Biomedical Named Entity Recognition Method and System

Abstract

The present invention discloses a method and system for biomedical named entity recognition; obtaining a biomedical statement to be recognized; dividing the biomedical statement into biomedical words and biomedical characters; inputting the biomedical statement, biomedical words, and biomedical characters into a trained biomedical named entity recognition model, and outputting the biomedical named entity recognition result; the trained model respectively extracts the word embedding feature, radical feature, radical feature, and grammatical feature of the characters from the biomedical statement, biomedical words, and biomedical characters; splicing the features to obtain a first spliced feature; performing a self-attention mechanism process on the first spliced feature to obtain a processed first spliced feature; splicing the stroke feature with the processed first spliced feature to obtain a second spliced feature; encoding and decoding the second spliced feature to obtain a biomedical named entity recognition label.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and particularly to a method and system for biomedical named entity recognition. Background Art

[0002] The statements in this section only mention the background art related to the present invention and do not necessarily constitute prior art.

[0003] Named Entity Recognition (NER) is an important task in natural language processing. Its main function is to identify entities with specific meanings in the text and identify entity categories according to requirements in a specific field. For example, in the medical field, named entity recognition mainly identifies the patient's name, home address, body parts, symptoms, treatment methods, etc.

[0004] Named entities generally refer to entities with specific meanings or strong referentiality in the text, usually including two parts: identifying the boundary of the entity and determining the type of the entity. Therefore, the entity range is very wide, and text nouns required in their respective fields can be called entities.

[0005] The main task of Biomedical Named Entity Recognition is to identify named entities such as genes, diseases, and chemical drugs in biomedical field texts. However, these entities are usually composed of very long words, each entity has multiple variant names, and the numbers and letters that make up the entity may represent different meanings. Therefore, compared with entity recognition in ordinary fields, biomedical named entity recognition is more difficult.

[0006] The research methods of named entity recognition mainly include rule - and dictionary - based methods, traditional machine learning - based methods, deep learning - based methods, and multi - task learning - based methods. Rule - and dictionary - based methods rely too much on the establishment of the initial database and can only be applied to entity recognition in specific fields, resulting in too one - sided application fields. Traditional machine learning methods are based on large - scale labeled data sets, mainly including supervised learning, semi - supervised learning, and unsupervised learning. However, traditional supervised learning methods require a lot of time to design features, and these features determine the performance of the model. Summary of the Invention

[0007] To solve the deficiencies of the prior art, the present invention provides a method and system for biomedical named entity recognition; this method can more accurately identify biomedical named entities and improve medical information.

[0008] In the first aspect, the present invention provides a method for biomedical named entity recognition;

[0009] A method for biomedical named entity recognition, including:

[0010] Obtain a biomedical statement to be recognized; divide the biomedical statement into biomedical words and biomedical characters;

[0011] Input the biomedical statement, biomedical words, and biomedical characters into the trained biomedical named entity recognition model, and output the biomedical named entity recognition result;

[0012] Among them, the working principle of the trained biomedical named entity recognition model includes: respectively extracting the word embedding feature, radical feature, radical feature, and grammar feature of the characters from the biomedical statement, biomedical words, and biomedical characters; splicing the word embedding feature, radical feature, radical feature, and grammar feature of the characters to obtain the first splicing feature; then, performing self-attention mechanism processing on the first splicing feature to obtain the processed first splicing feature; splicing the stroke feature with the processed first splicing feature to obtain the second splicing feature; encoding and decoding the second splicing feature to obtain the biomedical named entity recognition label.

[0013] In a second aspect, the present invention provides a biomedical named entity recognition system;

[0014] The biomedical named entity recognition system includes:

[0015] An acquisition module, which is configured to: obtain a biomedical statement to be recognized; divide the biomedical statement into biomedical words and biomedical characters;

[0016] A recognition module, which is configured to: input the biomedical statement, biomedical words, and biomedical characters into the trained biomedical named entity recognition model, and output the biomedical named entity recognition result;

[0017] Among them, the working principle of the trained biomedical named entity recognition model includes: respectively extracting the word embedding feature, radical feature, radical feature, and grammar feature of the characters from the biomedical statement, biomedical words, and biomedical characters; splicing the word embedding feature, radical feature, radical feature, and grammar feature of the characters to obtain the first splicing feature; then, performing self-attention mechanism processing on the first splicing feature to obtain the processed first splicing feature; splicing the stroke feature with the processed first splicing feature to obtain the second splicing feature; encoding and decoding the second splicing feature to obtain the biomedical named entity recognition label.

[0018] In a third aspect, the present invention also provides an electronic device, including:

[0019] A memory for non - transiently storing computer - readable instructions; and

[0020] A processor for running the computer - readable instructions,

[0021] wherein, when the computer - readable instructions are run by the processor, the method described in the first aspect above is executed.

[0022] In a fourth aspect, the present invention also provides a storage medium that non - transiently stores computer - readable instructions, wherein when the non - transient computer - readable instructions are executed by a computer, the instructions for executing the method described in the first aspect are executed.

[0023] In a fifth aspect, the present invention also provides a computer program product, including a computer program, where the computer program is used to implement the method described in the first aspect above when running on one or more processors.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] It can fully extract the features of biomedical entities; further effectively identify biomedical named entities through the accurately extracted features; and can effectively improve medical information. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0027] Figure 1 It is a structural diagram of the named - entity recognition model for Example 1;

[0028] Figure 2 It is a schematic diagram of the connection relationship of the pre - trained language model, self - attention mechanism (Self - Attention) and Bi - GRU for Example 1;

[0029] Figures 3(a)-3(c) It is the 3 - gram, 4 - gram, 5 - gram obtained for each character in Example 1;

[0030] Figure 4 It is a schematic diagram of the internal structure of the long - short - term memory network LSTM for Example 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] It should be noted that the following detailed description is exemplary and is intended to provide a further description of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0033] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0034] All data acquisition in this embodiment is a legal application of data on the basis of compliance with laws, regulations and user consent.

[0035] Deep learning avoids cumbersome feature engineering by automatically retrieving word and character features. Therefore, deep learning-based methods are widely used in named entity recognition. Among them, the model based on the recurrent neural network is the general standard for named entity recognition. The Recurrent Neural Network (RNN) can capture the context information of a sentence and is good at sequence tasks, but can only retain recent information. Therefore, the Long-Short Term Memory model (LSTM) and the Bidirectional Long Short-Term Memory model (Bi-LSTM) are proposed to retain bidirectional historical information.

[0036] Embodiment 1

[0037] This embodiment provides a method for biomedical named entity recognition;

[0038] As Figure 1 shown, the method for biomedical named entity recognition includes:

[0039] S101: Obtain the biomedical sentence to be recognized; divide the biomedical sentence into biomedical words and biomedical characters;

[0040] S102: Input the biomedical sentence, biomedical words and biomedical characters into the trained biomedical named entity recognition model, and output the biomedical named entity recognition result;

[0041] Among them, the working principle of the trained biomedical named entity recognition model includes: extracting the word embedding features, radical features, radical head features, and syntactic features of characters from biomedical sentences, biomedical vocabulary, and biomedical characters respectively; splicing the word embedding features, radical features, radical head features, and syntactic features of characters to obtain the first spliced feature; then, performing self-attention mechanism processing on the first spliced feature to obtain the processed first spliced feature; splicing the stroke feature with the processed first spliced feature to obtain the second spliced feature; encoding and decoding the second spliced feature to obtain the biomedical named entity recognition label.

[0042] Further, the division of the biomedical sentence into biomedical vocabulary is based on a dictionary-based method or a string matching method.

[0043] Further, the division of the biomedical sentence into biomedical characters is based on a string matching method.

[0044] Further, the extraction of the word embedding features, radical features, radical head features, and syntactic features of characters from biomedical sentences, biomedical vocabulary, and biomedical characters respectively specifically includes:

[0045] Extracting the word embedding features of characters by performing word embedding feature extraction on biomedical vocabulary.

[0046] Extracting the radical and radical head features of characters by performing radical and radical head feature extraction on biomedical characters.

[0047] Extracting the syntactic features of characters by performing syntactic feature extraction on biomedical sentences.

[0048] Extracting the stroke features of characters by performing stroke feature extraction on biomedical characters.

[0049] Further, the network structure of the trained biomedical named entity recognition model includes:

[0050] Four parallel branches;

[0051] Among them, the first branch is the pre-trained language model BioBERT. The input end of the pre-trained language model BioBERT is used to input biomedical vocabulary, and the output end of the pre-trained language model BioBERT is used to output the embedding features of characters. The output end of the pre-trained language model BioBERT is connected to the input end of the first splicing unit.

[0052] Among them, the second branch is a convolutional neural network CNN. The input end of the convolutional neural network CNN is used to input the mapping information of biomedical characters, and the output end of the convolutional neural network CNN is used to output the radical and component features of the characters. The output end of the convolutional neural network CNN is connected to the input end of the first splicing unit;

[0053] Among them, the third branch is a language model N-Gram. The input end of the language model N-Gram is used to input biomedical sentences, and the output end of the language model N-Gram is used to output syntactic features. The output end of the language model N-Gram is connected to the input end of the first self-attention mechanism layer, and the output end of the first self-attention mechanism layer is connected to the input end of the first splicing unit;

[0054] Among them, the output end of the first splicing unit is connected to the input end of the second self-attention mechanism layer;

[0055] Among them, the fourth branch is a long short-term memory model LSTM. The input end of the long short-term memory model LSTM is used to input biomedical characters, and the output end of the long short-term memory model LSTM is used to output stroke features. The output ends of the long short-term memory model LSTM and the second self-attention mechanism layer are both connected to the input end of the second splicing unit;

[0056] The output end of the second splicing unit is connected to the input end of the Bi-GRU; the output end of the Bi-GRU is connected to the input end of the conditional random field model CRF; the output end of the conditional random field model CRF is used to output the recognition labels of biomedical named entities.

[0057] Furthermore, for the trained biomedical named entity recognition model, its training process includes:

[0058] Construct a training set; the training set is a biomedical sentence with known biomedical named entity recognition labels;

[0059] Input the training set into the biomedical named entity recognition model, train the biomedical named entity recognition model, and stop training when the loss function reaches the minimum value or the number of iterations exceeds the set number of times to obtain the trained biomedical named entity recognition model.

[0060] Furthermore, for the extraction of word embedding features of biomedical vocabulary, the word embedding features of characters are extracted; specifically including:

[0061] Use the pre-trained language model BioBERT to extract word embedding features of biomedical vocabulary and extract the word embedding features of characters.

[0062] Exemplarily, the pre-trained language model BioBERT is trained using a large amount of existing medical literature knowledge. Each character of the input sequence is processed by the BioBERT model to obtain the corresponding output c i (i = 1, 2, 3....). Since BioBERT is a dynamic word embedding, it can dynamically adjust the word embedding of each character through the fine-tuning method. Therefore, it is better than the static word embedding methods such as word2vec and glove for representing characters.

[0063] Furthermore, for the bio-medical characters, the radical and stroke features are extracted; specifically, it includes:

[0064] Using the Chinese character feature extraction tool char_featurizer, the radical or stroke of each character in the input sequence is extracted to obtain the radical or stroke information;

[0065] Mapping the radical or stroke information to a radical or stroke vector;

[0066] Using the convolutional neural network CNN to extract features from the radical or stroke vector to extract the radical and stroke features of the character.

[0067] Exemplarily, for the bio-medical characters, the radical and stroke features are extracted; specifically, it includes:

[0068] Using the Chinese character feature extraction tool char_featurizer to extract the radicals and strokes of each character in the input sequence. For example, given a character "ammonia", by using the Chinese character feature extraction tool to disassemble the character by radicals, it can be disassembled into "qi" and "an". Map the obtained radical and stroke information to a feature vector, and then send it into the convolutional neural network (CNN) for feature extraction, so as to obtain the radical and stroke representation information r i (i = 1, 2, 3…). The specific calculation formula is as follows:

[0069]

[0070] Among them, b is the bias term, f(x) is the activation function, and the input sequence is represented as: X = [x1, x2, …, x s , where s represents the number of radicals and strokes, x t ∈R d is the d-dimensional feature vector of the t-th radical or stroke, x t:t+k-1 is the radical or stroke x t , x t+1 , …, x t+k-1 is the concatenation of, w ∈ R k×dis a convolution kernel.

[0071] Furthermore, for the biological and medical statement, extracting grammatical features specifically includes:

[0072] Using the language model N-Gram to perform word segmentation on the biological and medical statement;

[0073] For the result of word segmentation, using the first self-attention mechanism to extract grammatical features.

[0074] Exemplarily, for the biological and medical statement, extracting grammatical features specifically includes:

[0075] Using the n-gram grammar model (n = 3, 4, 5) to perform word segmentation on the biological and medical text sentence, and then processing the 3-gram, 4-gram, and 5-gram knowledge obtained for each character using the self-attention mechanism, as Figures 3(a)-3(c) shown, to obtain b i (i = 1, 2, 3...).

[0076] Furthermore, for the biological and medical character, extracting stroke features specifically includes:

[0077] Using the long short-term memory model LSTM to extract stroke features from the biological and medical statement.

[0078] As Figure 4 shown, for each character in the given sequence, obtaining the stroke information of each character, and then respectively processing the stroke information of each character through the long short-term memory network (LSTM) to obtain the corresponding output value h i (i = 1, 2, 3...), and taking the arithmetic mean of h i :

[0079] For each character in the given sequence, obtaining the stroke information of each character is to disassemble the strokes of Chinese characters step by step according to the stroke order.

[0080] The LSTM model mainly consists of three parts: an input gate, a forget gate, and an output gate. The relevant calculation formulas of the LSTM layer are as follows:

[0081] x t = σ(W s ·x t + U s ·h t -1 + b s ) (3)

[0082] m t = σ(Wm ·x t +U m ·h t -1 + b m ) (4)

[0083] n t =σ(W n ·x t +U n ·h t-1 +b n ) (5)

[0084]

[0085]

[0086]

[0087] In the formula, σ is the sigmoid activation function, s t is the input gate, m t is the forget gate, n t is the output gate, C t is the memory cell, h t represents the output value of the hidden layer at time t, W, U, and b are parameters during the neural network training process, is the dot product operation, x t represents the input value of the network at time t.

[0088] Furthermore, the word embedding features of the characters, the radical features of the characters, the radical features of the characters, and the grammar features are concatenated to obtain the first concatenated feature; among them, the concatenation is performed in a serial manner.

[0089] Exemplarily, r i , b i , c i are concatenated to obtain

[0090] Furthermore, the first concatenated feature is processed by the self-attention mechanism to obtain the processed first concatenated feature; specifically including:

[0091] The self-attention mechanism maps the input information to different spaces to obtain three matrices Q (Query), K (Key), and V (Value) respectively composed of the query vector query, the key vector key, and the value vector value.

[0092] First, calculate the dot product between Q and K, and then divide by to obtain the correlation weight matrix coefficient between Q and K;

[0093] Then, the obtained relevant weight matrix coefficients are normalized using the softmax function;

[0094] Finally, the normalized result is multiplied by the matrix V to obtain the vector sequence representation a of the current node of self-attention i (i = 1, 2, 3…).

[0095] The specific calculation formula is as follows:

[0096]

[0097] Among them, Q, K, and V are matrices respectively composed of vectors obtained by different linear transformations of the same input, and D k is the dimension between the query vector and the key vector, and softmax(·) is an activation function for column-wise normalization.

[0098] Exemplarily, is input into the self-attention mechanism (Self-Attention) for processing.

[0099] Furthermore, the stroke feature is concatenated with the processed first concatenated feature to obtain a second concatenated feature; specifically, the concatenation is performed in a series connection manner.

[0100] Exemplarily, s i is concatenated with a i to obtain

[0101] Furthermore, the second concatenated feature is encoded and decoded to obtain a biomedical named entity recognition label; specifically, it includes:

[0102] The second concatenated feature is input into a Bi-GRU for encoding processing;

[0103] The result of the encoding processing is input into a conditional random field model CRF to output a biomedical named entity recognition label.

[0104] Exemplarily, the second concatenated feature is encoded and decoded to obtain a biomedical named entity recognition label; specifically, it includes:

[0105] is fed into a Bi-GRU network for processing. The result information of the processing is then decoded using CRF, and corresponding labels are assigned to each character.

[0106] ​In a bidirectional GRU network, there is a hidden layer for forward propagation and a hidden layer for backward propagation. The input layer is connected to the forward propagation network and the backward propagation network respectively. The hidden layer states in both directions are passed to the output layer. Therefore, the output information contains both the forward information and the backward information of the input sequence. The specific calculation formula of GRU is as follows:

[0107] r t = σ(W r ·[h t-1 , x t ) (9)

[0108] z t = σ(W z ·[h t-1 , x t ) (10)

[0109]

[0110]

[0111] Among them, x t is the input data, h t is the output of the GRU cell, r t , z t are the reset gate and the update gate at time t respectively, σ is the Sigmoid function, W r , W Z , W h are the weight matrices of the reset gate, the update gate and the candidate hidden state respectively, is the candidate state at time t.

[0112] This method mainly includes five parts. As Figure 2 shown, the first part preprocesses biomedical vocabulary through a pre-trained language model (BioBERT) in the biomedical field. The second part is to obtain the radical and component information of each character and use a convolutional neural network (CNN) to obtain semantic representations. The third part is to perform n-gram tokenization on biomedical text sentences, and then concatenate the n-gram features with the word embeddings of the characters obtained by BioBERT and the radical and component features obtained by CNN. Then, it is sent to the self-attention mechanism for processing. The fourth part processes the stroke information of each character through a long short-term memory network (LSTM). The fifth part is to concatenate the stroke feature s i with the weight coefficient a i obtained by the self-attention mechanism, and then send it to Bi-GRU for processing. The output result is decoded by the CRF layer and labeled with the corresponding labels.

[0113] Example 2

[0114] This embodiment provides a biomedical named entity recognition system;

[0115] The biomedical named entity recognition system includes:

[0116] An acquisition module, which is configured to: acquire a biomedical statement to be recognized; divide the biomedical statement into biomedical words and biomedical characters;

[0117] A recognition module, which is configured to: input the biomedical statement, biomedical words, and biomedical characters into the trained biomedical named entity recognition model, and output the biomedical named entity recognition result;

[0118] Among them, the working principle of the trained biomedical named entity recognition model includes: respectively extracting the word embedding feature, radical feature, radical feature, and grammar feature of the characters from the biomedical statement, biomedical words, and biomedical characters; splicing the word embedding feature, radical feature, radical feature, and grammar feature of the characters to obtain a first splicing feature; then, performing a self-attention mechanism process on the first splicing feature to obtain a processed first splicing feature; splicing the stroke feature with the processed first splicing feature to obtain a second splicing feature; encoding and decoding the second splicing feature to obtain the biomedical named entity recognition label.

[0119] It should be noted here that the above acquisition module and recognition module correspond to steps S101 to S102 in Embodiment 1. The examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0120] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0121] The proposed system can be implemented in other ways. For example, the above-described system embodiments are merely illustrative. For example, the above module division is only a logical function division. In actual implementation, there can be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0122] Embodiment 3

[0123] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device runs, the processor executes the one or more computer programs stored in the memory, so that the electronic device executes the method described in Embodiment 1 above.

[0124] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0125] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0126] In the implementation process, each step of the above method may be completed by the integrated logic circuit in the hardware of the processor or the instructions in the form of software.

[0127] The method in Embodiment 1 may be directly embodied as being executed and completed by the hardware processor, or executed and completed by the combination of the hardware and software modules in the processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0128] Those of ordinary skill in the art can realize that, in combination with the units and algorithm steps of the examples described in this embodiment, they can be implemented by electronic hardware or the combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0129] Embodiment 4

[0130] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method described in Embodiment 1 is completed.

[0131] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for biomedical named entity recognition, characterized in that, Including: Obtain a biomedical statement to be recognized; divide the biomedical statement into biomedical vocabulary and biomedical characters; Input the biomedical statement, biomedical vocabulary, and biomedical characters into the trained biomedical named entity recognition model, and output the biomedical named entity recognition result; Among them, the working principle of the trained biomedical named entity recognition model includes: respectively extracting the word embedding feature, radical feature, radical feature, and syntactic feature of the character from the biomedical statement, biomedical vocabulary, and biomedical characters; splicing the word embedding feature, radical feature, radical feature, and syntactic feature of the character to obtain the first splicing feature; then, performing self-attention mechanism processing on the first splicing feature to obtain the processed first splicing feature; splicing the stroke feature with the processed first splicing feature to obtain the second splicing feature; encoding and decoding the second splicing feature to obtain the biomedical named entity recognition label.

2. The method for biomedical named entity recognition according to claim 1, characterized in that, Respectively extract the word embedding feature, radical feature, radical feature, and syntactic feature of the character from the biomedical statement, biomedical vocabulary, and biomedical characters; Specifically including: Extract the word embedding feature of the character by performing word embedding feature extraction on the biomedical vocabulary; Extract the radical and radical features of the character by performing radical and radical feature extraction on the biomedical characters; Extract the syntactic feature by performing syntactic feature extraction on the biomedical statement; Extract the stroke feature by performing stroke feature extraction on the biomedical characters.

3. The method for biomedical named entity recognition according to claim 1, characterized in that, The network structure of the trained biomedical named entity recognition model includes: Four parallel branches; Among them, the first branch is the pre-trained language model BioBERT. The input end of the pre-trained language model BioBERT is used to input biomedical vocabulary, and the output end of the pre-trained language model BioBERT is used to output the embedding feature of the character; the output end of the pre-trained language model BioBERT is connected to the input end of the first splicing unit; Among them, the second branch is the convolutional neural network CNN. The input end of the convolutional neural network CNN is used to input the mapping information of the biomedical characters, and the output end of the convolutional neural network CNN is used to output the radical and radical features of the character; the output end of the convolutional neural network CNN is connected to the input end of the first splicing unit; Among them, the third branch is the language model N-Gram. The input end of the language model N-Gram is used to input the biomedical statement, and the output end of the language model N-Gram is used to output the syntactic feature; the output end of the language model N-Gram is connected to the input end of the first self-attention mechanism layer, and the output end of the first self-attention mechanism layer is connected to the input end of the first splicing unit; Among them, the output end of the first splicing unit is connected to the input end of the second self-attention mechanism layer; Among them, the fourth branch is a long short-term memory model LSTM. The input end of the long short-term memory model LSTM is used to input biomedical characters, and the output end of the long short-term memory model LSTM is used to output stroke features; the output end of the long short-term memory model LSTM and the output end of the second self-attention mechanism layer are both connected to the input end of the second splicing unit; The output end of the second splicing unit is connected to the input end of the Bi-GRU; the output end of the Bi-GRU is connected to the input end of the conditional random field model CRF; the output end of the conditional random field model CRF is used to output the recognition labels of biomedical named entities.

4. The method for biomedical named entity recognition according to claim 1, characterized in that, For the trained biomedical named entity recognition model, its training process includes: Constructing a training set; the training set is a biomedical sentence with known biomedical named entity recognition labels; Inputting the training set into the biomedical named entity recognition model to train the biomedical named entity recognition model. When the loss function reaches the minimum value or the number of iterations exceeds the set number of times, stop training to obtain the trained biomedical named entity recognition model.

5. The method for biomedical named entity recognition according to claim 2, characterized in that, Performing word embedding feature extraction on biomedical vocabulary to extract the word embedding features of characters; Specifically including: Using the pre-trained language model BioBERT to perform word embedding feature extraction on biomedical vocabulary to extract the word embedding features of characters; Performing syntactic feature extraction on biomedical sentences to extract syntactic features; specifically including: Using the language model N-Gram to perform word segmentation on biomedical sentences; Performing syntactic feature extraction on the word segmentation result using the first self-attention mechanism to extract syntactic features; Performing stroke feature extraction on biomedical characters to extract stroke features; specifically including: Using the long short-term memory model LSTM to perform stroke feature extraction on biomedical sentences to extract stroke features; Performing splicing processing on the word embedding features of characters, the radical features of characters, the radical features of characters, and syntactic features to obtain a first splicing feature; among them, the splicing is performed in a concatenated manner.

6. The method for biomedical named entity recognition according to claim 2, characterized in that, Performing radical and radical feature extraction on biomedical characters to extract the radical and radical features of characters; specifically including: Using the Chinese character feature extraction tool char_featurizer to extract the radicals or radicals of each character in the input sequence to obtain the information of the radicals or radicals; Mapping the information of the radicals or radicals to vectors of the radicals or radicals; Performing feature extraction on the vectors of the radicals or radicals through a convolutional neural network CNN to extract the radical and radical features of characters.

7. The biological and medical named entity recognition method according to claim 1, wherein, Performing self-attention mechanism processing on the first splicing feature to obtain the processed first splicing feature; specifically including: The self-attention mechanism maps the input information to different spaces to obtain three matrices Q, K, and V composed of query vectors query, key vectors key, and value vectors value respectively; First, calculate the dot product between Q and K, and then divide by to obtain the correlation weight matrix coefficient of Q and K; Then using the softmax function to normalize the obtained relevant weight matrix coefficients; Finally, multiplying the normalized result by matrix V to obtain the vector sequence representation of the current node of self-attention; Encoding and decoding the second splicing feature to obtain a biomedical named entity recognition label; specifically including: Inputting the second splicing feature into a Bi-GRU for encoding processing; Inputting the result of the encoding processing into a conditional random field model CRF to output a biomedical named entity recognition label.

8. A biological and medical named entity recognition system, wherein, Including: An acquisition module configured to: acquire a biomedical statement to be recognized; divide the biomedical statement into biomedical words and biomedical characters; An identification module configured to: input the biomedical statement, biomedical words, and biomedical characters into a trained biomedical named entity recognition model, and output a biomedical named entity recognition result; Among them, the working principle of the trained biomedical named entity recognition model includes: respectively extracting the word embedding feature, radical feature, radical feature, and grammatical feature of the character from the biomedical statement, biomedical words, and biomedical characters; splicing the word embedding feature, radical feature, radical feature, and grammatical feature of the character to obtain a first splicing feature; then, performing self-attention mechanism processing on the first splicing feature to obtain a processed first splicing feature; splicing the stroke feature with the processed first splicing feature to obtain a second splicing feature; encoding and decoding the second splicing feature to obtain a biomedical named entity recognition label.

9. An electronic device, comprising: A memory for non-temporarily storing computer-readable instructions; And A processor for running the computer-readable instructions, Among them, when the computer-readable instructions are run by the processor, the method described in any one of the above claims 1-7 is executed.

10. A storage medium, wherein, Non-temporarily store computer-readable instructions, where when the computer-readable instructions are executed by a computer, the instructions for executing the method described in any one of claims 1-7.

Images