance is represented by the product of the vector subscript difference and the unit spacing, with a small positive number added to the denominator in square root form for smoothing. This smoothing process suppresses interference from distant words in the center density calculation, making the contribution of nearby words more significant. The sum of these weighted values ​​represents the word root density state in the actual structure. Simultaneously, a reference density state based on average weights is calculated using the same structure. Subtracting the two and taking the absolute value quantifies the offset strength of the word root expression. A normalization factor eliminates the unevenness in magnitude caused by differences in the number of matches, ensuring the comparability of the index across different word roots or texts. The overall structure integrates information from different dimensions through multiplication, achieves distance-sensitive control through the square root, and expresses the global offset trend through summation and addition / subtraction, reflecting the structural distribution characteristics of word roots in the semantic space;

[0168] The word root matching frequency index is used to measure the concentration of a specific word root in a vector structure, the strength of its semantic weight, and its relative distribution with respect to the structural center. The higher the value, the more frequently the word root appears in the current semantic context, and the more concentrated it is in the position close to the semantic core area, accompanied by a higher keyword weight. This reflects that the word root has strong semantic representativeness and organizational stability in the text structure. This index comprehensively considers the repetition frequency of the word root in the keywords, the importance weight of the keywords, and its spatial position relative to the center. It describes the distribution trend and semantic contribution of the word root from multiple dimensions, which helps to identify representative, stable, or clustered semantic units in subsequent tasks.

[0169] S403: Based on the frequency index and positional spacing distribution value in the root word distribution statistical parameter set, determine the coverage of the corresponding keywords in each vector group, combine the mapping relationship between the keyword index and the original input content, integrate the matching structure in all vectors and encode it to generate the semantic coverage structure of the input symptoms.

[0170] Based on the data in the root word distribution statistical parameter set, keyword coverage analysis is performed on each vector group in the semantic matrix. First, the root word code list contained in each vector group is read and compared with the root word codes mapped to the user-input keywords. The number of matching codes is counted and calculated as a ratio to the total number of codes in that group. For example, if a vector group contains D101, F203, and L301, and the input keyword mapping contains D101, D110, and D112, then D101 is a successful match, indicating a match count of 1. The total number of codes is 3, and the coverage is 33.3%. If a group has 2 matching codes, the coverage is 66.7%. Each... The calculation results are recorded as corresponding coverage values, and a correspondence table is established. The correspondence between keywords and their root codes, and between root codes and matching combination codes, is summarized and encoded in a structured manner. For example, "cough" is mapped to D101, and D101 matches the combination code Z101F203L301 with a matching coverage of 66.7%. The generated structure entry code is S01, indicating that the input keyword is partially covered in this combination unit. All matching results between input keywords and combination units are encoded and collected in this way to form a unified input symptom semantic coverage structure, providing a basis for subsequent scoring or path optimization analysis.

[0171] Please see Figure 6 The specific steps of S5 are as follows:

[0172] S501: Based on the input symptom semantic coverage structure and the symptom corresponding path sorting results, extract the acupoint node index that matches the semantic content, filter the sequence in the path that meets the node continuity condition, and obtain the set of semantically continuous matching paths.

[0173] The input natural language symptoms need to be converted into multi-dimensional vectors using a semantic model. Commonly used models include Word2Vec or BERT. The converted results are then mapped into a pre-defined semantic coverage structure, which is a directed graph where nodes represent specific symptoms and edges represent semantic similarity connections. Semantic similarity is calculated using the cosine of the angle between vectors. When the result is higher than a set threshold, such as 0.75, the two nodes are considered related. Taking "cough" as an example, after semantic model conversion, the "Feishu" node is matched, with a similarity of 0.81, meeting the matching criteria and thus recorded as a selectable acupoint node. For each path connected by symptoms, the semantics of each acupoint node needs to be analyzed to see if it matches the input symptom, and whether adjacent acupoint nodes maintain a continuous connection in the path. For example, if a path consists of four nodes, and three of them are sequentially and effectively connected without interruption, it is considered a continuous path. Conversely, if there are non-directly connected nodes or logical breaks in the path, they need to be removed. This process involves path retrieval and judgment logic of the node structure graph. Finally, all paths that satisfy semantic matching and have continuous nodes are retained to form a set of semantically continuous matching paths.

[0174] S502: Call the path sequence in the semantic continuous matching path set, calculate the frequency of acupoint node combinations in the sequence, summarize the correspondence between node combinations and frequencies, and obtain the acupoint combination frequency statistics matrix;

[0175] In the semantically continuous matching path set, the acupoint node combinations of each path need to be analyzed one by one. Continuous node pairs or triples are extracted by setting a sliding window length. For example, a four-node path can form three binary combinations, each considered a statistical unit with an initial frequency of 0. During the traversal of all paths, the frequencies of identical combinations are summed. If the same combination appears in all three paths, the frequency value is 3. The entire set can be constructed as a key-value mapping table, where the key is the acupoint combination and the value is the number of occurrences, forming a frequency statistics dictionary. Further, a frequency matrix model is constructed. The frequency of each combination needs to be filtered according to a set threshold. For example, if the threshold is set to 2, all combinations below this value will not proceed to the next processing stage. Taking the combination "Feishu, Zhongfu" as an example, if its frequency is 3, it meets the requirement and is retained as a subsequent sorting object. The frequency information of all combinations forms an acupoint combination frequency matrix, used for sorting and extraction in the next stage.

[0176] S503: Based on the frequency values ​​in the acupoint combination frequency statistics matrix, the acupoint node combinations are arranged in order, the sorted path index content is extracted and a corresponding number sequence is established, and an acupoint retrieval path output sequence is generated.

[0177] In the acquired frequency statistics, all combinations are sorted in descending order of frequency, with priority determined by frequency values. The sorting method can employ quicksort or a priority queue structure. Each combination is assigned a unique number after sorting, for example, starting from 001 and incrementing. These numbers are then linked to acupoint combinations, forming a lookup table between numbers and path combinations. For example, the combination "Zhongfu, Tanzhong" has a frequency of 4, "Feishu, Zhongfu" 3, and "Tanzhong, Neiguan" 2. After sorting, they are sequentially numbered 001, 002, and 003, forming the output sequence. This numbered sequence constitutes the complete path output content based on the sorting results. This output sequence can be directly used as an acupoint retrieval index, combined with input symptoms to form path recommendations or preliminary scheme construction. The output path sequence completely covers high-frequency, continuous, and semantically consistent acupoint combinations, providing a stable basis for subsequent path recognition modules and recommendation systems.

[0178] Please see Figure 7 A knowledge retrieval system for the field of traditional Chinese medicine and acupuncture based on deep learning, comprising:

[0179] The acupoint structure construction module obtains the ordinal information of acupoint nodes in the meridian path, extracts the semantic description content between adjacent acupoints and the meridian indications, identifies and classifies indication roots, location roots and direction roots, combines roots and acupoint ordinal information to construct structured fragments, and marks the semantic transmission direction between acupoints to generate acupoint path vector semantic fragments.

[0180] The symptom path ranking module extracts acupoint records that match the user-input symptom keywords based on the set of acupoint nodes in the semantic fragment of acupoint path vector, reads the efficacy level information and treatment cycle description of the associated acupoints, and sorts the acupoint nodes according to the semantic correspondence between keywords and indications to generate symptom path ranking results.

[0181] The combined vector generation module extracts the main treatment root, direction root, and location root corresponding to the acupoint path vector semantic fragment based on the acupoint nodes in the symptom-corresponding path sorting results. It organizes the vector representation structure according to word class, integrates them into a combined representation set after unifying the format, and generates an acupoint combined semantic vector matrix.

[0182] The semantic structure induction module retrieves the standard term positions of the symptom keywords entered by the user based on the root dimension information in the semantic vector matrix of acupoint combination, performs induction and organization on the frequency and distribution density of the matching root words in the vector, records the correspondence between the input content and the root word vector, and generates the semantic coverage structure of the input symptom.

[0183] The retrieval path output module selects acupoint combinations that simultaneously meet the semantic matching conditions and path continuity features based on the acupoint node information in the input symptom semantic coverage structure and the symptom corresponding path sorting results. The acupoints are then arranged according to their frequency of occurrence in the structural fragments to generate an acupoint retrieval path output sequence.

[0184] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion, characterized by, Comprise the following steps: S1: Obtain the order of acupoint nodes in the meridian path, combine the semantic description of adjacent acupoints with the expression of meridian treatment, perform vocabulary classification and root mapping, construct semantic structure fragments and label the transmission direction, and generate acupoint path vector semantic fragments; S2: Based on the acupoint node set in the acupoint path vector semantic fragment, extract the acupoint record corresponding to the user's symptom keywords, read the treatment cycle and treatment effect level, analyze the response characteristics of the symptoms, and generate a symptom corresponding path sorting result; S3: Based on the acupoint nodes in the symptom corresponding path sorting result, extract the treatment root, direction root and positioning root in the semantic structure fragment, combine them into a standard format, and gather them into a vector set to generate an acupoint combination semantic vector matrix; The specific steps of S3 are: S301: Based on the acupoint nodes with high ranking in the symptom corresponding path sorting result, retrieve the treatment root, direction root and positioning root associated with the acupoint in the semantic structure fragment, extract the term encoding of the three types of roots and complete the deduplication processing, and generate an acupoint associated root set; S302: According to the term content in the acupoint associated root set, combine the treatment root, direction root and positioning root according to the part-of-speech classification standard, establish an encoding sequence after uniform combination format, and generate a three-type root combination encoding sequence; S303: Call all combination units in the three-type root combination encoding sequence, extract the corresponding semantic vector value, and arrange all combination vectors in sequence to build a data frame structure to establish an acupoint combination semantic vector matrix; S4: Based on the root dimension in the acupoint combination semantic vector matrix, retrieve the user's symptom keyword position, summarize the root matching frequency and distribution density, and generate an input symptom semantic coverage structure; The specific steps of S4 are: S401: Based on the root dimension in the acupoint combination semantic vector matrix, retrieve the corresponding appearance position of the user's input symptom keyword in the vector structure, extract the keyword index information that overlaps with the root dimension, and summarize the records in index order to obtain a keyword matching position set; S402: Call the index data in the keyword matching position set, calculate the root matching frequency index according to the number of roots in the vector structure and the repetition frequency of keywords, calculate the position interval distribution value combined with the position sequence of the matching position in the vector, and establish a root distribution statistical parameter set; S403: According to the frequency index and position interval distribution value in the root distribution statistical parameter set, judge the coverage degree of the corresponding keywords in each vector, integrate the matching structure in all vectors and encode it according to the mapping relationship between the keyword index and the original input content, and generate an input symptom semantic coverage structure; S5: According to the input symptom semantic coverage structure and the symptom corresponding path sorting result, filter the acupoint combination that meets the semantic matching and path continuity conditions, arrange them according to the path appearance frequency, and generate an acupoint retrieval path output sequence. 2.The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 1, characterized in that, The acupoint path vector semantic segment includes acupoint node sequence information, adjacent acupoint semantic description, meridian main treatment expression, vocabulary classification, root mapping, semantic conduction direction, the symptom corresponding path sorting result includes acupoint response characteristics, curative effect grade, treatment cycle description, the acupoint combination semantic vector matrix includes main treatment root, direction root, positioning root, the input symptom semantic coverage structure includes root matching frequency, root distribution density, semantic correlation degree, and the acupoint retrieval path output sequence includes semantic matching condition, path continuity condition and path appearance frequency. 3.The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 1, characterized in that, The specific steps of S1 are: S101: acquire the sequence information of acupoint nodes in the meridian path, call the corresponding relationship between node number and meridian number, identify the sequential position of the front and rear nodes in the path, extract the acupoint pair with stable structure according to the continuous distribution of index values, and label the connection direction to generate the acupoint connection sequence pair information amount; S102: based on the acupoint connection sequence pair information amount, call the semantic description of adjacent acupoints and the meridian main treatment expression, filter co-occurrence terms and perform root mapping, generate a root set between adjacent acupoints according to the term frequency and root fitting degree, and obtain the adjacent acupoint semantic root pair quantity value; S103: according to the adjacent acupoint semantic root pair quantity value, combine the connection sequence and direction information, set the root weight and generate the root index sequence, establish the semantic guidance of the segment in the path vector according to the weight value distribution, and generate the acupoint path vector semantic segment.

4. The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 3, characterized in that, The specific steps of S2 are: S201: based on the acupoint node set included in the acupoint path vector semantic segment, retrieve the symptom keywords input by the user, and perform term comparison with the acupoint name and main treatment content in the acupoint node set, filter the acupoint records with a higher fitting degree than the symptom matching threshold, extract the coding and semantic identification information of the matched acupoints, and generate a symptom-associated acupoint record set; S202: call the acupoint coding information in the symptom-associated acupoint record set, locate the corresponding data frame in the retrieval sample, extract the curative effect grade and treatment cycle parameters included in each record, construct an acupoint response characteristic vector group, organize a matrix structure, and generate an acupoint response performance matrix; S203: according to the curative effect grade and treatment cycle parameters in the acupoint response performance matrix, construct a score value sequence and complete standardization processing, extract the score results of each acupoint, complete path sorting according to the score value order, and generate a symptom corresponding path sorting result. 5.The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 1, characterized in that, The specific steps of S5 are: S501: based on the input symptom semantic coverage structure and the symptom corresponding path sorting result, extract the acupoint node index matched with the semantic content, filter the sequence that meets the node continuity condition in the path, obtain a semantic continuous matching path set; S502: call the path sequence in the semantic continuous matching path set, calculate the appearance frequency of acupoint node combination in the sequence, summarize the corresponding relationship between node combination and frequency, and obtain an acupoint combination frequency statistical matrix; S503: According to the frequency value in the frequency statistical matrix of the acupoint combination, the acupoint node combination is sequentially arranged, the sorted path index content is extracted and the corresponding numbered sequence is established, and an acupoint retrieval path output sequence is generated. 6.The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 1, wherein, The acupoint node refers to the acupoint name or code based on the standard meridian system of the human body, which comes from the standard literature and data resources of "Needle and Acupuncture" and "Meridian and Acupoint Positioning"; The meridian path refers to the ordered arrangement link of acupoints constructed according to the theory of traditional Chinese medicine meridians; The semantic description refers to the natural language text content from traditional Chinese medicine acupuncture classics, acupuncture teaching materials or acupoint database; The vocabulary classification refers to the process of dividing the extracted words in the semantic description of acupoints according to the function and purpose, which is divided into three categories: treatment, positioning and direction; The root mapping refers to the process of reducing words to basic semantic units, which unifies similar semantic words of different forms through root dictionary or language rules; The structure fragment refers to a structural semantic record composed of acupoint nodes, adjacent acupoint semantic roots and transmission direction elements; The transmission direction refers to the order of therapeutic effect or physiological information flow direction between acupoint nodes in the meridian path. 7.The deep learning-based knowledge retrieval method in the field of traditional Chinese acupuncture and moxibustion according to claim 1, characterized in that, The therapeutic effect grade refers to the qualitative evaluation of acupoint treatment results in samples, which is set as recovery, significant effect, effective and ineffective according to "Evaluation Criteria for Traditional Chinese Medicine Disease Treatment"; The treatment cycle refers to the length of time from the start of acupoint intervention treatment to the first relief of symptoms, recorded in days; The response characteristics refer to the comprehensive performance of acupoints in different symptom retrieval samples, including intervention time, relief time and therapeutic effect grade; The treatment root refers to a set of roots representing the object of acupoint action; The direction root refers to a set of roots describing the treatment direction or blood flow trend; The positioning root refers to a set of roots indicating the position of acupoints in the human body, corresponding to anatomical regions or clinical acupoint taking regions; The matching frequency refers to the number of times a user input keyword appears in a semantic vector matrix; The distribution density refers to the distribution concentration of a keyword in a vector dimension; The input symptom semantic coverage structure refers to the matching relationship set formed by the user input keyword in the acupoint semantic vector matrix; The path continuity condition refers to the continuity rule of acupoint node sequence arrangement in the meridian path; The path frequency refers to the number of times a target acupoint combination path is called or appears in a retrieval sample. 8.A deep learning-based knowledge retrieval system in the field of traditional Chinese acupuncture and moxibustion, characterized by, The system is used to realize the knowledge retrieval method of traditional Chinese medicine acupuncture field based on deep learning according to any one of claims 1-7, and the system comprises: The acupoint structure construction module obtains the sequence information of acupoint nodes in the meridian path, extracts the semantic description content and meridian treatment expression between adjacent acupoints, identifies and classifies the treatment root, positioning root and direction root, combines the roots with the acupoint sequence information to construct structured fragments, and labels the semantic transmission direction between acupoints to generate acupoint path vector semantic fragments; The symptom path ordering module extracts acupoint records matching the user input symptom keywords based on the acupoint node set in the acupoint path vector semantic segment, reads the efficacy level information and treatment cycle description of the associated acupoints, performs ordering processing on the acupoint nodes according to the semantic correspondence between the keywords and the treatment information, and generates a symptom corresponding path ordering result; The combination vector generation module extracts the corresponding treatment root, direction root and positioning root in the acupoint path vector semantic segment based on the acupoint nodes in the symptom corresponding path ordering result, organizes the vector representation structure according to the word class, integrates it into a combined representation set after uniform formatting, and generates an acupoint combination semantic vector matrix; The semantic structure induction module retrieves the standard word position of the user input symptom keywords based on the root dimension information in the acupoint combination semantic vector matrix, performs induction and arrangement on the occurrence frequency and distribution density of the matching roots in the vector, records the corresponding relationship between the input content and the root vector, and generates an input symptom semantic coverage structure; The retrieval path output module filters acupoint combinations that simultaneously satisfy the semantic matching condition and the path continuity feature according to the acupoint node information in the input symptom semantic coverage structure and the symptom corresponding path ordering result, arranges them according to the frequency of occurrence in the structure segment, and generates an acupoint retrieval path output sequence.

Images