CN110321550A - A kind of name entity recognition method and device towards Chinese medical book document - Google Patents
A kind of name entity recognition method and device towards Chinese medical book document Download PDFInfo
- Publication number
- CN110321550A CN110321550A CN201910340359.6A CN201910340359A CN110321550A CN 110321550 A CN110321550 A CN 110321550A CN 201910340359 A CN201910340359 A CN 201910340359A CN 110321550 A CN110321550 A CN 110321550A
- Authority
- CN
- China
- Prior art keywords
- entity
- word
- corpus
- training
- chinese medicine
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The embodiment of the present invention discloses a kind of name entity recognition method and device towards Chinese medical book document, which comprises the entity word for arranging at least one entity type obtains a first traditional Chinese medical science field vocabulary comprising entity type to be identified;Using the automatic short phrase picking technology of AutoPhrase, short phrase picking is carried out from Chinese medicine ancient Chinese prose corpus, obtains the second traditional Chinese medical science field vocabulary;According to scheduled time mark strategy, the entity occurred in the Chinese medicine ancient Chinese prose corpus is marked out;Obtain the labeled data of Chinese medicine ancient Chinese prose corpus;Training dataset, validation data set, test data set are generated, training dataset is output in trained file, validation data set and test data set are output in test file;Data are read in from the trained file, test file, according to the automatic Named Entity Extraction Model of reading data training, the Chinese medicine ancient Chinese prose corpus are predicted, the result identified;The entity identified according to result.
Description
Technical field
The present invention relates to Chinese language processing field more particularly to a kind of name entity recognition methods towards Chinese medical book document
And device.
Background technique
With the development of technology, need to be named Chinese medicine literature of ancient book Entity recognition processing.Current method all needs
Will largely artificial labeled data or design feature, however the mark of traditional Chinese medical science field and characteristic Design need domain knowledge, institute
It is higher with cost.
Summary of the invention
In view of this, the embodiment of the present invention provides a kind of name entity recognition method and dress towards Chinese medical book document
It sets, can be improved the automatization level of the name Entity recognition of Chinese medical book document.
A kind of name entity recognition method towards Chinese medical book document, comprising:
S1, the entity word for arranging at least one entity type, obtaining one includes the first of entity type to be identified
Traditional Chinese medical science field vocabulary;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;
S2, using the automatic short phrase picking technology of AutoPhrase, carry out short phrase picking from Chinese medicine ancient Chinese prose corpus, obtain institute
Possible entity word, obtains the second traditional Chinese medical science field vocabulary, and the second traditional Chinese medical science field vocabulary includes entity word;
S3, in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary, according to scheduled time mark strategy,
Mark out the entity occurred in the Chinese medicine ancient Chinese prose corpus;
S4, in conjunction with the Chinese medicine ancient Chinese prose corpus return mark result and tie/break connects/disconnects dimension model, obtain
Cure the labeled data of ancient Chinese prose corpus;
S5, in conjunction with the labeled data and according to the obtained pre-training model Word of Chinese medicine ancient Chinese prose corpus training
The insertion of Embedding word, generates training dataset, validation data set, test data set, and training dataset is output to training text
In part, validation data set and test data set are output in test file;
S6, data are read in from the trained file, test file, it is automatic according to reading data training AutoNER
Named Entity Extraction Model, and the automatic Named Entity Extraction Model of the AutoNER obtained using training, it is ancient to the Chinese medicine
Literary corpus predicted, the result identified;The entity identified according to result.
A kind of name entity recognition device towards Chinese medical book document, comprising:
Finishing unit arranges the entity word of at least one entity type, obtains one and includes entity type to be identified
The first traditional Chinese medical science field vocabulary;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;
It excavates unit and carries out short phrase picking from Chinese medicine ancient Chinese prose corpus using the automatic short phrase picking technology of AutoPhrase,
All possible entity word is obtained, the second traditional Chinese medical science field vocabulary is obtained, the second traditional Chinese medical science field vocabulary includes entity word;
Unit is marked, in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary, according to scheduled time
Mark strategy, marks out the entity occurred in the Chinese medicine ancient Chinese prose corpus;
Processing unit connects/disconnects dimension model in conjunction with time mark result and tie/break of the Chinese medicine ancient Chinese prose corpus,
Obtain the labeled data of Chinese medicine ancient Chinese prose corpus;
Output unit, in conjunction with the labeled data and the pre-training model obtained according to Chinese medicine ancient Chinese prose corpus training
The insertion of Word Embedding word, generates training dataset, validation data set, test data set, training dataset is output to
In training file, validation data set and test data set are output in test file;
Predicting unit reads in data from the trained file, test file, according to reading data training
The automatic Named Entity Extraction Model of AutoNER, and the automatic Named Entity Extraction Model of the AutoNER obtained using training,
The Chinese medicine ancient Chinese prose corpus is predicted, the result identified;The entity identified according to result.
In the present invention, it can be named Entity recognition towards Chinese medical book document, improve automatic processing level.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the flow diagram of name entity recognition method of the embodiment of the present invention towards Chinese medical book document;
Fig. 2 is the schematic diagram of the name entity recognition method towards Chinese medical book document in application scenarios of the present invention;
Fig. 3 is that the embodiment of the present invention executes the result schematic diagram after step 1;
Fig. 4 is that the embodiment of the present invention executes the result schematic diagram after step 2;
Fig. 5 is that the embodiment of the present invention executes the result schematic diagram after step 4;
Fig. 6 be the present invention in pre-training model word2vec word to vector result;
Fig. 7 is that the embodiment of the present invention executes the result schematic diagram after step 6;
Fig. 8 is the schematic diagram of model training evaluation index in the present invention.
Fig. 9 is the connection schematic diagram of name entity recognition device of the embodiment of the present invention towards Chinese medical book document.
Specific embodiment
The embodiment of the present invention is described in detail with reference to the accompanying drawing.
It will be appreciated that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
Its embodiment, shall fall within the protection scope of the present invention.
For convenience of description, description apparatus above is to be divided into various units/modules with function to describe respectively.Certainly, exist
Implement to realize each unit/module function in the same or multiple software and or hardware when the present invention.
As shown in Figure 1, a kind of name entity recognition method towards Chinese medical book document, comprising:
S1, the entity word for arranging at least one entity type, obtaining one includes the first of entity type to be identified
Traditional Chinese medical science field vocabulary;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;As shown in figure 3, being S1
The result of step: what this step generated is the vocabulary of belt type,
Format a: word and corresponding type (it is pulse condition, first letter of pinyin that ZZ, which is symptom, MX)
S2, using the automatic short phrase picking technology of AutoPhrase, carry out short phrase picking from Chinese medicine ancient Chinese prose corpus, obtain institute
Possible entity word, obtains the second traditional Chinese medical science field vocabulary, and the second traditional Chinese medical science field vocabulary includes entity word;Such as Fig. 4
Shown, be the result of AutoPhrase short phrase picking: what this step generated is the vocabulary of not belt type, format: according to phrase
Quality sorts from high to low.
S3, in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary, according to scheduled time mark strategy,
Mark out the entity occurred in the Chinese medicine ancient Chinese prose corpus;
S4, in conjunction with the Chinese medicine ancient Chinese prose corpus return mark result and tie/break connects/disconnects dimension model, obtain
Cure the labeled data of ancient Chinese prose corpus;As shown in figure 5, format: showing complete a word.It is possible that " I None S ", which represents the word not,
Entity;The type that " O None D " represents the possible entity is unknown;" I+ corresponding types+S " represents the beginning of single entity,
It is break;Centre or the end that " O+ corresponding types+S " represents single entity, are tie;
S5, in conjunction with the labeled data and according to the obtained pre-training model Word of Chinese medicine ancient Chinese prose corpus training
The insertion of Embedding word, generates training dataset, validation data set, test data set, and training dataset is output to training text
In part, verify data and test data set are output in test file;As shown in fig. 6, pre-training model word2vec word to
Measure result: should be the result is that converting a vector for a word or a word;Such as Fig. 6 ' ' word be converted into one 200 dimension
Vector, 48355 representatives have 48355 words;Training, verifying, test set are exactly three files, and the content of the inside is by the knot in S4
Fruit is converted into pure digi-tal, the i.e. format of computer capacity understanding in conjunction with the result of word2vec.
S6, data are read in from the trained file, test file, it is automatic according to reading data training AutoNER
Named Entity Extraction Model, and the automatic Named Entity Extraction Model of the AutoNER obtained using training, it is ancient to the Chinese medicine
Literary corpus predicted, the result identified;The entity identified according to result.As shown in fig. 7, being obtained for model prediction
Result format: the first and second column are line number of the entity in original language material, and three column are entities, and four, five column are entity types
Id and entity type.
Wherein, the step S1 includes:
S101, the entity word for arranging at least one entity type, the arrangement includes: to clear up entity, is deleted
Space, punctuation mark;
S102, the entity for having ambiguousness in existing vocabulary is deleted, and carries out the operation of duplicate removal to the entity of same type, obtained
First traditional Chinese medical science field vocabulary of the entity composition of type to be identified.
Wherein, the step S2 includes:
S201, setting the automatic short phrase picking script of AutoPhrase in input and output path, short phrase picking word frequency with
And the Thread Count of program operation;
S202, according to the Chinese medicine ancient Chinese prose corpus, the deactivated vocabulary of Chinese is safeguarded, word that addition needs to filter and
Word;
S203, the TCM Major term in Chinese medicine ancient Chinese prose corpus is added in vocabulary to promote the quality of short phrase picking, so
After carry out short phrase picking, generate the second traditional Chinese medical science field vocabulary, the second traditional Chinese medical science field vocabulary be according to phrase mass fraction from
The vocabulary of high to Low sequence.
Wherein, the step S3 includes:
301, merge the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary;First Chinese medicine is led
Domain vocabulary reads in entity and corresponding type;
S302, the second traditional Chinese medical science field vocabulary is cleared up, deletes front and back space, newline, and filter out list
The entity of word;Two credible threshold values are set, and the entity formed respectively for word and more words is screened, and is more than the credible threshold
Value, then be merged into final vocabulary, and entity word is saved when merging, and corresponding entity type is set as NULL;
S303, according to mark strategy is returned, for each sentence in Chinese medicine ancient Chinese prose, return to the entity being wherein likely to occur and right
The entity type answered;
Described time mark strategy includes:
Short word situation is contained at least two for a long word, using the strategy of priority of long word, and it is corresponding to mark word
Entity type;Situation is formed by least two short words for long word, the entity type of the short word of each of the long word is marked
For NULL;The case where at least two word intersections conflict, the entity type of described two words is all labeled as NULL.
Wherein, the step S4 includes:
S401, Chinese medicine ancient Chinese prose corpus is read in, the Chinese medicine ancient Chinese prose corpus is segmented, word segmentation result is obtained;
S402, use "." two Chinese medicine ancient Chinese prose corpus are subjected to subordinate sentences, retain after subordinate sentence it is original ".", and filter out institute
The sentence that simple sentence number of words in Chinese medicine ancient Chinese prose corpus is less than 4 is stated, without mark;
S403, it converts corpus to the dimension model that tie/break is connected/disconnected, adds "<s>O in one beginning
None S " row is as one opening flag;For after simple sentence participle as a result, index index is traversed, if the index
Index then illustrates not arriving entity word, i.e., not equal to the starting index index of next entity of the result marked in third step
The index is indexed into the mould that word all in the starting index index of next entity word is designated as " word+I None S "
Formula, and a word a line;On the contrary, illustrating if the index index is equal to the starting index index of next entity word
Have found the entity word;If the entity word is corresponding with type, i.e., first word of the entity word is labeled as "
Subsequent all words of the entity word are labeled as " word+O+ corresponding types+S " by word+I+ corresponding types+S ";
If the entity word, without corresponding types, all words for including by the entity word are labeled as " word+O None
The mode of D ";The operation is recycled, until subsequent all words are designated as " word+I without possible entity word in simple sentence
None S";Finally, finally filling "<eof>I None S " row as one end mark in simple sentence;In this dimension model
The mark of word accounts for a line;
In the dimension model, it is not possible entity that " I None S ", which represents the entity word,;"O None D"
The type for representing the possible entity of entity word is unknown;The beginning that " I+ corresponding types+S " represents single entity is
Break is disconnected;Centre or the end that " O+ corresponding types+S " represents single entity word are tie connection;So far it completes
Conversion of the Chinese medicine ancient Chinese prose corpus to the tie/break dimension model connected/disconnected.
Wherein, the step S5 includes:
S501, pre-training, the word2vec word distinguished to vector are carried out according to the result after Chinese medicine ancient Chinese prose and participle
As a result, the namely corresponding words vector of Chinese medicine corpus;
S502, the mark that will occur in annotation results: '<s>', '<unk>', '<>', '<\n>' also training obtain it is corresponding
Word2vec word is respectively added in the words vector to vector, obtains complete words vector with this;
S503, according to word, the word occurred in the Chinese medicine ancient Chinese prose corpus, the complete words vector is filtered,
Leave behind word, the word occurred in Chinese medicine ancient Chinese prose corpus;Multiple Chinese medicine corpus are handled using a complete words vector;
The mistake being likely to occur in S504, detection annotation results;If it was found that there is the number less than 4 parts in annotation results
According to then judging are as follows: there are problems for mark;
The entity type occurred in S505, automatic detection annotation results, and return and obtain all types of dictionaries, carry out letter
Single sequence, to obtain the Embedding insertion of all entity type and corresponding type;And by tie/break connection/
It disconnects making simply to sort and connects/disconnects corresponding test data to obtain corresponding dictionary to get to tie/break
Embedding insertion;
S506, it is embedded according to the annotation results and words vector of input and the Embedding, obtains training data
Collection;Sentence is ranked up according to sentence length, finally the result by the Embedding insertion of training labeled data is output to instruction
Practice and collects in corresponding trained file;
S507, according to the verification result of input, test result and words vector, be verified data set and test data
Collection;And be ranked up according to sentence length, finally return to the Embedding insertion of verify data and test data as a result, simultaneously
The Embedding insertion and verifying that Embedding insertion and tie/break by words vector, entity type connect/disconnect
Data set and the Embedding of test data set insertion are output in test file.
Wherein, the step S6 includes:
S601, file path, test file path, the outgoing route of model, will be trained obtained in the S5 step
Habit rate, hidden layer dimension, forgetting rate, term vector dimension, epoch stage and optimizer algorithm are passed in training pattern;
S602, in model training, reading test file first reads in type label, tie/break and connects/disconnects mark
Label and validation data set and test data set;Then training file is read in, training dataset is read in;NER model is set, then
Trained words vector in advance is loaded, the parameter of corresponding optimizer is set;And in the training process every one section of wheel number or
Test obtains accuracy rate and recall rate and F1 value to a part of data of person's training i.e. on verifying collection, and record optimal F1 value with
And the corresponding checkpoint checkpoint, and tested on test set;If it was found that the effect on verifying collection does not increase
Afterwards, i.e. adjusting parameter;After multiple checks, best checkpoint and best evaluation evaluation index numerical value can be obtained;
S603, the model path by the best checkpoint of S602 final entry, finally saved and the Chinese medicine for needing to predict are ancient
In the program of the incoming coding forecast set of literary corpus, it will predict that corpus carries out Embedding according to incoming prediction corpus and model
Insertion, the predictive data set after being encoded;
S604, the model path by the checkpoint of predictive data set and record, finally saved, hidden layer dimension, forgetting rate,
In the program of the incoming prediction of term vector dimensional parameter;Predictive data set is read in, NER Named Entity Extraction Model is set, for every
Sentence input predicted respectively as a result, and being saved in decoding result file;
S605, the decoding result file for obtaining prediction are read in, and are obtained different types of entity according to type-collection and are protected
It is stored in different files, while saving its position in Chinese medicine ancient Chinese prose corpus, corresponding type and entity word itself;
So far the name Entity recognition task towards Chinese medical book document is completed.
Application scenarios of the invention are described below.
As shown in Fig. 2, being a kind of name entity recognition techniques towards Chinese medical book document of the present invention, including step
It is rapid:
S1, the Chinese medicine for the entity type that one identifies comprising needs is obtained by arranging the entity word of various entity types
Field vocabulary;If the entities such as symptom, medicinal material, prescription need to be identified, then needs arrangement to obtain one in this step and include above three
The vocabulary of class entity, format are that an entity corresponds to corresponding type;
S2, short phrase picking is carried out from Chinese medicine ancient Chinese prose corpus using AutoPhrase technology, obtains all possible entity,
And it arranges and obtains the traditional Chinese medical science field vocabulary for not including entity type;
S3, mark strategy is returned in conjunction with above two vocabulary and according to design, marks out and is likely to occur in Chinese medicine ancient Chinese prose corpus
Entity;
S4, in conjunction with Chinese medicine ancient Chinese prose corpus return mark result and tie/break dimension model obtain the labeled data of ancient Chinese prose;
S5, instruction is generated in conjunction with labeled data and according to the pre-training model Word Embedding that original language material training obtains
Practice, verifying, test data;Training set: it is used to training pattern;Verifying collection: for the effect for adjusting model parameter to be optimal;
Test set:
S6, test model performance, training AutoNER model simultaneously are predicted to obtain using the model that training obtains to corpus
The result of identification;The entity identified according to result.Format: the first and second column are line number of the entity in original language material, three column
It is entity, four, five column are the id and entity type of entity type.
It is further comprising the steps of in the implementation steps of step S1:
S101, the entity word for arranging various entity types, including entity is cleared up, delete space, punctuation mark
Deng;
S102, the biggish entity of ambiguousness in existing vocabulary is deleted, such as the entity of individual character, the entity and number of words of multiple meaning
Excessive entity finally carries out the operation of duplicate removal to the entity of same type, obtains the word for the entity composition for needing the type identified
Table.
It is further comprising the steps of in the implementation steps of step S2:
What S201, the input and output path in modification AutoPhrase script, the word frequency of short phrase picking and program were run
Thread Count;
S202, safeguard that addition needs the word and word filtered according to deactivated vocabulary of the original language material to Chinese;
S203, the high quality phrase in Chinese medicine ancient Chinese prose corpus is added in vocabulary to promote the quality of short phrase picking, then
Carry out short phrase picking, short phrase picking the result is that the vocabulary of the not belt type to be sorted from high to low according to phrase mass fraction.
It is further comprising the steps of in the implementation steps of step S3:
301, before marking out the entity being likely to occur in original language material, merge two generated in one or two steps first
A vocabulary.Firstly for the vocabulary of belt type, it is only necessary to read in entity and corresponding type;
S302, the entity excavated for AutoPhrase, first clear up the word of the part, and it is empty to delete front and back
Lattice, newline, and the entity of individual character is filtered out, because its singularity is too big.And two believable threshold values are set, respectively for
The entity of word and more words composition is screened, and the threshold value of Manual definition's confidence level can be merged into final vocabulary more than threshold value
In, entity need to be only saved when merging, entity type is set as NULL;
S303, then illustrate that designs returns mark strategy:
Entity returns mark and is roughly divided into three kinds of situations:
(1), a long word includes multiple short words, for example hypodynia includes hypodynia in abdomen
(2), long word is fully composed by multiple short words, for example is suffered from vomiting and diarrhoea to be dropped by upper Tuhe and be formed
(3), two word intersection conflicts, such as diet are divided into two kinds of diet, deficiency of food point-scores less.
For the first case, using the strategy of priority of long word, and the type of upper entity is marked;For second of feelings
The each word of the long word is labeled as NULL by condition;For the third situation, multiple words are all designated as NULL;
S304, according to mark strategy is returned, for each sentence in Chinese medical book document, return the entity being wherein likely to occur and
Corresponding entity type, including its accurate entity type and NULL, for the part NULL, though entity is not indicated in mark
Type will not have an impact model in model training, but can based on context predict to obtain part possibility in prediction
Entity type.
It is further comprising the steps of in the implementation steps of step S4:
S401, original language material is read in first, after then using the segmenting method towards Chinese medical book document to obtain participle
Result;
S402, use "." two corpus are subjected to subordinate sentences, retain after subordinate sentence it is original ".", and filter out in original language material
Simple sentence number of words is less than 4 sentence, without mark;
S403, the dimension model for converting corpus to tie/break, in one beginning addition "<s>O None S " row
As one opening flag.For after simple sentence participle as a result, traversal indexes index, if the index is not equal to third step
The starting index of next entity of the result of middle mark then illustrates not arrive entity, i.e., by the index to next entity
All words are designated as the mode of " word+I None S ", and a word a line in starting index;If the opposite index is equal to next
The starting index of a entity, then explanation has found the entity, if the entity has type, i.e., marks first word of the entity
For " word+I+ corresponding types+S ", subsequent all words of the entity are labeled as " word+O+ corresponding types+S ";
If the entity, without type, all words for including by the entity are labeled as the mode of " word+O None D ".Circulation should
Operation, until subsequent all words are designated as " word+I None S " without possible entity in simple sentence.Finally simple sentence most
"<eof>I None S " row is filled afterwards as one end mark.
The mark of word accounts for a line in this dimension model.In this dimension model, " I None S " represent the word be not can
The entity of energy;The type that " O None D " represents the possible entity is unknown;" I+ corresponding types+S " represents opening for single entity
Begin, is break;Centre or the end that " O+ corresponding types+S " represents single entity, are tie;So far Chinese medicine is completed
Corpus to tie/break dimension model conversion.
It is further comprising the steps of in the implementation steps of step S5:
S501, it pre-training is carried out according to the result after original language material and participle respectively obtains word2vec's as a result, also
It is the corresponding word of Chinese medicine corpus, term vector;
S502, the mark that will occur in annotation results, such as '<s>', '<unk>', '<>', '<\n>' also training corresponded to
Word2vec, and be respectively added in words vector obtained in the previous step, complete words vector obtained with this;
Word, term vector that previous step training obtains are filtered, only by S503, the word according to occurring in original language material, word
The word occurred in original language material, word are left, this step can be filled into original word term vector not in the corpus of this training
Word, the word of appearance, to reduce memory consumption.And a more complete word, term vector can be used to multiple Chinese medicine corpus
It is handled, eliminates the process that training every time requires re -training words vector;
The mistake being likely to occur in S504, detection annotation results, if having the number less than 4 parts in discovery annotation results
According to, then illustrate mark there are problems;
The entity type occurred in S505, automatic detection annotation results, and return and obtain all types of dictionaries, carry out letter
Single sequence, to obtain the Embedding of all entity type and corresponding type;And tie/break is made simple
Sequence arrives the corresponding Embedding of tie/break to obtain corresponding dictionary;
S506, training data is obtained according to the annotation results and words vector and the Embedding of previous step of input
Collection.And be ranked up according to sentence length, it is corresponding that the result of the Embedding of training labeled data is finally output to training set
Training file in;
S507, verifying annotation results, test annotation results and words vector according to input.And it is carried out according to sentence length
Sequence finally returns to verifying labeled data and tests the Embedding's as a result, and by words vector, entity class of labeled data
The Embedding of the Embedding of the Embedding and tie/break of type and verifying and test data is output to test
In file.
It is further comprising the steps of in the implementation steps of step S6:
S601, will obtained in S5 step training file path, test file path, the outgoing route of model, learning rate,
Hidden layer dimension, droprate, term vector dimension, epoch and optimizer algorithm etc. are passed in training pattern;Model and algorithm
Specifically: in AutoNER on existing model training program, it is added to the process of load word vector, it is other the same.The mould
Type uses the network structure based on Recognition with Recurrent Neural Network RNN, is obtained by incoming data set and model parameter training
AutoNER model.
S602, in model training, reading test file first reads in type label, tie/break label and verifying
Collection and test set;Then training file is read in, training set is read in;Be arranged NER model, then load in advance trained words to
Amount, is arranged the parameter of corresponding optimizer, such as type, the initial learning rate of optimizer, such as;Finally start training for model
Journey;And every one section of wheel number or a part of data of training, test obtains accuracy rate and calls together i.e. on verifying collection in the training process
Rate and F1 value are returned, and records optimal F1 value and corresponding checkpoint, and tested on test set;If it was found that
After the effect on verifying collection does not increase, i.e. adjusting parameter, such as learning rate;After multiple check, it can be obtained optimal
Checkpoint and optimal evaluation evaluation index numerical value;
S603, the model path by the checkpoint of S602 final entry, finally saved and the original language for needing to predict
In the program for expecting incoming encode forecast set, it will predict that corpus carries out Embedding, obtains according to incoming prediction corpus and model
Predictive data set after to coding;
S604, the model path by the checkpoint of predictive data set and record, finally saved, hidden layer dimension,
In the program of the incoming prediction of the parameters such as droprate, term vector dimension;Predictive data set is read in, NER model is set, for every
Input predicted respectively as a result, and being saved in decoding result file;
S605, the decoding result file for obtaining prediction are read in, and are obtained different types of entity according to type-collection and are protected
It is stored in different files, while saving its position in original language material, corresponding type and entity itself;So far i.e. complete
At the name Entity recognition task towards Chinese medical book document.
In the present invention, model training evaluation index is as shown in Figure 8.
Original Chinese ancient Chinese prose corpus are as follows:
Taishan leaf grandson Xiang Guoyou avoids as taboo beneficial bud person, and body such as bright charcoal when prolonging remaining examine, opisthotonos, coma is in silence.Examine complete, Xiang Guowen
It says;Does arteries and veins dissipate no? can you rescue no? it is remaining to say;In a panic without labor, without delivering, Gu Mai does not dissipate still this typhoid fever Grand Tutor, and disease gesture is also light, can answer
Hand takes effect.Prime minister says;It is easier said than done, i.e., it with treatise on Febrile Diseases, can deliver within 1st or two, preferably conciliate within 3rd or four, it is 14 days modern,
An Ganbiao? it is remaining to say;There is exterior syndrome table arteries and veins Ju to see, though more time again, still might as well with opening pores method solution it.High fever like the burning coal over the body, a sweat are
It dissipates, is monarch with Rhizoma Et Radix Notopterygii then, pueraria lobata, radix bupleuri, cimicifugae foetidae are minister, and Rhizoma Chuanxiong, purple perilla, radix paeoniae rubra are assistant, and Chinese ephedra is to make, about 1 Qian Chongxian
Ginger five, even must be Bulbus Allii Fistulosi five, two bowls of water, decoct one bowl.Hot drink and sweat such as rain are infused, and disease is recover, without tired medicine again.Public affairs, which are sighed, to be said;Preceding doctor
Fire syndrome is takeed for, it is more to take freshener, so disease is more, public celestial, the evening what is met each other.It is remaining to say;Without very surprise, but feel the pulse not
Accidentally, the appropriate ear of medication weight.
After subordinate sentence:
Taishan leaf grandson Xiang Guoyou avoids as taboo beneficial bud person, and body such as bright charcoal when prolonging remaining examine, opisthotonos, coma is in silence.
Examine complete, prime minister, which asks, to be said;Does arteries and veins dissipate no? can you rescue no? it is remaining to say;In a panic without labor, this typhoid fever is without delivering, and Gu Mai is still by Grand Tutor
It does not dissipate, disease gesture is also light, hand can be answered to take effect.
Prime minister says;It is easier said than done, i.e., it with treatise on Febrile Diseases, can deliver within 1st or two, preferably conciliate within 3rd or four, it is 14 days modern
, An Ganbiao? it is remaining to say;There is exterior syndrome table arteries and veins Ju to see, though more time again, still might as well with opening pores method solution it.
High fever like the burning coal over the body, a sweat dissipate, and are monarch with Rhizoma Et Radix Notopterygii then, and pueraria lobata, radix bupleuri, cimicifugae foetidae are minister, and Rhizoma Chuanxiong, purple perilla, radix paeoniae rubra are assistant,
Chinese ephedra is to make, and about 1 money weight fresh gingers five even must be Bulbus Allii Fistulosi five, two bowls of water, decocts one bowl.
Hot drink and sweat such as rain are infused, and disease is recover, without tired medicine again.
Public affairs, which are sighed, to be said;Preceding doctor takes for fire syndrome, and it is more to take freshener, so disease is more, public celestial, the evening what is met each other.
It is remaining to say;Without very surprise, but feeling the pulse does not miss, the appropriate ear of medication weight.
Result after participle:
Taishan leaf grandson Xiang Guoyou avoids as taboo beneficial bud person, and body such as bright charcoal when prolonging remaining examine, opisthotonos, coma is in silence.
Examine complete, prime minister, which asks, to be said;Does arteries and veins dissipate no? can you rescue no? it is remaining to say;In a panic without labor, this typhoid fever is without delivering, and Gu Mai is still by Grand Tutor
It does not dissipate, disease gesture is also light, hand can be answered to take effect.
Prime minister says;It is easier said than done, i.e., it with treatise on Febrile Diseases, can deliver within 1st or two, preferably conciliate within 3rd or four, it is 14 days modern
, An Ganbiao? it is remaining to say;There is exterior syndrome table arteries and veins Ju to see, though more time again, still might as well with opening pores method solution it.
High fever like the burning coal over the body, a sweat dissipate, and are monarch with Rhizoma Et Radix Notopterygii then, and pueraria lobata, radix bupleuri, cimicifugae foetidae are minister, and Rhizoma Chuanxiong, purple perilla, radix paeoniae rubra are assistant,
Chinese ephedra is to make, and about 1 money weight fresh gingers five even must be Bulbus Allii Fistulosi five, two bowls of water, decocts one bowl.
Hot drink and sweat such as rain are infused, and disease is recover, without tired medicine again.
Public affairs, which are sighed, to be said;Preceding doctor takes for fire syndrome, and it is more to take freshener, so disease is more, public celestial, the evening what is met each other.
It is remaining to say;Without very surprise, but feeling the pulse does not miss, the appropriate ear of medication weight.
As shown in figure 9, being a kind of name entity recognition device towards Chinese medical book document of the present invention, comprising:
Finishing unit 41 arranges the entity word of at least one entity type, obtains one and includes entity class to be identified
First traditional Chinese medical science field vocabulary of type;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;
Unit 42 is excavated, using the automatic short phrase picking technology of AutoPhrase, phrase digging is carried out from Chinese medicine ancient Chinese prose corpus
Pick, obtains all possible entity word, obtains the second traditional Chinese medical science field vocabulary, the second traditional Chinese medical science field vocabulary includes entity word
Language;
Unit 43 is marked, in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary, according to scheduled
Mark strategy is returned, the entity occurred in the Chinese medicine ancient Chinese prose corpus is marked out;
Processing unit 44 connects/disconnects mark mould in conjunction with time mark result and tie/break of the Chinese medicine ancient Chinese prose corpus
Formula obtains the labeled data of Chinese medicine ancient Chinese prose corpus;
Output unit 45, in conjunction with the labeled data and the pre-training model obtained according to Chinese medicine ancient Chinese prose corpus training
The insertion of Word Embedding word, generates training dataset, validation data set, test data set, training dataset is output to
In training file, validation data set and test data set are output in test file;
Predicting unit 46 reads in data from the trained file, test file, according to reading data training
The automatic Named Entity Extraction Model of AutoNER, and the automatic Named Entity Extraction Model of the AutoNER obtained using training,
The Chinese medicine ancient Chinese prose corpus is predicted, the result identified;The entity identified according to result.
The present invention provides a kind of name entity recognition techniques towards Chinese medical book document, by using short phrase picking technology
AutoPhrase, the segmenting method (patent applied for) towards Chinese medical book document and modified English entity recognition techniques
AutoNER carries out the name Entity recognition for Chinese medicine ancient Chinese prose corpus.The described method includes: by arranging various entity types
Entity word obtains one and includes the traditional Chinese medical science field vocabulary for the entity type for needing to identify;Using AutoPhrase technology from Chinese medicine
The traditional Chinese medical science field vocabulary of occurred phrase composition is obtained in ancient Chinese prose corpus;By the design of above two vocabulary knot merga pass
Strategy is returned and is marked back in original Chinese medicine ancient Chinese prose corpus;Chinese medicine is obtained in conjunction with time mark result and newest tie/break dimension model
The labeled data of ancient Chinese prose corpus;In conjunction with labeled data and the pre-training Word Embedding life obtained according to original language material training
At training, verifying, test data;Training AutoNER model and the model obtained using training carry out Chinese medicine ancient Chinese prose corpus pre-
Measure the result of identification;The entity identified according to result.Name entity recognition techniques of the invention are based on remote supervisory
Method does not need mark corpus, does not need manually to extract feature yet, it is only necessary to the vocabulary of entity type to be identified, it can be right
Entity in Chinese medical book document carries out identification efficiently, intelligent.Method proposed by the present invention can use remote supervisory, only
The vocabulary for needing to use traditional Chinese medical science field does not need manually, cost to be greatly saved.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, described program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be
Magnetic disk, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by those familiar with the art, all answers
It is included within the scope of the present invention.Therefore, protection scope of the present invention should be subject to the protection scope in claims.
Claims (8)
1. a kind of name entity recognition method towards Chinese medical book document characterized by comprising
S1, the entity word for arranging at least one entity type obtain first Chinese medicine comprising entity type to be identified
Field vocabulary;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;
S2, using the automatic short phrase picking technology of AutoPhrase, carry out short phrase picking from Chinese medicine ancient Chinese prose corpus, obtain it is all can
The entity word of energy, obtains the second traditional Chinese medical science field vocabulary, the second traditional Chinese medical science field vocabulary includes entity word;
S3, it is marked in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary according to scheduled time mark strategy
The entity occurred in the Chinese medicine ancient Chinese prose corpus out;
S4, in conjunction with the Chinese medicine ancient Chinese prose corpus return mark result and tie/break connect/disconnect dimension model, obtain Chinese medicine Gu
The labeled data of literary corpus;
S5, in conjunction with the labeled data and according to the obtained pre-training model Word of Chinese medicine ancient Chinese prose corpus training
The insertion of Embedding word, generates training dataset, validation data set, test data set, and training dataset is output to training text
In part, validation data set and test data set are output in test file;
S6, data are read in from the trained file, test file, is named automatically according to reading data training AutoNER
Entity recognition model, and the automatic Named Entity Extraction Model of the AutoNER obtained using training, to the Chinese medicine ancient Chinese prose language
Expect to be predicted, the result identified;The entity identified according to result.
2. the method according to claim 1, wherein the step S1 includes:
S101, the entity word for arranging at least one entity type, the arrangement includes: to clear up entity, deletion space,
Punctuation mark;
S102, the entity for having ambiguousness in existing vocabulary is deleted, and carries out the operation of duplicate removal to the entity of same type, obtained wait know
First traditional Chinese medical science field vocabulary of the entity composition of other type.
3. the method according to claim 1, wherein the step S2 includes:
S201, setting the automatic short phrase picking script of AutoPhrase in input and output path, short phrase picking word frequency and journey
The Thread Count of sort run;
S202, according to the Chinese medicine ancient Chinese prose corpus, the deactivated vocabulary of Chinese is safeguarded, addition needs the word and word filtered;
S203, add the TCM Major term in Chinese medicine ancient Chinese prose corpus in vocabulary to promote the quality of short phrase picking, then into
Row short phrase picking, generate the second traditional Chinese medical science field vocabulary, the second traditional Chinese medical science field vocabulary be according to phrase mass fraction from height to
The vocabulary of low sequence.
4. the method according to claim 1, wherein the step S3 includes:
301, merge the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary;For the first traditional Chinese medical science field word
Table reads in entity and corresponding type;
S302, the second traditional Chinese medical science field vocabulary is cleared up, deletes front and back space, newline, and filter out individual character
Entity;Two credible threshold values are set, it is more than the credible threshold value that the entity formed respectively for word and more words, which is screened,
It is then merged into final vocabulary, entity word is saved when merging, corresponding entity type is set as NULL;
S303, according to mark strategy is returned, for each sentence in Chinese medicine ancient Chinese prose, return to the entity being wherein likely to occur and corresponding
Entity type;
Described time mark strategy includes:
Short word situation is contained at least two for a long word, using the strategy of priority of long word, and marks the corresponding reality of word
Body type;Situation is formed by least two short words for long word, the entity type of the short word of each of the long word is labeled as
NULL;The case where at least two word intersections conflict, the entity type of described two words is all labeled as NULL.
5. the method according to claim 1, wherein the step S4 includes:
S401, Chinese medicine ancient Chinese prose corpus is read in, the Chinese medicine ancient Chinese prose corpus is segmented, word segmentation result is obtained;
S402, use "." two Chinese medicine ancient Chinese prose corpus are subjected to subordinate sentences, retain after subordinate sentence it is original ".", and filter out in described
The sentence that simple sentence number of words in ancient Chinese prose corpus is less than 4 is cured, without mark;
S403, it converts corpus to the dimension model that tie/break is connected/disconnected, adds "<s>O None in one beginning
S " row is as one opening flag;For after simple sentence participle as a result, index index is traversed, if the index index is not
Equal to the starting index index of next entity of the result marked in third step, then illustrate not arriving entity word, i.e., it will be described
Index indexes the mode that word all in the starting index index of next entity word is designated as " word+I None S ", and
And word a line;On the contrary, illustrating to have found if the index index is equal to the starting index index of next entity word
The entity word;If the entity word is corresponding with type, i.e., first word of the entity word is labeled as " word+I+
Subsequent all words of the entity word are labeled as " word+O+ corresponding types+S " by corresponding types+S ";
If the entity word, without corresponding types, all words for including by the entity word are labeled as " word+O None D "
Mode;The operation is recycled, until without possible entity word subsequent all words to be designated as to " word+I None in simple sentence
S";Finally, finally filling "<eof>I None S " row as one end mark in simple sentence;Word in this dimension model
Mark account for a line;
In the dimension model, it is not possible entity that " I None S ", which represents the entity word,;" O None D " is represented
The type of the possible entity of entity word is unknown;The beginning that " I+ corresponding types+S " represents single entity is break disconnected
It opens;Centre or the end that " O+ corresponding types+S " represents single entity word are tie connection;So far Chinese medicine Gu is completed
Conversion of the literary corpus to the tie/break dimension model connected/disconnected.
6. the method according to claim 1, wherein the step S5 includes:
S501, according to Chinese medicine ancient Chinese prose and participle after result carry out pre-training, respectively obtain word2vec word to vector as a result,
The namely corresponding words vector of Chinese medicine corpus;
S502, the mark that will occur in annotation results: '<s>', '<unk>', '<>', '<\n>' also training obtain it is corresponding
Word2vec word is respectively added in the words vector to vector, obtains complete words vector with this;
S503, according to word, the word occurred in the Chinese medicine ancient Chinese prose corpus, the complete words vector is filtered, is only stayed
Word, the word occurred in lower Chinese medicine ancient Chinese prose corpus;Multiple Chinese medicine corpus are handled using a complete words vector;
The mistake being likely to occur in S504, detection annotation results;If it was found that there is the data less than 4 parts in annotation results,
Judgement are as follows: there are problems for mark;
The entity type occurred in S505, automatic detection annotation results, and return and obtain all types of dictionaries, it carries out simple
Sequence, to obtain the Embedding insertion of all entity type and corresponding type;And tie/break is connected/disconnected
Make simple sequence and connects/disconnects corresponding test data Embedding to obtain corresponding dictionary to get to tie/break
Insertion;
S506, it is embedded according to the annotation results and words vector of input and the Embedding, obtains training dataset;Root
Sentence is ranked up according to sentence length, finally the result by the Embedding insertion of training labeled data is output to training set
In corresponding trained file;
S507, according to the verification result of input, test result and words vector, be verified data set and test data set;And
It is ranked up according to sentence length, finally return to the Embedding insertion of verify data and test data as a result, and by words
The Embedding insertion and validation data set that vector, the Embedding insertion of entity type and tie/break are connected/disconnected
Embedding insertion with test data set is output in test file.
7. the method according to claim 1, wherein the step S6 includes:
S601, will obtained in the S5 step training file path, test file path, the outgoing route of model, learning rate,
Hidden layer dimension, forgetting rate, term vector dimension, epoch stage and optimizer algorithm are passed in training pattern;
S602, in model training, reading test file first, read in type label, tie/break connect/disconnect label with
And validation data set and test data set;Then training file is read in, training dataset is read in;NER model is set, is then loaded
Preparatory trained words vector, is arranged the parameter of corresponding optimizer;And in the training process every one section of wheel number or instruction
Practicing a part of data, test obtains accuracy rate and recall rate and F1 value i.e. on verifying collection, and records optimal F1 value and right
The checkpoint checkpoint answered, and tested on test set;If it was found that verifying collection on effect do not increase after, i.e.,
Adjusting parameter;After multiple checks, best checkpoint and best evaluation evaluation index numerical value can be obtained;
S603, the model path by the best checkpoint of S602 final entry, finally saved and the Chinese medicine ancient Chinese prose language for needing to predict
In the program of the incoming coding forecast set of material, it will predict that corpus carries out Embedding insertion according to incoming prediction corpus and model,
Predictive data set after being encoded;
S604, the model path by the checkpoint of predictive data set and record, finally saved, hidden layer dimension, forgetting rate, word to
In the program for measuring the incoming prediction of dimensional parameter;Predictive data set is read in, NER Named Entity Extraction Model is set, it is defeated for every
Enter being predicted respectively as a result, and being saved in decoding result file;
S605, the decoding result file for obtaining prediction are read in, and are obtained different types of entity according to type-collection and are saved in
In different files, while saving its position in Chinese medicine ancient Chinese prose corpus, corresponding type and entity word itself;So far
Complete the name Entity recognition task towards Chinese medical book document.
8. a kind of name entity recognition device towards Chinese medical book document characterized by comprising
Finishing unit arranges the entity word of at least one entity type, obtains the comprising entity type to be identified
One traditional Chinese medical science field vocabulary;The first traditional Chinese medical science field vocabulary includes entity word and corresponding entity type;
Unit is excavated to carry out short phrase picking using the automatic short phrase picking technology of AutoPhrase from Chinese medicine ancient Chinese prose corpus, obtain
All possible entity word, obtains the second traditional Chinese medical science field vocabulary, and the second traditional Chinese medical science field vocabulary includes entity word;
Unit is marked, in conjunction with the first traditional Chinese medical science field vocabulary and the second traditional Chinese medical science field vocabulary, according to scheduled time mark plan
Slightly, the entity occurred in the Chinese medicine ancient Chinese prose corpus is marked out;
Processing unit connects/disconnects dimension model in conjunction with time mark result and tie/break of the Chinese medicine ancient Chinese prose corpus, obtains
The labeled data of Chinese medicine ancient Chinese prose corpus;
Output unit, in conjunction with the labeled data and the pre-training model Word obtained according to Chinese medicine ancient Chinese prose corpus training
The insertion of Embedding word, generates training dataset, validation data set, test data set, and training dataset is output to training text
In part, validation data set and test data set are output in test file;
Predicting unit reads in data from the trained file, test file, certainly according to reading data training AutoNER
Dynamic Named Entity Extraction Model, and the automatic Named Entity Extraction Model of the AutoNER obtained using training, to the Chinese medicine
Ancient Chinese prose corpus predicted, the result identified;The entity identified according to result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910340359.6A CN110321550A (en) | 2019-04-25 | 2019-04-25 | A kind of name entity recognition method and device towards Chinese medical book document |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910340359.6A CN110321550A (en) | 2019-04-25 | 2019-04-25 | A kind of name entity recognition method and device towards Chinese medical book document |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110321550A true CN110321550A (en) | 2019-10-11 |
Family
ID=68113398
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910340359.6A Pending CN110321550A (en) | 2019-04-25 | 2019-04-25 | A kind of name entity recognition method and device towards Chinese medical book document |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110321550A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259626A (en) * | 2020-01-16 | 2020-06-09 | 上海国民集团健康科技有限公司 | Traditional Chinese medicine entity recognition algorithm |
CN112507703A (en) * | 2020-12-07 | 2021-03-16 | 医渡云(北京)技术有限公司 | Medical entity identification method, device, medium and electronic equipment |
CN113111656A (en) * | 2020-01-13 | 2021-07-13 | 腾讯科技(深圳)有限公司 | Entity identification method, entity identification device, computer readable storage medium and computer equipment |
CN113283244A (en) * | 2021-07-20 | 2021-08-20 | 湖南达德曼宁信息技术有限公司 | Pre-training model-based bidding data named entity identification method |
CN114154514A (en) * | 2022-02-07 | 2022-03-08 | 北京三子健康科技有限公司 | Traditional Chinese medicine syndrome type identification method and system |
CN114912450A (en) * | 2022-07-18 | 2022-08-16 | 天津恒达文博科技股份有限公司 | Information generation method and device, training method, electronic device and storage medium |
CN116525100A (en) * | 2023-04-26 | 2023-08-01 | 脉景(杭州)健康管理有限公司 | Traditional Chinese medicine prescription reverse verification method and system based on label system |
CN116796750A (en) * | 2023-08-24 | 2023-09-22 | 宁波甬恒瑶瑶智能科技有限公司 | NER model-based gene literature information extraction method, system and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9734193B2 (en) * | 2014-05-30 | 2017-08-15 | Apple Inc. | Determining domain salience ranking from ambiguous words in natural speech |
CN107092674A (en) * | 2017-04-14 | 2017-08-25 | 福建工程学院 | The automatic abstracting method and system of a kind of Chinese medicine acupuncture field event trigger word |
US9836453B2 (en) * | 2015-08-27 | 2017-12-05 | Conduent Business Services, Llc | Document-specific gazetteers for named entity recognition |
CN108829681A (en) * | 2018-06-28 | 2018-11-16 | 北京神州泰岳软件股份有限公司 | A kind of name entity extraction method and device |
-
2019
- 2019-04-25 CN CN201910340359.6A patent/CN110321550A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9734193B2 (en) * | 2014-05-30 | 2017-08-15 | Apple Inc. | Determining domain salience ranking from ambiguous words in natural speech |
US9836453B2 (en) * | 2015-08-27 | 2017-12-05 | Conduent Business Services, Llc | Document-specific gazetteers for named entity recognition |
CN107092674A (en) * | 2017-04-14 | 2017-08-25 | 福建工程学院 | The automatic abstracting method and system of a kind of Chinese medicine acupuncture field event trigger word |
CN108829681A (en) * | 2018-06-28 | 2018-11-16 | 北京神州泰岳软件股份有限公司 | A kind of name entity extraction method and device |
Non-Patent Citations (3)
Title |
---|
JINGBO SHANG ET.AL: "Automated Phrase Mining from Massive Text Corpora", 《HTTPS://ARXIV.ORG/ABS/1702.04457V2》 * |
JINGBO SHANG ET.AL: "Learning Named Entity Tagger using Domain-Specific Dictionary", 《HTTPS://ARXIV.ORG/ABS/1809.03599》 * |
高甦 有情: "基于深度学习的中医典籍命名实体识别研究", 《情报工程》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113111656A (en) * | 2020-01-13 | 2021-07-13 | 腾讯科技(深圳)有限公司 | Entity identification method, entity identification device, computer readable storage medium and computer equipment |
CN113111656B (en) * | 2020-01-13 | 2023-10-31 | 腾讯科技(深圳)有限公司 | Entity identification method, entity identification device, computer readable storage medium and computer equipment |
CN111259626A (en) * | 2020-01-16 | 2020-06-09 | 上海国民集团健康科技有限公司 | Traditional Chinese medicine entity recognition algorithm |
CN112507703A (en) * | 2020-12-07 | 2021-03-16 | 医渡云(北京)技术有限公司 | Medical entity identification method, device, medium and electronic equipment |
CN113283244A (en) * | 2021-07-20 | 2021-08-20 | 湖南达德曼宁信息技术有限公司 | Pre-training model-based bidding data named entity identification method |
CN113283244B (en) * | 2021-07-20 | 2021-10-01 | 湖南达德曼宁信息技术有限公司 | Pre-training model-based bidding data named entity identification method |
CN114154514A (en) * | 2022-02-07 | 2022-03-08 | 北京三子健康科技有限公司 | Traditional Chinese medicine syndrome type identification method and system |
CN114912450A (en) * | 2022-07-18 | 2022-08-16 | 天津恒达文博科技股份有限公司 | Information generation method and device, training method, electronic device and storage medium |
CN114912450B (en) * | 2022-07-18 | 2022-09-27 | 天津恒达文博科技股份有限公司 | Information generation method and device, training method, electronic device and storage medium |
CN116525100A (en) * | 2023-04-26 | 2023-08-01 | 脉景(杭州)健康管理有限公司 | Traditional Chinese medicine prescription reverse verification method and system based on label system |
CN116796750A (en) * | 2023-08-24 | 2023-09-22 | 宁波甬恒瑶瑶智能科技有限公司 | NER model-based gene literature information extraction method, system and storage medium |
CN116796750B (en) * | 2023-08-24 | 2023-11-10 | 宁波甬恒瑶瑶智能科技有限公司 | NER model-based gene literature information extraction method, system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110321550A (en) | A kind of name entity recognition method and device towards Chinese medical book document | |
CN110968700B (en) | Method and device for constructing domain event map integrating multiple types of affairs and entity knowledge | |
CN108073569B (en) | Law cognition method, device and medium based on multi-level multi-dimensional semantic understanding | |
CN108345647A (en) | Domain knowledge map construction system and method based on Web | |
Shapiro et al. | Language and society in South Asia | |
CN107076567A (en) | Multilingual image question and answer | |
CN104137097B (en) | Predicate template gathering device, and specified phrase pair gathering device | |
US20160350288A1 (en) | Multilingual embeddings for natural language processing | |
CN111914532B (en) | Chinese composition scoring method | |
CN106104519B (en) | Phrase is to collection device and computer-readable storage medium | |
CN105808524A (en) | Patent document abstract-based automatic patent classification method | |
CN106104524B (en) | Complex predicate template collection device and recording medium | |
CN109960727A (en) | For the individual privacy information automatic testing method and system of non-structured text | |
CN103092966A (en) | Vocabulary mining method and device | |
CN107145514A (en) | Chinese sentence pattern sorting technique based on decision tree and SVM mixed models | |
CN108664512A (en) | Text object sorting technique and device | |
CN106844340A (en) | News in brief generation and display methods, apparatus and system based on artificial intelligence | |
CN105701076A (en) | Thesis plagiarism detection method and system | |
CN109740164A (en) | Based on the matched electric power defect rank recognition methods of deep semantic | |
CN108763211A (en) | The automaticabstracting and system of knowledge are contained in fusion | |
CN105701085A (en) | Network duplicate checking method and system | |
Josserand | The narrative structure of hieroglyphic texts at Palenque | |
CN105701086A (en) | Method and system for detecting literature through sliding window | |
CN109657043A (en) | Automatically generate the method, apparatus, equipment and storage medium of article | |
CN110060749B (en) | Intelligent electronic medical record diagnosis method based on SEV-SDG-CNN |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191011 |
|
RJ01 | Rejection of invention patent application after publication |