CN111949792B - Medicine relation extraction method based on deep learning - Google Patents

Medicine relation extraction method based on deep learning Download PDF

Info

Publication number
CN111949792B
CN111949792B CN202010811218.0A CN202010811218A CN111949792B CN 111949792 B CN111949792 B CN 111949792B CN 202010811218 A CN202010811218 A CN 202010811218A CN 111949792 B CN111949792 B CN 111949792B
Authority
CN
China
Prior art keywords
vector
sentence
medicine
representing
drug
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010811218.0A
Other languages
Chinese (zh)
Other versions
CN111949792A (en
Inventor
刘勇国
何家欢
杨尚明
李巧勤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010811218.0A priority Critical patent/CN111949792B/en
Publication of CN111949792A publication Critical patent/CN111949792A/en
Application granted granted Critical
Publication of CN111949792B publication Critical patent/CN111949792B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/50—Molecular design, e.g. of drugs

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medicinal Chemistry (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Computing Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention discloses a medicine relation extraction method based on deep learning, which is characterized in that a RDkit tool is utilized to convert a medicine molecular formula into a molecular diagram structure, the characteristics of medicine molecules are expressed, the text characteristics of a sample are extracted at the same time, the medicine molecule characteristics and the text characteristics of the sample are combined, then the medicine relations are classified by utilizing a full-connection layer softmax, the physicochemical properties of medicines in sentences are adopted, the extraction accuracy can be improved, and the problems that the existing method is difficult to cover all text scenes and excessively depends on an external natural language processing tool are solved.

Description

Medicine relation extraction method based on deep learning
Technical Field
The invention relates to the field of extraction of pharmaceutical chemical entity relations, in particular to a deep learning-based extraction method of a pharmaceutical relation.
Background
The extraction of the relationship of the pharmaceutical chemical entities refers to the automatic extraction of the relationship between the pharmaceutical entities from the text, which can assist pharmaceutical researchers in developing new drugs, assist doctors in making reasonable treatment schemes for patients and is also the basis for constructing a pharmaceutical chemical knowledge database. The existing extraction method of interaction relation of medicinal entities mainly comprises two types: rule-based methods and supervised machine learning-based methods.
Early research mostly employed rule-based approaches because early drug relationships extracted an authoritative annotated corpus that lacked. The sentence structure for expressing action relationship in the method is fixed and limited, namely, most sentences with action relationship description have the same or similar sentence structure. The method analyzes the syntax of the sentences, detects the syntax structure of the sentences, extracts the interacting drug pairs from the short sentences according to the description rules formulated by pharmacists, and classifies the drug pair relationship.
Since DDIExtraction2011 and DDIExtraction2013 evaluations, supervised-based machine learning methods were used for the extraction of the interaction relationships of the pharmaceutical entities, the most important of which is feature-based methods that treat the relationship extraction as a classification problem, explicitly represent candidate relationship instances as a feature vector with various types of features, and then classify the candidate relationship instances using a supervised machine learning model.
The rule-based method has a good extraction effect only for simple short sentences because it is difficult to formulate a proper rule for complex long sentences. However, the sentences of the documents in the pharmaceutical field are complex long sentences, many of the descriptive sentences contain more than two drugs, and the sentences contain a large number of isotopologues, parallel structures and other complex structures. The rule-based approach is less accurate with the current large amount of data. The formulation of the rules is time-consuming and labor-consuming and requires the participation of personnel in the professional field; furthermore, it is difficult for manually-programmed rules to cover all application text scenarios. The method based on supervised machine learning has better performance and portability, but the method depends on external natural language processing tools, and if the external tools make mistakes, errors can be propagated to influence the performance.
Disclosure of Invention
Aiming at the defects in the prior art, the medicine relation extraction method based on deep learning provided by the invention solves the problems that the existing method is difficult to cover all text scenes and excessively depends on an external natural language processing tool.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that:
the deep learning-based medicine relation extraction method comprises the following steps:
s1, acquiring documents related to the medicine, dividing the text content of the documents into sentences by taking the sentences as basic units, and taking each sentence as an initial sample;
s2, reserving an initial sample containing two or more drug nouns, and labeling the reserved sample to obtain a labeled sample;
s3, adding a position attribute relative to the medicine for each word according to the position relation between the word and the medicine in the labeled sample to obtain a position feature vector corresponding to each word;
s4, obtaining and converting the SMILES expressions of all the drug molecules into graph structures, and obtaining the drug molecule characteristic vector of each drug in the graph structures;
s5, representing the words in the text as vectors, replacing the corresponding words with the vectors, and vectorizing each sentence;
S6, inputting the vectorized sentence into a deep learning network to obtain a text feature vector corresponding to the sentence;
s7, connecting the text characteristic vector and the medicine molecule characteristic vector corresponding to each sentence in series to obtain an integral characteristic vector corresponding to each sentence;
s8, inputting the integral characteristic vector corresponding to each sentence into a full-connection layer to obtain a vector represented by nonlinearity;
and S9, classifying the vectors represented by the nonlinearity by adopting a softmax function to obtain the probability of each classification, and taking the class with the highest probability as the medicine-pair relation obtained by identification to finish the extraction of the medicine relation.
Further, in step S2, labeling the retained sample, and the specific method for obtaining the labeled sample includes:
according to DDIExtraction2013 challenge rules, tags are classified into 5 classes, which are respectively: advice, action, drug mechanism, positive and irrelevant.
Further, the specific method of step S3 is:
acquiring the position relation between each word and the medicine in the labeled sample, establishing a vector with the number of elements equal to the number of the medicines, and setting the value of the nth element in the vector as m if the word is in m positions before the nth medicine; and if the word is m positions behind the nth medicine, setting the value of the nth element in the vector as-m, traversing each medicine to obtain the position feature vector corresponding to the word, and further obtaining the position feature vector corresponding to each word.
Further, the specific method of step S4 includes the following sub-steps:
s4-1, obtaining SMILES expressions of all drug molecules from a database DrugBank;
s4-2, converting the drug molecule SMILES expression into a graph structure by using an RDkit tool and taking each atom of the drug as a node and taking an element bond between atoms as an edge;
s4-3, randomly initializing all element bonds and atoms in the graph structure into a vector, and according to a formula:
Figure BDA0002631028100000031
obtaining the vector representation of the v-th atom and element bond after the t-th iteration
Figure BDA0002631028100000032
Where σ (-) is the sigmod activation function; ht-1Is a parameter matrix;
Figure BDA0002631028100000033
is the vector representation of the v atom and element bond after the t-1 iteration;
Figure BDA0002631028100000034
representing the vector representation of the w atom and element bond after the t-1 iteration; n (v) represents a set of atoms and elemental bonds adjacent to the v atom in the diagram structure;
s4-4, according to the formula:
Figure BDA0002631028100000041
acquiring a drug molecular characteristic vector of a drug corresponding to the v-th atom, and further acquiring a drug molecular characteristic vector of each drug in a graph structure; wherein softmax (·) is a softmax function; wtIs a parameter matrix.
Further, the specific method of step S5 is:
and training text contents by adopting a word2vec model, representing words in the text as vectors, taking each vector as an element of a sentence vector according to the front-back position relation of the words to obtain a vector representing each sentence, and vectorizing each sentence.
Further, in step S6, the deep learning network is a two-way long-short term memory model, where the expression of the two-way long-short term memory model is:
ip＝σ(Wxixp+Whihp-1+bi)
fp＝σ(Wxfxp+Whfhp-1+bf)
cp＝fpcp-1+iptanh(Wxcxp+Whchp-1+bc)
op＝σ(Wxoxp+Whohp-1+bo)
hp＝optanh(cp)
wherein ipRepresents the output of the input gate; σ (-) is a sigmod activation function; wxiA parameter matrix representing between the input and the input gate; x is the number ofpIs the input of the model; whiIs a parameter matrix between the hidden layer and the input gate; h isp-1A hidden layer output representing a last input word in the sentence; biAn offset vector representing an input gate; f. ofpAn output representing a forgetting gate; wxfA parameter matrix representing between the input and the forgetting gate; whfA parameter matrix representing between the hidden layer and the forgetting gate; bfAn offset vector representing a forgetting gate; c. CpPresentation memory sheetThe output of the element; c. Cp-1The memory unit output corresponding to the last word is shown; tanh (-) is a tanh activation function; wxcA parameter matrix representing the input and memory cells; whcA parameter matrix representing between the hidden layer and the memory cell; bcAn offset vector representing a memory cell; opRepresents the output of the output gate; wxoRepresenting a parameter matrix between the output gate and the input; whoRepresenting a parameter matrix between the output gate and the hidden layer; boAn offset vector representing an output gate; h ispRepresenting the output of the two-way long-short term memory model.
Further, the specific method of step S7 is:
and taking the text characteristic vector corresponding to each sentence as a first element, and sequentially arranging the drug molecule characteristic vectors corresponding to the drugs in the sentence after the text characteristic vectors of the sentence according to the sequence of the drugs in the sentence to obtain the overall characteristic vector corresponding to each sentence.
Further, the specific method of step S8 is:
inputting the global feature vector corresponding to each sentence into the full-connected layer, according to the formula:
X'＝tanh(W'X+b')
obtaining a vector X' represented by nonlinearity; wherein tanh (-) is a tanh activation function; w' is a full link layer parameter; b' is the offset of the full connection layer; x is input.
Further, the specific method of step S9 is:
according to the formula:
Figure BDA0002631028100000051
classifying the vector X' of the non-linear representation to obtain an output comprising a probability for each classification
Figure BDA0002631028100000052
The class with the highest probability is used as the medicine-pair relation obtained by identification, and medicine relation extraction is completed; wherein softmax (. cndot.) is softman ax function; w' is a classification parameter matrix; b "is the classification parameter offset.
The invention has the beneficial effects that:
1. the method utilizes an RDkit tool to convert the molecular formula of the medicine into a molecular diagram structure, then expresses the characteristics of the medicine molecules, extracts the text characteristics of the sample, combines the characteristics of the medicine molecules with the text characteristics of the sample, and classifies the medicine relations by utilizing a full-link softmax.
2. The data of the invention comes from the medicine literature, so that the lag of updating the medicine relation data can be effectively reduced, the acquisition speed of the medicine relation information is accelerated, the learning cost and the learning burden of medical workers are reduced, the cognitive level of the medical workers on medicine knowledge is improved, and the potential risk of various adverse drug reactions to patients in the medicine taking process is reduced.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
As shown in fig. 1, the deep learning-based drug relationship extraction method includes the following steps:
s1, acquiring documents related to the medicine, dividing the text content of the documents into sentences by taking the sentences as basic units, and taking each sentence as an initial sample;
S2, reserving an initial sample containing two or more drug nouns, and labeling the reserved sample to obtain a labeled sample;
s3, adding a position attribute relative to the medicine for each word according to the position relation between the word and the medicine in the labeled sample to obtain a position feature vector corresponding to each word;
s4, obtaining and converting the SMILES expressions of all the drug molecules into graph structures, and obtaining the drug molecule characteristic vector of each drug in the graph structures;
s5, representing the words in the text as vectors, replacing the corresponding words with the vectors, and vectorizing each sentence;
s6, inputting the vectorized sentence into a deep learning network to obtain a text feature vector corresponding to the sentence;
s7, connecting the text characteristic vector and the medicine molecule characteristic vector corresponding to each sentence in series to obtain an integral characteristic vector corresponding to each sentence;
s8, inputting the integral characteristic vector corresponding to each sentence into a full-connection layer to obtain a vector represented by nonlinearity;
and S9, classifying the vectors represented by the nonlinearity by adopting a softmax function to obtain the probability of each classification, and taking the class with the highest probability as the medicine-pair relation obtained by identification to finish the extraction of the medicine relation.
In step S2, labeling the retained sample, and the specific method for obtaining the labeled sample is as follows: according to DDIExtraction2013 challenge rules, tags are classified into 5 classes, which are respectively: advice, action, drug mechanism, positive and irrelevant.
The specific method of step S3 is: acquiring the position relation between each word and the medicine in the labeled sample, establishing a vector with the number of elements equal to the number of the medicines, and setting the value of the nth element in the vector as m if the word is in m positions before the nth medicine; and if the word is m positions behind the nth medicine, setting the value of the nth element in the vector as-m, traversing each medicine to obtain the position feature vector corresponding to the word, and further obtaining the position feature vector corresponding to each word.
The specific method of step S4 includes the following substeps:
s4-1, acquiring SMILES expressions of all drug molecules from a database DrugBank;
s4-2, converting the drug molecule SMILES expression into a graph structure by using each atom of the drug as a node and an element bond between atoms as an edge through an RDkit tool;
s4-3, randomly initializing all element bonds and atoms in the graph structure into a vector according to a formula:
Figure BDA0002631028100000071
Obtaining the vector representation of the v-th atom and element bond after the t-th iteration
Figure BDA0002631028100000072
Where σ (-) is the sigmod activation function; ht-1Is a parameter matrix;
Figure BDA0002631028100000073
is the vector representation of the v atom and element bond after the t-1 iteration;
Figure BDA0002631028100000074
representing the vector representation of the w atom and element bond after the t-1 iteration; n (v) represents a set of atoms and elemental bonds adjacent to the v atom in the diagram structure;
s4-4, according to the formula:
Figure BDA0002631028100000081
acquiring a drug molecular characteristic vector of a drug corresponding to the v-th atom, and further acquiring a drug molecular characteristic vector of each drug in a graph structure; wherein softmax (·) is a softmax function; wtIs a parameter matrix.
The specific method of step S5 is: and training text contents by adopting a word2vec model, representing words in the text as vectors, taking each vector as an element of a sentence vector according to the front-back position relation of the words to obtain a vector representing each sentence, and vectorizing each sentence.
In step S6, the deep learning network is a bidirectional long-short term memory model, where the expression of the bidirectional long-short term memory model is:
ip＝σ(Wxixp+Whihp-1+bi)
fp＝σ(Wxfxp+Whfhp-1+bf)
cp＝fpcp-1+iptanh(Wxcxp+Whchp-1+bc)
op＝σ(Wxoxp+Whohp-1+bo)
hp＝optanh(cp)
wherein ipRepresents the output of the input gate; σ (-) is a sigmod activation function; wxiA parameter matrix representing between the input and the input gate; x is the number of pIs the input of the model; whiIs a parameter matrix between the hidden layer and the input gate; h isp-1A hidden layer output representing a last input word in the sentence; biAn offset vector representing an input gate; f. ofpAn output representing a forgetting gate; wxfA parameter matrix representing between the input and the forgetting gate; whfA parameter matrix representing between the hidden layer and the forgetting gate; bfAn offset vector representing a forgetting gate; c. CpRepresenting the output of the memory cell; c. Cp-1The memory unit output corresponding to the last word is shown; tanh (-) is a tanh activation function; wxcA parameter matrix representing the input and memory cells; whcA parameter matrix representing between the hidden layer and the memory cell; bcAn offset vector representing a memory cell; opRepresents the output of the output gate; wxoRepresenting a parameter matrix between the output gate and the input; whoRepresenting a parameter matrix between the output gate and the hidden layer; boAn offset vector representing an output gate; h ispRepresenting outputs of a two-way long-short term memory modelAnd (6) discharging.
The specific method of step S7 is: and taking the text characteristic vector corresponding to each sentence as a first element, and sequentially arranging the drug molecule characteristic vectors corresponding to the drugs in the sentence after the text characteristic vectors of the sentence according to the sequence of the drugs in the sentence to obtain the overall characteristic vector corresponding to each sentence.
The specific method of step S8 is: inputting the global feature vector corresponding to each sentence into the full-connected layer, according to the formula:
X'＝tanh(W'X+b')
obtaining a vector X' represented by nonlinearity; wherein tanh (-) is a tanh activation function; w' is a full link layer parameter; b' is the offset of the full connection layer; x is input.
The specific method of step S9 is: according to the formula:
Figure BDA0002631028100000091
classifying the vector X' of the non-linear representation to obtain an output comprising a probability for each classification
Figure BDA0002631028100000092
The class with the highest probability is used as the medicine-pair relation obtained by identification, and medicine relation extraction is completed; wherein softmax (·) is a softmax function; w' is a classification parameter matrix; b "is the classification parameter offset.
In one embodiment of the present invention, the drug literature is available from PubMed. The bidirectional long-short term memory model respectively calculates hidden vectors from front to back of text sentences
Figure BDA0002631028100000093
And computing the hidden vector from back to front
Figure BDA0002631028100000094
The last output in two directions is respectively
Figure BDA0002631028100000095
And
Figure BDA0002631028100000096
are connected in series to obtain the text characteristic vector H of the sentenceS。
In summary, the invention utilizes the RDKit tool to convert the molecular formula of the drug into the molecular diagram structure, then expresses the characteristics of the drug molecules, extracts the text characteristics of the sample, combines the characteristics of the drug molecules with the text characteristics of the sample, and then classifies the drug relationships by utilizing the full-link softmax, and adopts the physicochemical properties of the drug in the sentence, thereby improving the extraction accuracy and solving the problems that the existing method is difficult to cover all text scenes and excessively depends on the external natural language processing tool.

Claims (9)

1. A medicine relation extraction method based on deep learning is characterized by comprising the following steps:
s1, obtaining documents related to the medicine, dividing the text content of the documents into sentences by taking the sentences as basic units, and taking each sentence as an initial sample;
s2, reserving an initial sample containing two or more drug nouns, and labeling the reserved sample to obtain a labeled sample;
s3, adding a position attribute relative to the medicine for each word according to the position relation between the word and the medicine in the labeled sample to obtain a position feature vector corresponding to each word;
s4, obtaining and converting the SMILES expressions of all the drug molecules into graph structures, and obtaining the drug molecule feature vector of each drug in the graph structures;
s5, representing the words in the text as vectors, and substituting the vectors for the corresponding words to vectorize each sentence;
s6, inputting the vectorized sentence into a deep learning network to obtain a text feature vector corresponding to the sentence;
s7, connecting the text characteristic vector and the medicine molecule characteristic vector corresponding to each sentence in series to obtain an integral characteristic vector corresponding to each sentence;
s8, inputting the integral characteristic vector corresponding to each sentence into a full-connection layer to obtain a vector represented by nonlinearity;
And S9, classifying the vectors represented by the nonlinearity by adopting a softmax function to obtain the probability of each classification, and taking the class with the highest probability as the medicine-pair relation obtained by identification to finish the extraction of the medicine relation.
2. The method for extracting drug relationship based on deep learning of claim 1, wherein the step S2 is to label the retained samples, and the specific method for obtaining the labeled samples is as follows:
according to the DDIExtraction2013 challenge rule, the tags are divided into 5 classes, which are respectively: advice, action, drug mechanism, positive and irrelevant.
3. The method for extracting drug relationship based on deep learning of claim 1, wherein the specific method of step S3 is as follows:
acquiring the position relation between each word and the medicine in the labeled sample, establishing a vector with the number of elements equal to the number of the medicines, and setting the value of the nth element in the vector as m if the word is in m positions before the nth medicine; and if the word is m positions behind the nth medicine, setting the value of the nth element in the vector as-m, traversing each medicine to obtain the position feature vector corresponding to the word, and further obtaining the position feature vector corresponding to each word.
4. The deep learning-based drug relationship extraction method as claimed in claim 1, wherein the specific method of step S4 includes the following sub-steps:
s4-1, acquiring SMILES expressions of all drug molecules from a database DrugBank;
s4-2, converting the drug molecule SMILES expression into a graph structure by using each atom of the drug as a node and an element bond between atoms as an edge through an RDkit tool;
s4-3, randomly initializing all element bonds and atoms in the graph structure into a vector, and according to a formula:
Figure FDA0002631028090000021
obtaining the vector representation of the v-th atom and element bond after the t-th iteration
Figure FDA0002631028090000022
Where σ (-) is the sigmod activation function; ht-1Is a parameter matrix;
Figure FDA0002631028090000023
is the vector representation of the v atom and element bond after the t-1 iteration;
Figure FDA0002631028090000024
representing the vector representation of the w atom and element bond after the t-1 iteration; n (v) represents a set of atoms and elemental bonds adjacent to the v atom in the diagram structure;
s4-4, according to the formula:
Figure FDA0002631028090000025
acquiring a drug molecular characteristic vector of a drug corresponding to the v-th atom, and further acquiring a drug molecular characteristic vector of each drug in a graph structure; wherein softmax (·) is a softmax function; wtIs a parameter matrix.
5. The deep learning-based drug relationship extraction method according to claim 1, wherein the specific method of step S5 is:
and training text contents by adopting a word2vec model, representing words in the text as vectors, taking each vector as an element of a sentence vector according to the front-back position relation of the words to obtain a vector representing each sentence, and vectorizing each sentence.
6. The deep learning based drug relationship extraction method as claimed in claim 1, wherein the deep learning network in step S6 is a two-way long-short term memory model, wherein the expression of the two-way long-short term memory model is:
ip＝σ(Wxixp+Whihp-1+bi)
fp＝σ(Wxfxp+Whfhp-1+bf)
cp＝fpcp-1+iptanh(Wxcxp+Whchp-1+bc)
op＝σ(Wxoxp+Whohp-1+bo)
hp＝optanh(cp)
wherein ipRepresents the output of the input gate; σ (-) is a sigmod activation function; wxiA parameter matrix representing between the input and the input gate; x is the number ofpIs the input of the model; whiIs a parameter matrix between the hidden layer and the input gate; h isp-1A hidden layer output representing a last input word in the sentence; biAn offset vector representing an input gate; f. ofpAn output representing a forgetting gate; wxfA parameter matrix representing between the input and the forgetting gate; whfA parameter matrix representing between the hidden layer and the forgetting gate; bfAn offset vector representing a forgetting gate; c. C pRepresenting the output of the memory cell; c. Cp-1The memory unit output corresponding to the last word is shown; tanh (-) is a tanh activation function; wxcA parameter matrix representing the input and memory cells; whcA parameter matrix representing between the hidden layer and the memory cell; bcAn offset vector representing a memory cell; opRepresents the output of the output gate; wxoRepresenting a parameter matrix between the output gate and the input; whoRepresenting a parameter matrix between the output gate and the hidden layer; boAn offset vector representing an output gate; h ispRepresenting the output of the two-way long-short term memory model.
7. The method for extracting drug relationship based on deep learning of claim 1, wherein the specific method of step S7 is as follows:
and taking the text characteristic vector corresponding to each sentence as a first element, and sequentially arranging the drug molecule characteristic vectors corresponding to the drugs in the sentence after the text characteristic vectors of the sentence according to the sequence of the drugs in the sentence to obtain the overall characteristic vector corresponding to each sentence.
8. The method for extracting drug relationship based on deep learning of claim 1, wherein the specific method of step S8 is as follows:
inputting the integral characteristic vector corresponding to each sentence into a full connection layer, and according to a formula:
X'＝tanh(W'X+b')
Obtaining a vector X' represented by nonlinearity; wherein tanh (-) is a tanh activation function; w' is a full link layer parameter; b' is the offset of the full connection layer; x is input.
9. The method for extracting drug relationship based on deep learning of claim 1, wherein the specific method of step S9 is as follows:
according to the formula:
Figure FDA0002631028090000041
classifying the vector X' of the non-linear representation to obtain an output comprising a probability for each classification
Figure FDA0002631028090000042
The class with the highest probability is used as the medicine-pair relation obtained by identification, and medicine relation extraction is completed; wherein softmax (·) is a softmax function; w' is a classification parameter matrix;b "is the classification parameter offset.
CN202010811218.0A 2020-08-13 2020-08-13 Medicine relation extraction method based on deep learning Active CN111949792B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010811218.0A CN111949792B (en) 2020-08-13 2020-08-13 Medicine relation extraction method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010811218.0A CN111949792B (en) 2020-08-13 2020-08-13 Medicine relation extraction method based on deep learning

Publications (2)

Publication Number Publication Date
CN111949792A CN111949792A (en) 2020-11-17
CN111949792B true CN111949792B (en) 2022-05-31

Family

ID=73331750

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010811218.0A Active CN111949792B (en) 2020-08-13 2020-08-13 Medicine relation extraction method based on deep learning

Country Status (1)

Country Link
CN (1) CN111949792B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112820359A (en) * 2021-02-24 2021-05-18 北京中医药大学东直门医院 Liver injury prediction method, apparatus, device, medium, and program product

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106933985A (en) * 2017-02-20 2017-07-07 广东省中医院 A kind of analysis of core side finds method
CN107092594A (en) * 2017-04-19 2017-08-25 厦门大学 Bilingual recurrence self-encoding encoder based on figure
CN108520275A (en) * 2017-06-28 2018-09-11 浙江大学 A kind of regular system of link information based on adjacency matrix, figure Feature Extraction System, figure categorizing system and method
CN109446338A (en) * 2018-09-20 2019-03-08 大连交通大学 Drug disease relationship classification method neural network based
CN109783618A (en) * 2018-12-11 2019-05-21 北京大学 Pharmaceutical entities Relation extraction method and system based on attention mechanism neural network
CN109801687A (en) * 2019-01-15 2019-05-24 合肥工业大学 A kind of construction method and system of the causality knowledge base towards medicine
CN109934852A (en) * 2019-04-01 2019-06-25 重庆理工大学 A kind of video presentation method based on object properties relational graph
CN110223742A (en) * 2019-06-14 2019-09-10 中南大学 The clinical manifestation information extraction method and equipment of Chinese electronic health record data
CN111078889A (en) * 2019-12-20 2020-04-28 大连理工大学 Method for extracting relationships among medicines based on attention of various entities and improved pre-training language model
CN111125434A (en) * 2019-11-26 2020-05-08 北京理工大学 Relation extraction method and system based on ensemble learning
CN111339774A (en) * 2020-02-07 2020-06-26 腾讯科技（深圳）有限公司 Text entity relation extraction method and model training method
CN111429977A (en) * 2019-09-05 2020-07-17 中国海洋大学 Novel molecular similarity search algorithm based on graph structure attention

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11490849B2 (en) * 2017-01-03 2022-11-08 AventuSoft, LLC System and method of marking cardiac time intervals from the heart valve signals

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106933985A (en) * 2017-02-20 2017-07-07 广东省中医院 A kind of analysis of core side finds method
CN107092594A (en) * 2017-04-19 2017-08-25 厦门大学 Bilingual recurrence self-encoding encoder based on figure
CN108520275A (en) * 2017-06-28 2018-09-11 浙江大学 A kind of regular system of link information based on adjacency matrix, figure Feature Extraction System, figure categorizing system and method
CN109446338A (en) * 2018-09-20 2019-03-08 大连交通大学 Drug disease relationship classification method neural network based
CN109783618A (en) * 2018-12-11 2019-05-21 北京大学 Pharmaceutical entities Relation extraction method and system based on attention mechanism neural network
CN109801687A (en) * 2019-01-15 2019-05-24 合肥工业大学 A kind of construction method and system of the causality knowledge base towards medicine
CN109934852A (en) * 2019-04-01 2019-06-25 重庆理工大学 A kind of video presentation method based on object properties relational graph
CN110223742A (en) * 2019-06-14 2019-09-10 中南大学 The clinical manifestation information extraction method and equipment of Chinese electronic health record data
CN111429977A (en) * 2019-09-05 2020-07-17 中国海洋大学 Novel molecular similarity search algorithm based on graph structure attention
CN111125434A (en) * 2019-11-26 2020-05-08 北京理工大学 Relation extraction method and system based on ensemble learning
CN111078889A (en) * 2019-12-20 2020-04-28 大连理工大学 Method for extracting relationships among medicines based on attention of various entities and improved pre-training language model
CN111339774A (en) * 2020-02-07 2020-06-26 腾讯科技（深圳）有限公司 Text entity relation extraction method and model training method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
深度学习实体关系抽取研究综述;鄂海红等;《软件学报》;20190328;第30卷(第6期);1793-1818 *
藏药药理命名实体识别;何家欢等;《医学信息学杂质》;20200425;第41卷(第4期);30-36 *
面向医学文本的实体关系抽取研究综述;昝红英等;《郑州大学学报（理学版）》;20200804;第52卷(第4期);1-15 *

Also Published As

Publication number Publication date
CN111949792A (en) 2020-11-17

Similar Documents

Publication Publication Date Title
CN109783618B (en) Attention mechanism neural network-based drug entity relationship extraction method and system
Beam et al. Clinical concept embeddings learned from massive sources of multimodal medical data
Wang et al. Dependency-based long short term memory network for drug-drug interaction extraction
US7672987B2 (en) System and method for integration of medical information
Lee et al. Machine learning in relation to emergency medicine clinical and operational scenarios: an overview
Javed et al. An efficient pattern recognition based method for drug-drug interaction diagnosis
CN108630322A (en) Drug interaction modeling and methods of risk assessment, terminal device and storage medium
US20220068482A1 (en) Interactive treatment pathway interface for guiding diagnosis or treatment of a medical condition
CN116860987A (en) Domain knowledge graph construction method and system based on generation type large language model
CN115293161A (en) Reasonable medicine taking system and method based on natural language processing and medicine knowledge graph
CN110609910A (en) Medical knowledge graph construction method and device, storage medium and electronic equipment
CN111949792B (en) Medicine relation extraction method based on deep learning
Hu et al. A novel neural network model fusion approach for improving medical named entity recognition in online health expert question-answering services
Goenaga et al. A section identification tool: towards hl7 cda/ccr standardization in spanish discharge summaries
CN116383413B (en) Knowledge graph updating method and system based on medical data extraction
CN114882970B (en) Medicine interaction effect prediction method based on pre-training model and molecular diagram
CN109871414A (en) Biomedical entity relationship classification method based on the context vector kernel of graph
Chen et al. Extraction of entity relations from Chinese medical literature based on multi-scale CRNN
Zubke et al. Using openEHR archetypes for automated extraction of numerical information from clinical narratives
Bhatia et al. An efficient modular framework for automatic LIONC classification of MedIMG using unified medical language
Xing et al. Biorel: a large-scale dataset for biomedical relation extraction
Butala et al. Natural language parser for physician’s handwritten prescription
Miranda et al. Deep Learning for Multi-Label ICD-9 Classification of Hospital Discharge Summaries
Nguyen et al. Medical Prescription Recognition Using Heuristic Clustering and Similarity Search
Tsai et al. Application of Named Entity Recognition by Self-Attention BiLSTM-CRF to Chinese Prescription Document

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant