A multimodal analysis method for mining traditional Chinese medicine knowledge

Abstract

The present invention relates to the technical field of TCM knowledge mining, and discloses a TCM knowledge mining method of multimodal analysis, the method comprising: collecting TCM text data and audio data and performing encoding processing; performing inter-modal entity alignment and fusion on the encoding vectors of different models, and performing TCM knowledge entity recognition on the fusion results; performing association relationship mining based on the identified TCM knowledge entities, obtaining the relationship between TCM knowledge entities, and forming a TCM knowledge network diagram. The present invention combines the maximum similarity and similarity correction degree of the text vector and the audio vector to perform similarity correction, takes the different modal vectors with the highest corrected similarity as the vectors under the same entity, realizes the aggregation processing of different modal vectors corresponding to the knowledge entity information, performs association strength calculation on the identified TCM knowledge entities, and forms a TCM knowledge network diagram representing the association relationship between different entities as the mining result.

Description

Technical Field

[0001] The present invention relates to the technical field of traditional Chinese medicine knowledge mining, and in particular to a traditional Chinese medicine knowledge mining method of multimodal analysis. Background Art

[0002] Traditional Chinese medicine knowledge mining is mainly based on text and statistical analysis. Although this method can extract key information from Chinese medicine classics to a certain extent, it has the following three limitations: First, the information is one-sided. Text data alone cannot fully reflect the multidimensional characteristics of Chinese medicine, such as visual, auditory and other perceptual information in clinical practice; second, knowledge is difficult to obtain. Some valuable Chinese medicine knowledge may be scattered in unstructured and multimedia data, and traditional analysis methods are difficult to effectively extract; third, lack of interactivity: single modal analysis cannot fully explore the potential connections and interactions between different data, reducing the depth and breadth of knowledge mining. With the gradual maturity of multimodal data processing technology, new ideas and methods are provided for Chinese medicine knowledge mining. Multimodal data includes text, sound, etc., covering more sensory information, and is expected to help reveal the inherent laws and characteristics of Chinese medicine. In view of this, the present invention proposes a Chinese medicine knowledge mining method of multimodal analysis, combining text description and sound information to conduct multimodal analysis to mine potential Chinese medicine knowledge. Summary of the invention

[0003] In view of this, the present invention provides a method for mining TCM knowledge by multimodal analysis, the purpose of which is to: 1) collect TCM text data and audio data describing TCM knowledge information, and encode them respectively, calculate similarity for different modal vectors, and perform similarity correction based on the maximum similarity of text vectors and audio vectors and the similarity correction degree, take different modal vectors with the highest corrected similarity as vectors under the same entity, and fuse the vector sequence after entity alignment to obtain an aggregated entity vector, obtain a vector representing a knowledge entity, and realize the aggregation processing of modal vectors corresponding to different knowledge entity information; 2) map the aggregated entity vector, and perform multiple rounds of residual iteration on the mapping result, retain the valid information in the aggregated entity vector, filter the invalid information, and calculate the association probability between the aggregated entity vector and different TCM knowledge entities, realize TCM knowledge entity recognition, calculate the association strength between TCM knowledge entities according to the TCM knowledge text vectors and audio vectors corresponding to different TCM knowledge entities, and form a TCM knowledge network diagram based on the association strength between TCM knowledge entities, characterize the association between different TCM knowledge in text semantic information and audio semantic information, and realize TCM knowledge mining.

[0004] To achieve the above object, the present invention provides a method for mining traditional Chinese medicine knowledge by multimodal analysis, comprising the following steps:

[0005] S1: Collect TCM text data and audio data, and encode them respectively to obtain TCM knowledge text vector sequence and audio vector sequence;

[0006] S2: Perform inter-modal entity alignment on the encoded TCM knowledge text vector sequence and audio vector sequence, and fuse the vector sequences after entity alignment to obtain an aggregated entity vector;

[0007] S3: Perform TCM knowledge entity recognition based on the fused aggregated entity vector to obtain the named entity of the current aggregated entity vector;

[0008] S4: Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities and form a TCM knowledge network diagram.

[0009] As a further improvement method of the present invention:

[0010] Optionally, collecting TCM text data and audio data in step S1 includes:

[0011] Collect TCM text data and audio data to form TCM text data set and TCM knowledge audio data set respectively:

[0012] data1={x n |n∈[1,N]}

[0013] data2={y m (t)|m∈[1,M]}

[0014] y m (t)=(y m (t1),y m (t2),...,y m (t s ),...,y m (t S ))

[0015] in:

[0016] data1 represents the Chinese medicine text data set, x n represents the nth group of TCM text data collected, N represents the total number of TCM text data collected, and each group of TCM text data is a description text of a TCM;

[0017] data2 represents the audio data set of traditional Chinese medicine knowledge, y m(t) represents the mth group of TCM knowledge audio data collected, t represents the time sequence information of the audio data, M represents the total number of TCM knowledge audio data collected, and each group of TCM knowledge audio data is a description audio of a TCM;

[0018] y m (t1),y m (t2),...,y m (t s ),...,y m (t S ) represents the audio data of traditional Chinese medicine knowledge y m (t) is the signal value at the S signal sampling time, S represents the preset length of the TCM knowledge audio data, y m (t s ) represents the audio data of traditional Chinese medicine knowledge y m (t) at the sth signal sampling time t s The signal value of the adjacent signal sampling moments is Δt; in the embodiment of the present invention, if the length of the collected audio data of traditional Chinese medicine knowledge is less than S signal sampling moments, the missing signal value is padded with 0;

[0019] The collected TCM text data and TCM knowledge audio data are encoded respectively to obtain TCM knowledge text vector sequence and TCM knowledge audio vector sequence.

[0020] Optionally, encoding the collected TCM text data to obtain a TCM knowledge text vector sequence includes:

[0021] The collected TCM text data is coded, where the TCM text data x n The encoding process is as follows:

[0022] Use word segmentation tools to analyze traditional Chinese medicine text data x n Perform word segmentation to obtain traditional Chinese medicine text data x n The word sequence of:

[0023]

[0024] in:

[0025] Represents Chinese medicine text data x n The i-th word segmentation result in num n Represents Chinese medicine text data x n The total number of word segmentation results, i∈[1,num n ];

[0026] Perform one-hot encoding on any segmentation result in the segmentation sequence to obtain a segmentation sequence after one-hot encoding, where the segmentation result The one-hot encoding result is

[0027] The one-hot encoding result is represented by a weighted encoding combined with word frequency information, where the one-hot encoding result The weighted coding formula is:

[0028]

[0029] in:

[0030] Indicates the result of one-hot encoding The weighted coding representation result of ;

[0031] Indicates word segmentation results In traditional Chinese medicine text data x n The frequency of occurrence in

[0032] W represents the encoding matrix parameters, and T represents the transpose;

[0033] Composition of TCM text data x n The corresponding TCM knowledge text vector C n :

[0034]

[0035] And form a sequence of TCM knowledge text vectors:

[0036] C=(C1,C2,...,C n ,...,C N )

[0037] in:

[0038] C represents a sequence of traditional Chinese medicine knowledge text vectors composed of N groups of traditional Chinese medicine knowledge text vectors.

[0039] Optionally, encoding the collected audio data of traditional Chinese medicine knowledge to obtain a sequence of audio vectors of traditional Chinese medicine knowledge includes:

[0040] The collected audio data of traditional Chinese medicine knowledge is encoded and processed, wherein the audio data of traditional Chinese medicine knowledge y m The encoding process of (t) is:

[0041] Construct the windowing function w(s):

[0042]

[0043] Audio data on traditional Chinese medicine knowledge m (t) is windowed, then y m (t s ) is as follows:

[0044] y′ m (t s )=y m (t s )w(s)

[0045] in:

[0046] y′ m (t s ) represents y m (t s )’s windowing result;

[0047] Construct the audio data y′ of traditional Chinese medicine knowledge after windowing m (t):

[0048] y′ m (t) = (y′ m (t1),y′ m (t2),...,y′ m (t s ),...,y′ m (t S ))

[0049] For the audio data y′ of traditional Chinese medicine knowledge after windowing m (t) Perform Fourier transform of point E, then y′ m (t) The transformation formula at Fourier transform point E is:

[0050]

[0051] in:

[0052] j represents the imaginary unit;

[0053] exp(·) represents an exponential function with a natural constant as the base;

[0054] E∈[1,S];

[0055] Y m (E) represents y′ m (t) the transformation result at Fourier transform point number E;

[0056] Generate a filter bank of U filters, where the center response frequency of the u-th filter is f u ;

[0057] Input the transformation result into the filter bank to obtain the audio data y of Chinese medicine knowledgem (t) The corresponding audio vector of traditional Chinese medicine knowledge:

[0058] R m =[R m (1),R m (2),...,R m (u),...,R m (U)]

[0059]

[0060]

[0061] in:

[0062] R m Represents audio data of traditional Chinese medicine knowledge y m (t) The corresponding audio vector of TCM knowledge;

[0063] W u (E) represents the u-th filter pair Y m (E) filter coefficient;

[0064] R e (·) indicates extracting real data;

[0065] Constructing the audio vector sequence of traditional Chinese medicine knowledge:

[0066] R=(R1,R2,...,R m ,...,R M )

[0067] in:

[0068] R represents a sequence of traditional Chinese medicine knowledge audio vectors composed of M groups of traditional Chinese medicine knowledge audio vectors.

[0069] Optionally, in step S2, the encoded TCM knowledge text vector sequence and the audio vector sequence are subjected to inter-modal entity alignment, including:

[0070] The encoded TCM knowledge text vector sequence and audio vector sequence are subjected to inter-modal entity alignment, where the inter-modal entity alignment process is as follows:

[0071] S21: Calculate the similarity between the TCM knowledge text vector and the TCM knowledge audio vector, where the TCM knowledge text vector C n Audio vector of traditional Chinese medicine knowledge R m The similarity between them is:

[0072]

[0073] in:

[0074] sim(C n ,R m ) represents the TCM knowledge text vector C n Audio vector of traditional Chinese medicine knowledge R m The similarity between

[0075] ||·||2 represents the L2 norm;

[0076] S22: Correct the calculated similarity, where the similarity sim(C n ,R m ) is:

[0077]

[0078] in:

[0079] SIM(C n ,R m ) represents the similarity sim(C n ,R m )

[0080] MAX(C n ) represents the TCM knowledge text vector C n The maximum similarity with any audio vector of traditional Chinese medicine knowledge;

[0081] MAX(R m ) represents the audio vector R of TCM knowledge m The maximum similarity with any TCM knowledge text vector;

[0082] α1 represents the similarity correction of text vectors, and α2 represents the similarity correction of audio vectors;

[0083] S23: For any TCM knowledge audio vector, select the TCM knowledge text vector with the maximum corrected similarity as the alignment result to obtain the vector sequence after entity alignment:

[0084] {g n =(C n ,Q n )|n∈[1,N]}

[0085]

[0086] in:

[0087] g n Represents the nth group of vector sequences after entity alignment;

[0088] Q n Represents the text vector C of traditional Chinese medicine knowledgen The alignment result sequence is Represents the text vector C of traditional Chinese medicine knowledge n Sum n alignment results, each alignment result corresponds to a set of TCM knowledge audio vectors, Sum n Represents the text vector C of traditional Chinese medicine knowledge n The total number of alignment results;

[0089] The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The aggregate entity vector obtained by fusion is G n .

[0090] Optionally, fusing the entity aligned vector sequences to obtain an aggregated entity vector includes:

[0091] The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The fusion formula is:

[0092]

[0093] in:

[0094] G n Represents the nth group of vector sequence g n The aggregate entity vector obtained by fusion.

[0095] Optionally, performing TCM knowledge entity recognition based on the fused aggregated entity vector in step S3 includes:

[0096] Based on the fused aggregate entity vector, TCM knowledge entity recognition is performed, where the aggregate entity vector G n The process of TCM knowledge entity recognition is as follows:

[0097] S31: Aggregate entity vector G n Mapping is performed to obtain the aggregate entity vector G n The mapping features of , where the mapping formula is:

[0098]

[0099] in:

[0100] W1 represents the mapping matrix parameters;

[0101] Represents the aggregate entity vector G n The mapping characteristics of

[0102] S32: Mapping features obtained by mapping Perform attention calculation to obtain mapping features The attention weights are:

[0103]

[0104] in:

[0105] Representation mapping features The attention weight of

[0106] S33: Mapping features Perform attention weighting to obtain the attention-weighted mapping features

[0107]

[0108] S34: Mapping features Perform K rounds of residual iterative mapping to obtain the aggregate entity vector G n The residual mapping characteristics of (K), where the k-th round residual iterative mapping formula is:

[0109]

[0110] in:

[0111] Representation mapping features The k-th round residual iterative mapping result;

[0112] W2 represents the residual mapping parameter;

[0113] S35: Residual Mapping Features Calculate the aggregate entity vector G n The probability of association with different TCM knowledge entities, where the aggregate entity vector G n The probability of association with the zth Chinese medicine knowledge entity is:

[0114]

[0115] in:

[0116] P(G n ,z) represents the aggregate entity vector G n The probability of association with the z-th TCM knowledge entity, z∈[1,Z], where Z represents the number of TCM knowledge entities, and each TCM knowledge entity corresponds to a TCM name;

[0117] Represents the association mapping matrix parameters of the z-th TCM knowledge entity;

[0118] ReLU(·) represents the ReLU activation function;

[0119] Select the TCM knowledge entity with the highest association probability as the aggregate entity vector G n The results of entity recognition of traditional Chinese medicine knowledge.

[0120] Optionally, in step S4, performing association mining based on the identified TCM knowledge entities to obtain the relationships between the TCM knowledge entities includes:

[0121] Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities. The association mining process between N TCM knowledge entities is as follows:

[0122] S41: Obtain the entity aligned vector sequence corresponding to N traditional Chinese medicine knowledge entities, where the vector sequence corresponding to the nth traditional Chinese medicine knowledge entity is g n =(C n ,Q n );

[0123] S42: merging vector sequences with the same TCM knowledge entity to obtain V groups of merged vector sequences, and accumulating and averaging the TCM knowledge text vectors in the merged vector sequences to obtain a merged TCM knowledge text vector for each group of vector sequences;

[0124] S43: Calculate the similarity of the merged TCM knowledge text vectors of any two groups of vector sequences, where the similarity of the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences is:

[0125]

[0126] in:

[0127] cor(v,b) represents the similarity between the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences;

[0128] Con v represents the merged TCM knowledge text vector in the vth group of vector sequences, Con b represents the merged TCM knowledge text vector in the b-th group of vector sequences;

[0129] S44: Calculate the association strength between any two TCM knowledge entities, where the association strength between the TCM knowledge entities corresponding to the vth group of vector sequences and the bth group of vector sequences is:

[0130]

[0131] in:

[0132] L(v,b) represents the strength of association between the TCM knowledge entities corresponding to the vth vector sequence and the bth vector sequence.

[0133] AVG(Con v ) represents the combined TCM knowledge text vector Con v The average corrected similarity between all TCM knowledge audio vectors, AVG (Con v ,b) represents the merged TCM knowledge text vector Con v The average corrected similarity between the audio vectors of TCM knowledge in the b-th group of vector sequences, wherein the calculation process of the corrected similarity is from step S21 to step S22;

[0134] Based on the association strength between TCM knowledge entities, a TCM knowledge network diagram is formed, in which the nodes in the TCM knowledge network diagram are TCM knowledge entities, as well as the TCM text data and audio data corresponding to the TCM knowledge entities. Any two nodes constitute a network edge, and the weight of the network edge is the association strength between the corresponding two TCM knowledge entities.

[0135] In order to solve the above problem, the present invention provides an electronic device, the electronic device comprising:

[0136] A memory storing at least one instruction;

[0137] Communication interface, enabling electronic equipment to communicate; and

[0138] A processor executes instructions stored in the memory to implement the above-mentioned multimodal analysis method for mining traditional Chinese medicine knowledge.

[0139] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned multimodal analysis method for mining knowledge of traditional Chinese medicine.

[0140] Compared with the prior art, the present invention proposes a method for mining TCM knowledge based on multimodal analysis, which has the following advantages:

[0141] First, this scheme proposes an entity alignment method for different modal vectors, and performs inter-modal entity alignment on the encoded TCM knowledge text vector sequence and audio vector sequence. The inter-modal entity alignment process is as follows: calculate the similarity between the TCM knowledge text vector and the TCM knowledge audio vector, where the TCM knowledge text vector C n Audio vector of traditional Chinese medicine knowledge R m The similarity between them is:

[0142]

[0143] Where: sim(C n ,R m ) represents the TCM knowledge text vector C n Audio vector of traditional Chinese medicine knowledge R m The similarity between them; ||·||2 represents the L2 norm; the calculated similarity is corrected, where the similarity sim(C n ,R m ) is:

[0144]

[0145] Where: SIM(C n ,R m ) represents the similarity sim(C n ,R m ) is the corrected result; MAX(C n ) represents the TCM knowledge text vector C n The maximum similarity with any TCM knowledge audio vector; MAX(R m ) represents the audio vector R of TCM knowledge m The maximum similarity with any TCM knowledge text vector; α1 represents the similarity correction degree of the text vector, and α2 represents the similarity correction degree of the audio vector; for any TCM knowledge audio vector, the TCM knowledge text vector with the maximum similarity after correction is selected as the alignment result, and the vector sequence after entity alignment is obtained:

[0146] {g n =(C n ,Q n )|n∈[1,N]}

[0147]

[0148] Where: g n represents the nth group of vector sequences after entity alignment; Q n Represents the text vector C of traditional Chinese medicine knowledge n The alignment result sequence is Represents the text vector C of traditional Chinese medicine knowledge n Sum n alignment results, each alignment result corresponds to a set of TCM knowledge audio vectors, Sum n Represents the text vector C of traditional Chinese medicine knowledge n The total number of alignment results; the vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The aggregate entity vector obtained by fusion is Gn The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The fusion formula is:

[0149]

[0150] Where: G n Represents the nth group of vector sequence g n The fused aggregated entity vector. This solution collects TCM text data and audio data describing TCM knowledge information, and encodes them separately, calculates the similarity of different modal vectors, and corrects the similarity by combining the maximum similarity between the text vector and the audio vector and the similarity correction degree. The different modal vectors with the highest corrected similarity are taken as vectors under the same entity, and the vector sequence after entity alignment is fused to obtain the aggregated entity vector, and the vector representing a knowledge entity is obtained, realizing the aggregation processing of modal vectors corresponding to different knowledge entity information.

[0151] At the same time, this scheme proposes an entity recognition method to perform TCM knowledge entity recognition based on the fused aggregate entity vector, where the aggregate entity vector G n The TCM knowledge entity recognition process is as follows: aggregate entity vector G n Mapping is performed to obtain the aggregate entity vector G n The mapping features of , where the mapping formula is:

[0152]

[0153] Where: W1 represents the mapping matrix parameter; Represents the aggregate entity vector G n The mapping features of the mapping; the mapping features obtained by mapping Perform attention calculation to obtain mapping features The attention weights are:

[0154]

[0155] in: Representation mapping features The attention weight of the mapping feature Perform attention weighting to obtain the attention-weighted mapping features

[0156]

[0157] Mapping features Perform K rounds of residual iterative mapping to obtain the aggregate entity vector G n The residual mapping characteristics of The k-th round residual iterative mapping formula is:

[0158]

[0159] in: Representation mapping features The kth round of residual iterative mapping result; W2 represents the residual mapping parameter; based on the residual mapping feature Calculate the aggregate entity vector G n The probability of association with different TCM knowledge entities, where the aggregate entity vector G n The probability of association with the zth Chinese medicine knowledge entity is:

[0160]

[0161] Where: P(G n ,z) represents the aggregate entity vector G n The probability of association with the zth Chinese medicine knowledge entity, z

[0162] ∈[1,Z], Z represents the number of types of TCM knowledge entities, and each TCM knowledge entity corresponds to a TCM name; represents the association mapping matrix parameters of the zth TCM knowledge entity; ReLU(·) represents the ReLU activation function; the TCM knowledge entity with the highest association probability is selected as the aggregate entity vector G n The results of TCM knowledge entity recognition. This scheme maps the aggregated entity vectors and performs multiple rounds of residual iterations on the mapping results, retains the valid information in the aggregated entity vectors, filters the invalid information, and calculates the association probability between the aggregated entity vectors and different TCM knowledge entities to realize TCM knowledge entity recognition. According to the TCM knowledge text vectors and audio vectors corresponding to different TCM knowledge entities, the association strength between TCM knowledge entities is calculated. Based on the association strength between TCM knowledge entities, a TCM knowledge network diagram is formed to characterize the association between different TCM knowledge in text semantic information and audio semantic information, thereby realizing TCM knowledge mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0163] Figure 1 A schematic diagram of a flow chart of a method for mining traditional Chinese medicine knowledge using multimodal analysis provided by an embodiment of the present invention;

[0164] Figure 2 A schematic diagram of the structure of an electronic device for implementing a method for mining knowledge of traditional Chinese medicine using multimodal analysis provided by an embodiment of the present invention.

[0165] In the figure: 1 electronic device, 10 processor, 11 memory, 12 program, 13 communication interface.

[0166] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0167] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0168] The embodiment of the present application provides a method for mining knowledge of traditional Chinese medicine by multimodal analysis. The execution subject of the method for mining knowledge of traditional Chinese medicine by multimodal analysis includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for mining knowledge of traditional Chinese medicine by multimodal analysis can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0169] Embodiment 1:

[0170] S1: Collect TCM text data and audio data, and encode them separately to obtain TCM knowledge text vector sequence and audio vector sequence.

[0171] The step S1 includes collecting Chinese medicine text data and audio data, including:

[0172] Collect TCM text data and audio data to form TCM text data set and TCM knowledge audio data set respectively:

[0173] data1={x n |n∈[1,N]}

[0174] data2={y m (t)|m∈[1,M]}

[0175] y m (t)=(y m (t1),y m (t2),...,y m (t s ),...,y m (t S ))

[0176] in:

[0177] data1 represents the Chinese medicine text data set, x n represents the nth group of TCM text data collected, N represents the total number of TCM text data collected, and each group of TCM text data is a description text of a TCM;

[0178] data2 represents the audio data set of traditional Chinese medicine knowledge, y m (t) represents the mth group of TCM knowledge audio data collected, t represents the time sequence information of the audio data, M represents the total number of TCM knowledge audio data collected, and each group of TCM knowledge audio data is a description audio of a TCM;

[0179] y m (t1),y m (t2),...,y m (t s ),...,y m (t S ) represents the audio data of traditional Chinese medicine knowledge y m (t) is the signal value at the S signal sampling time, S represents the preset length of the TCM knowledge audio data, y m (t s ) represents the audio data of traditional Chinese medicine knowledge y m (t) at the sth signal sampling time t s The signal value of the adjacent signal sampling moments is Δt; in the embodiment of the present invention, if the length of the collected audio data of traditional Chinese medicine knowledge is less than S signal sampling moments, the missing signal value is padded with 0;

[0180] The collected TCM text data and TCM knowledge audio data are encoded respectively to obtain TCM knowledge text vector sequence and TCM knowledge audio vector sequence.

[0181] The encoding process of the collected TCM text data to obtain a TCM knowledge text vector sequence includes:

[0182] The collected TCM text data is coded, where the TCM text data x n The encoding process is as follows:

[0183] Use word segmentation tools to analyze traditional Chinese medicine text data x n Perform word segmentation to obtain traditional Chinese medicine text data x n The word sequence of:

[0184]

[0185] in:

[0186] Represents Chinese medicine text data x n The i-th word segmentation result in num n Represents Chinese medicine text data x n The total number of word segmentation results, i∈[1,num n ];

[0187] Perform one-hot encoding on any segmentation result in the segmentation sequence to obtain a segmentation sequence after one-hot encoding, where the segmentation result The one-hot encoding result is

[0188] The one-hot encoding result is represented by a weighted encoding combined with word frequency information, where the one-hot encoding result The weighted coding formula is:

[0189]

[0190] in:

[0191] Indicates the result of one-hot encoding The weighted coding representation result of ;

[0192] Indicates word segmentation results In traditional Chinese medicine text data x n The frequency of occurrence in

[0193] W represents the encoding matrix parameters, and T represents the transpose;

[0194] Composition of TCM text data x n The corresponding TCM knowledge text vector C n :

[0195]

[0196] And form a sequence of TCM knowledge text vectors:

[0197] C=(C1,C2,...,C n ,...,C N )

[0198] in:

[0199] C represents a sequence of traditional Chinese medicine knowledge text vectors composed of N groups of traditional Chinese medicine knowledge text vectors.

[0200] The encoding process of the collected audio data of traditional Chinese medicine knowledge to obtain the audio vector sequence of traditional Chinese medicine knowledge includes:

[0201] The collected audio data of traditional Chinese medicine knowledge is encoded and processed, wherein the audio data of traditional Chinese medicine knowledge y m The encoding process of (t) is:

[0202] Construct the windowing function w(s):

[0203]

[0204] Audio data on traditional Chinese medicine knowledge m (t) is windowed, then y m (t s ) is as follows:

[0205] y′ m (t s )=y m (t s )w(s)

[0206] in:

[0207] y′ m (t s ) represents y m (t s )’s windowing result;

[0208] Construct the audio data y′ of traditional Chinese medicine knowledge after windowing m (t):

[0209] y′ m (t) = (y′ m (t1),y′ m (t2),...,y′ m (t s ),...,y′ m (t S ))

[0210] For the audio data y′ of traditional Chinese medicine knowledge after windowing m (t) Perform Fourier transform of point E, then y′ m (t) The transformation formula at Fourier transform point E is:

[0211]

[0212] in:

[0213] j represents the imaginary unit;

[0214] exp(·) represents an exponential function with a natural constant as the base;

[0215] E∈[1,S];

[0216] Y m (E) represents y′ m (t) the transformation result at Fourier transform point number E;

[0217] Generate a filter bank of U filters, where the center response frequency of the u-th filter is f u ;

[0218] Input the transformation result into the filter bank to obtain the audio data y of Chinese medicine knowledge m (t) The corresponding audio vector of traditional Chinese medicine knowledge:

[0219] R m =[R m (1),R m (2),...,R m (u),...,R m (U)]

[0220]

[0221]

[0222] in:

[0223] R m Represents audio data of traditional Chinese medicine knowledge y m (t) The corresponding audio vector of TCM knowledge;

[0224] W u (E) represents the u-th filter pair Y m (E) filter coefficient;

[0225] Re(·) means extracting the real part data;

[0226] Constructing the audio vector sequence of traditional Chinese medicine knowledge:

[0227] R=(R1,R2,...,R m ,...,R M )

[0228] in:

[0229] R represents a sequence of traditional Chinese medicine knowledge audio vectors composed of M groups of traditional Chinese medicine knowledge audio vectors.

[0230] S2: Perform inter-modal entity alignment on the encoded TCM knowledge text vector sequence and audio vector sequence, and fuse the entity-aligned vector sequences to obtain the aggregated entity vector.

[0231] In the step S2, the encoded TCM knowledge text vector sequence and the audio vector sequence are subjected to inter-modal entity alignment, including:

[0232] The encoded TCM knowledge text vector sequence and audio vector sequence are subjected to inter-modal entity alignment, where the inter-modal entity alignment process is as follows:

[0233] S21: Calculate the similarity between the TCM knowledge text vector and the TCM knowledge audio vector, where the TCM knowledge text vector C nAudio vector of traditional Chinese medicine knowledge R m The similarity between them is:

[0234]

[0235] in:

[0236] sim(C n ,R m ) represents the TCM knowledge text vector C n Audio vector of traditional Chinese medicine knowledge R m The similarity between

[0237] ||·||2 represents the L2 norm;

[0238] S22: Correct the calculated similarity, where the similarity sim(C n ,R m ) is:

[0239]

[0240] in:

[0241] SIM(C n ,R m ) represents the similarity sim(C n ,R m )

[0242] MAX(C n ) represents the TCM knowledge text vector C n The maximum similarity with any audio vector of traditional Chinese medicine knowledge;

[0243] MAX(R m ) represents the audio vector R of TCM knowledge m The maximum similarity with any TCM knowledge text vector;

[0244] α1 represents the similarity correction of text vectors, and α2 represents the similarity correction of audio vectors;

[0245] S23: For any TCM knowledge audio vector, select the TCM knowledge text vector with the maximum corrected similarity as the alignment result to obtain the vector sequence after entity alignment:

[0246] {g n =(C n ,Q n )|n∈[1,N]}

[0247]

[0248] in:

[0249] g n Represents the nth group of vector sequences after entity alignment;

[0250] Q n Represents the text vector C of traditional Chinese medicine knowledge n The alignment result sequence is Represents the text vector C of traditional Chinese medicine knowledge n Sum n alignment results, each alignment result corresponds to a set of TCM knowledge audio vectors, Sum n Represents the text vector C of traditional Chinese medicine knowledge n The total number of alignment results;

[0251] The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The aggregate entity vector obtained by fusion is G n .

[0252] The step of fusing the entity aligned vector sequences to obtain an aggregated entity vector includes:

[0253] The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences g n The fusion formula is:

[0254]

[0255] in:

[0256] G n Represents the nth group of vector sequence g n The aggregate entity vector obtained by fusion.

[0257] S3: Perform TCM knowledge entity recognition based on the fused aggregated entity vector to obtain the named entity of the current aggregated entity vector.

[0258] In the step S3, the entity recognition of TCM knowledge is performed based on the fused aggregated entity vector, including:

[0259] Based on the fused aggregate entity vector, TCM knowledge entity recognition is performed, where the aggregate entity vector G n The process of TCM knowledge entity recognition is as follows:

[0260] S31: Aggregate entity vector G n Mapping is performed to obtain the aggregate entity vector G n The mapping characteristics of , where the mapping formula is:

[0261]

[0262] in:

[0263] W1 represents the mapping matrix parameters;

[0264] Represents the aggregate entity vector G n The mapping characteristics of

[0265] S32: Mapping features obtained by mapping Perform attention calculation to obtain mapping features The attention weights are:

[0266]

[0267] in:

[0268] Representation mapping features The attention weight of

[0269] S33: Mapping features Perform attention weighting to obtain the attention-weighted mapping features

[0270]

[0271] S34: Mapping features Perform K rounds of residual iterative mapping to obtain the aggregate entity vector G n The residual mapping characteristics of (K), where the k-th round residual iterative mapping formula is:

[0272]

[0273] in:

[0274] Representation mapping features The k-th round residual iterative mapping result;

[0275] W2 represents the residual mapping parameter;

[0276] S35: Residual Mapping Features Calculate the aggregate entity vector G n The probability of association with different TCM knowledge entities, where the aggregate entity vector G n The probability of association with the zth Chinese medicine knowledge entity is:

[0277]

[0278] in:

[0279] P(G n,z) represents the aggregate entity vector G n The probability of association with the z-th TCM knowledge entity, z∈[1,Z], where Z represents the number of TCM knowledge entities, and each TCM knowledge entity corresponds to a TCM name;

[0280] Represents the association mapping matrix parameters of the z-th TCM knowledge entity;

[0281] ReLU(·) represents the ReLU activation function;

[0282] Select the TCM knowledge entity with the highest association probability as the aggregate entity vector G n The results of entity recognition of traditional Chinese medicine knowledge.

[0283] S4: Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities and form a TCM knowledge network diagram.

[0284] In step S4, association mining is performed based on the identified TCM knowledge entities to obtain the relationships between the TCM knowledge entities, including:

[0285] Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities. The association mining process between N TCM knowledge entities is as follows:

[0286] S41: Obtain the entity aligned vector sequence corresponding to N traditional Chinese medicine knowledge entities, where the vector sequence corresponding to the nth traditional Chinese medicine knowledge entity is g n =(C n ,Q n );

[0287] S42: merging vector sequences with the same TCM knowledge entity to obtain V groups of merged vector sequences, and accumulating and averaging the TCM knowledge text vectors in the merged vector sequences to obtain a merged TCM knowledge text vector for each group of vector sequences;

[0288] S43: Calculate the similarity of the merged TCM knowledge text vectors of any two groups of vector sequences, where the similarity of the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences is:

[0289]

[0290] in:

[0291] cor(v,b) represents the similarity between the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences;

[0292] Con v represents the merged TCM knowledge text vector in the vth group of vector sequences, Con b represents the merged TCM knowledge text vector in the b-th group of vector sequences;

[0293] S44: Calculate the association strength between any two TCM knowledge entities, where the association strength between the TCM knowledge entities corresponding to the vth group of vector sequences and the bth group of vector sequences is:

[0294]

[0295] in:

[0296] L(v,b) represents the strength of association between the TCM knowledge entities corresponding to the vth vector sequence and the bth vector sequence.

[0297] AVG(Con v ) represents the combined TCM knowledge text vector Con v The average corrected similarity between all TCM knowledge audio vectors, AVG (Con v ,b) represents the combined TCM knowledge text vector Con v The average corrected similarity between the audio vectors of TCM knowledge in the b-th group of vector sequences, wherein the calculation process of the corrected similarity is from step S21 to step S22;

[0298] Based on the association strength between TCM knowledge entities, a TCM knowledge network diagram is formed, in which the nodes in the TCM knowledge network diagram are TCM knowledge entities, as well as the TCM text data and audio data corresponding to the TCM knowledge entities. Any two nodes constitute a network edge, and the weight of the network edge is the association strength between the corresponding two TCM knowledge entities.

[0299] Embodiment 2:

[0300] like Figure 2 , which is a schematic diagram of the structure of an electronic device for implementing a method for mining knowledge of traditional Chinese medicine using multimodal analysis provided by an embodiment of the present invention.

[0301] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0302] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 may be used not only to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but also to temporarily store data that has been output or is to be output.

[0303] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as the program 12 for realizing multimodal analysis of TCM knowledge mining), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0304] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to achieve connection and communication between internal components of the electronic device.

[0305] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0306] Figure 2 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0307] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0308] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0309] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0310] The program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0311] Collecting TCM text data and audio data, and encoding them respectively to obtain TCM knowledge text vector sequence and audio vector sequence;

[0312] Perform inter-modal entity alignment on the encoded TCM knowledge text vector sequence and audio vector sequence, and fuse the entity-aligned vector sequences to obtain an aggregated entity vector;

[0313] Based on the fused aggregated entity vector, traditional Chinese medicine knowledge entity recognition is performed to obtain the named entity of the current aggregated entity vector;

[0314] Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities and form a TCM knowledge network diagram.

[0315] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figure 1 to Figure 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0316] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0317] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0318] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for mining traditional Chinese medicine knowledge based on multimodal analysis, characterized in that: The method comprises: S1: Collect TCM text data and audio data, and encode them respectively to obtain TCM knowledge text vector sequence and audio vector sequence; Collect TCM text data and audio data to form TCM text data set and TCM knowledge audio data set respectively: ； ； ； in: Represents a collection of traditional Chinese medicine text data. represents the nth group of TCM text data collected, N represents the total number of TCM text data collected, and each group of TCM text data is a description text of a TCM; Represents a collection of audio data on traditional Chinese medicine knowledge. represents the mth group of TCM knowledge audio data collected, t represents the time sequence information of the audio data, M represents the total number of TCM knowledge audio data collected, and each group of TCM knowledge audio data is a description audio of a TCM; Represents audio data of traditional Chinese medicine knowledge The signal values ​​at the S signal sampling moments, S represents the preset length of the audio data of traditional Chinese medicine knowledge, Represents audio data of traditional Chinese medicine knowledge At the sth signal sampling time The signal value of the adjacent signal sampling time interval is ; The collected TCM text data and TCM knowledge audio data are respectively encoded to obtain a TCM knowledge text vector sequence and a TCM knowledge audio vector sequence; S2: Perform inter-modal entity alignment on the encoded TCM knowledge text vector sequence and audio vector sequence, and fuse the vector sequences after entity alignment to obtain an aggregated entity vector; The inter-modality entity alignment process is: S21: Calculate the similarity between the TCM knowledge text vector and the TCM knowledge audio vector, where the TCM knowledge text vector Audio vector with Chinese medicine knowledge The similarity between them is: ； in: Represents the text vector of traditional Chinese medicine knowledge Audio vector with Chinese medicine knowledge The similarity between represents the L2 norm; S22: Correct the calculated similarity, where the similarity The correction formula is: ； in: Indicates similarity The correction result of Represents the text vector of traditional Chinese medicine knowledge The maximum similarity with any audio vector of traditional Chinese medicine knowledge; Representing audio vectors of traditional Chinese medicine knowledge The maximum similarity with any TCM knowledge text vector; Represents the similarity correction degree of the text vector, Represents the similarity correction degree of the audio vector; S23: For any TCM knowledge audio vector, select the TCM knowledge text vector with the maximum corrected similarity as the alignment result to obtain the vector sequence after entity alignment: ； ； in: Represents the nth group of vector sequences after entity alignment; Represents the text vector of traditional Chinese medicine knowledge The alignment result sequence is Represents the text vector of traditional Chinese medicine knowledge of alignment results, each of which corresponds to a set of TCM knowledge audio vectors. Represents the text vector of traditional Chinese medicine knowledge The total number of alignment results; The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences The aggregate entity vector obtained by fusion is ; The vector sequences after entity alignment are fused to obtain the aggregated entity vector, where the nth group of vector sequences The fusion formula is: ； in: Represents the nth group of vector sequences The aggregate entity vector obtained by fusion; S3: Perform TCM knowledge entity recognition based on the fused aggregated entity vector to obtain the named entity of the current aggregated entity vector; Aggregate entity vector The process of TCM knowledge entity recognition is as follows: S31: Aggregate entity vector Mapping is performed to obtain the aggregate entity vector The mapping characteristics of , where the mapping formula is: ； in: Represents the mapping matrix parameters; Represents the aggregate entity vector The mapping characteristics of S32: Mapping features obtained by mapping Perform attention calculation to obtain mapping features The attention weights are: ； in: Representation mapping features The attention weight of S33: Mapping features Perform attention weighting to obtain the attention-weighted mapping features : S34: Mapping features Perform K rounds of residual iterative mapping to obtain the aggregate entity vector The residual mapping characteristics of , where the k-th round residual iterative mapping formula is: ； in: Representation mapping features No. Round residual iterative mapping result; represents the residual mapping parameters; S35: Residual Mapping Features , calculate the aggregate entity vector The probability of association with different TCM knowledge entities, where the aggregate entity vector The probability of association with the zth Chinese medicine knowledge entity is: ； in: Represents the aggregate entity vector The probability of association with the zth Chinese medicine knowledge entity, , Indicates the number of types of TCM knowledge entities, each of which corresponds to a TCM name; Indicates Association mapping matrix parameters of TCM knowledge entities; ReLU activation function. Select the TCM knowledge entity with the highest association probability as the aggregate entity vector The results of TCM knowledge entity recognition; S4: Based on the identified TCM knowledge entities, association mining is performed to obtain the relationships between TCM knowledge entities and form a TCM knowledge network diagram; The process of mining the association relationship between TCM knowledge entities is as follows: S41: Obtain the entity aligned vector sequence corresponding to N traditional Chinese medicine knowledge entities, where the vector sequence corresponding to the nth traditional Chinese medicine knowledge entity is ; S42: merging vector sequences with the same TCM knowledge entity to obtain V groups of merged vector sequences, and accumulating and averaging the TCM knowledge text vectors in the merged vector sequences to obtain a merged TCM knowledge text vector for each group of vector sequences; S43: Calculate the similarity of the merged TCM knowledge text vectors of any two groups of vector sequences, where the similarity of the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences is: ； in: Represents the similarity between the merged TCM knowledge text vectors in the vth group of vector sequences and the bth group of vector sequences; represents the merged TCM knowledge text vector in the vth group of vector sequences, represents the merged TCM knowledge text vector in the b-th group of vector sequences; S44: Calculate the association strength between any two TCM knowledge entities, where the association strength between the TCM knowledge entities corresponding to the vth group of vector sequences and the bth group of vector sequences is: ； in: Represents the strength of association between the TCM knowledge entities corresponding to the vth group of vector sequences and the bth group of vector sequences Represents the merged TCM knowledge text vector The average corrected similarity between all TCM knowledge audio vectors, Represents the merged TCM knowledge text vector The average corrected similarity between the audio vectors of TCM knowledge in the b-th group of vector sequences, wherein the calculation process of the corrected similarity is from step S21 to step S22; Based on the association strength between TCM knowledge entities, a TCM knowledge network diagram is formed, in which the nodes in the TCM knowledge network diagram are TCM knowledge entities, as well as the TCM text data and audio data corresponding to the TCM knowledge entities. Any two nodes constitute a network edge, and the weight of the network edge is the association strength between the corresponding two TCM knowledge entities.

2. The method for mining traditional Chinese medicine knowledge by multimodal analysis according to claim 1, characterized in that: The encoding process of the collected TCM text data to obtain a TCM knowledge text vector sequence includes: The collected TCM text data is coded and processed, where the TCM text data The encoding process is as follows: Using word segmentation tools to analyze traditional Chinese medicine text data Perform word segmentation to obtain traditional Chinese medicine text data The word sequence of: ； in: Represents traditional Chinese medicine text data The i-th word segmentation result in Represents traditional Chinese medicine text data The total number of word segmentation results, ; Perform one-hot encoding on any word segmentation result in the word segmentation sequence to obtain a word segmentation sequence after one-hot encoding, where the word segmentation result The one-hot encoding result is ; The one-hot encoding result is represented by a weighted encoding combined with word frequency information, where the one-hot encoding result The weighted coding formula is: ； in: Indicates the result of one-hot encoding The weighted coding representation result of ; Indicates word segmentation results In traditional Chinese medicine text data The frequency of occurrence in W represents the encoding matrix parameters, and T represents the transpose; Constructing TCM text data The corresponding TCM knowledge text vector : ； And form a sequence of TCM knowledge text vectors: ； in: Represents a sequence of traditional Chinese medicine knowledge text vectors composed of N groups of traditional Chinese medicine knowledge text vectors.

3. The method for mining traditional Chinese medicine knowledge by multimodal analysis according to claim 1, characterized in that: The encoding process of the collected audio data of traditional Chinese medicine knowledge to obtain the audio vector sequence of traditional Chinese medicine knowledge includes: The collected audio data of traditional Chinese medicine knowledge is encoded and processed, wherein the audio data of traditional Chinese medicine knowledge The encoding process is as follows: Constructing the windowing function : ； Audio data on traditional Chinese medicine knowledge If windowing is performed, The windowing result is: in: express The windowing result of Composition of audio data of traditional Chinese medicine knowledge after windowing : ； Audio data of traditional Chinese medicine knowledge after windowing Perform Fourier transform of point E, then The transformation formula at the Fourier transform point E is: ； in: j represents the imaginary unit; represents an exponential function with a natural constant as base; ； express The transformation result at Fourier transform point number E; Generate a filter bank of U filters, where the center response frequency of the u-th filter is ; Input the transformation result into the filter bank to obtain audio data of traditional Chinese medicine knowledge The corresponding audio vector of traditional Chinese medicine knowledge: ； ； ； in: Represents audio data of traditional Chinese medicine knowledge The corresponding audio vector of traditional Chinese medicine knowledge; represents the u-th filter pair The filter coefficient of Indicates the extraction of real data; Constructing the audio vector sequence of traditional Chinese medicine knowledge: ； in: R represents a sequence of traditional Chinese medicine knowledge audio vectors composed of M groups of traditional Chinese medicine knowledge audio vectors.

Images