CA3039280C - Method for recognizing network text named entity based on neural network probability disambiguation - Google Patents
Method for recognizing network text named entity based on neural network probability disambiguation Download PDFInfo
- Publication number
- CA3039280C CA3039280C CA3039280A CA3039280A CA3039280C CA 3039280 C CA3039280 C CA 3039280C CA 3039280 A CA3039280 A CA 3039280A CA 3039280 A CA3039280 A CA 3039280A CA 3039280 C CA3039280 C CA 3039280C
- Authority
- CA
- Canada
- Prior art keywords
- neural network
- word
- named entity
- word vector
- probability
- 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
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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Machine Translation (AREA)
- Character Discrimination (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A method for recognizing network text named entities based on neural network probability disambiguation comprising: carrying out word segmentation on an unlabeled corpus, using Word2Vec to extract a word vector; converting a sample corpus into a word feature matrix and windowing same; building a deep neural network to carry out training, and adding a softmax function into an output layer of the neural network to carry out normalization processing, so as to obtain a probability matrix of the named entity category corresponding to each word; and re-windowing the probability matrix, and using a conditional random field model to carry out disambiguation, so as to obtain a final named entity annotation. A probability disambiguation method is used in order to deal with the problems of a nonstandard grammatical structure and many wrongly written characters in the network text.
Description
TITLE OF THE INVENTION
METHOD FOR RECOGNIZING NETWORK TEXT NAMED ENTITY BASED ON
NEURAL NETWORK PROBABILITY DISAMBIGUATION
TECHNICAL FIELD
The present invention relates to processing and analysis of network text, particularly to a method for recognizing network text named entities based on neural network probability disambiguation.
BACKGROUND ART
Networks have driven the speed and scale of information collection and dissemination to an unprecedented level and brought global information sharing and interaction into reality, and have become an indispensable infrastructure in the information society. Modern communication and dissemination techniques have greatly improved the speed and breadth of information dissemination. However, there are accompanying problems and "side effects":
sometimes people are confused by the turbulent information, and it is very hard to obtain the precise information needed quickly and accurately from the vast sea of information. It is a prerequisite to analyze and obtain named entities, such as people, places, and organizations, etc., concerned by Internet users from within a mas of network text, in order to provide important support information for various higher-level applications such as online marketing, group emotion analysis, etc. Accordingly, network text named entity recognition has become an important core technique in network data processing and analysis.
Two kinds of methods for dealing with named entity recognition are considered in the research, i.e., rule-based method and statistics-based method. As the machine learning theory is consummated continuously and the computing performance is improved greatly, the statistics-based method is increasingly favored.
At present, statistical models and methods applied in named entity recognition mainly include:
hidden Markov model, decision tree, maximum entropy model, support vector machine, conditional random field and artificial neural network. Artificial neural networks can achieve a better result in named entity recognition than conditional random field, maximum entropy model, and other models, but conditional random field and maximum entropy models are still dominant practical models. For example, in the Patent Document No. CN201310182978.X, named entity recognition method and apparatus for MicroBlog text based on conditional random field and named entity library are proposed. In the Patent Document No.
CN200710098635.X, a named entity recognition method utilizing word features and using maximum entropy model to model is proposed. Artificial neural networks are difficult to use practically because they often require the conversion of words into vectors in a word vector space in the field of named entity recognition.
Consequently, artificial neural networks can not be applied in large-scale practical applications, because they are unable to obtain corresponding vectors for new words.
Owing to the above-mentioned present situation, there are mainly the following problems in named entity recognition for network text: firstly, it is unable to train a word vector space that contains all words in order to train a neural network, because there are a lot of network words, new words, and wrongly written or mispronounced characters in network text;
secondly, the accuracy of named entity recognition for network texts is degraded as a result of phenomena existing in network text, such as arbitrary language forms, non-standard grammatical structures, and wrongly written or mispronounced characters, etc.
SUMMARY OF THE INVENTION
The object of the invention is to overcome the drawbacks in the prior art. The present invention provides a network text named entity recognition method based on neural network probability disambiguation, which extracts word features incrementally without neural network retraining, and performs recognition with the aid of probability disambiguation. The method obtains a prediction probability matrix on the named entity category of a word from a neural network by training neural network, and performs disambiguation on the prediction matrix outputted from the neural network in a probability model, and thereby improves accuracy and precision of network text named entity recognition.
In order to attain the object described above, the technical scheme employed by the present invention is as follows.
The network text named entity recognition method is based on neural network probability disambiguation, performing word segmentation on an untagged corpus, utilizing Word2Vec to extract a word vector, converting sample corpora into a word feature matrix and windowing, building a deep neural network for training, adding a softmax function into an output layer of the neural network, and performing normalization, to acquire a probability matrix of named entity categorycorresponding to each word; re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
Specifically, the method comprises the following steps:
step 1: acquiring an untagged corpus by means of a web crawler, acquiring sample corpora with named entity tags from a corpus base, and performing word segmentation on the untagged corpus by a natural language tool;
step 2: performing word vector space training on the segmented untagged corpus and the sample corpora by a Word2Vec tool;
step 3: converting the text in the sample corpora into a word vector representing word features according to the trained Word2Vec model, windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to a neural network; converting the tags in the sample corpora into a one-hot
METHOD FOR RECOGNIZING NETWORK TEXT NAMED ENTITY BASED ON
NEURAL NETWORK PROBABILITY DISAMBIGUATION
TECHNICAL FIELD
The present invention relates to processing and analysis of network text, particularly to a method for recognizing network text named entities based on neural network probability disambiguation.
BACKGROUND ART
Networks have driven the speed and scale of information collection and dissemination to an unprecedented level and brought global information sharing and interaction into reality, and have become an indispensable infrastructure in the information society. Modern communication and dissemination techniques have greatly improved the speed and breadth of information dissemination. However, there are accompanying problems and "side effects":
sometimes people are confused by the turbulent information, and it is very hard to obtain the precise information needed quickly and accurately from the vast sea of information. It is a prerequisite to analyze and obtain named entities, such as people, places, and organizations, etc., concerned by Internet users from within a mas of network text, in order to provide important support information for various higher-level applications such as online marketing, group emotion analysis, etc. Accordingly, network text named entity recognition has become an important core technique in network data processing and analysis.
Two kinds of methods for dealing with named entity recognition are considered in the research, i.e., rule-based method and statistics-based method. As the machine learning theory is consummated continuously and the computing performance is improved greatly, the statistics-based method is increasingly favored.
At present, statistical models and methods applied in named entity recognition mainly include:
hidden Markov model, decision tree, maximum entropy model, support vector machine, conditional random field and artificial neural network. Artificial neural networks can achieve a better result in named entity recognition than conditional random field, maximum entropy model, and other models, but conditional random field and maximum entropy models are still dominant practical models. For example, in the Patent Document No. CN201310182978.X, named entity recognition method and apparatus for MicroBlog text based on conditional random field and named entity library are proposed. In the Patent Document No.
CN200710098635.X, a named entity recognition method utilizing word features and using maximum entropy model to model is proposed. Artificial neural networks are difficult to use practically because they often require the conversion of words into vectors in a word vector space in the field of named entity recognition.
Consequently, artificial neural networks can not be applied in large-scale practical applications, because they are unable to obtain corresponding vectors for new words.
Owing to the above-mentioned present situation, there are mainly the following problems in named entity recognition for network text: firstly, it is unable to train a word vector space that contains all words in order to train a neural network, because there are a lot of network words, new words, and wrongly written or mispronounced characters in network text;
secondly, the accuracy of named entity recognition for network texts is degraded as a result of phenomena existing in network text, such as arbitrary language forms, non-standard grammatical structures, and wrongly written or mispronounced characters, etc.
SUMMARY OF THE INVENTION
The object of the invention is to overcome the drawbacks in the prior art. The present invention provides a network text named entity recognition method based on neural network probability disambiguation, which extracts word features incrementally without neural network retraining, and performs recognition with the aid of probability disambiguation. The method obtains a prediction probability matrix on the named entity category of a word from a neural network by training neural network, and performs disambiguation on the prediction matrix outputted from the neural network in a probability model, and thereby improves accuracy and precision of network text named entity recognition.
In order to attain the object described above, the technical scheme employed by the present invention is as follows.
The network text named entity recognition method is based on neural network probability disambiguation, performing word segmentation on an untagged corpus, utilizing Word2Vec to extract a word vector, converting sample corpora into a word feature matrix and windowing, building a deep neural network for training, adding a softmax function into an output layer of the neural network, and performing normalization, to acquire a probability matrix of named entity categorycorresponding to each word; re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
Specifically, the method comprises the following steps:
step 1: acquiring an untagged corpus by means of a web crawler, acquiring sample corpora with named entity tags from a corpus base, and performing word segmentation on the untagged corpus by a natural language tool;
step 2: performing word vector space training on the segmented untagged corpus and the sample corpora by a Word2Vec tool;
step 3: converting the text in the sample corpora into a word vector representing word features according to the trained Word2Vec model, windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to a neural network; converting the tags in the sample corpora into a one-hot
2 form and taking them as outputs of the neural network; performing normalization on an output layer of the neural network with a softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or a named entity, adjusting the structure, depth, number of nodes, step length, activation function, and initial value parameters in the neural network, and selecting an activation function to train the neural network;
step 4: re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in a conditional random field model, utilizing an EM
algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: in the recognition process, first, converting the text to be recognized into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, converting the word into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc., windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network;
then, re-windowing a prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining a final named entity tag of the text to be recognized.
Preferably, the parameters of the Word2Vec tool are as follows: length of word vector: 200, number of iterations: 25, initial step length: 0.025, minimum step length:
0.0001, and a CBOW
model is selected.
Preferably, the parameters of the neural network are as follows: number of hidden layers: 2, number of hidden nodes: 150, step length: 0.01, batchSize: 40, activation function: sigmoid function.
Preferably, the tags in the sample corpora are converted into an one-hot form with the following method: converting the tags "/o", "In", and "/p" in the sample corpora into named entity tags "/Org-B", "Org-I", "/Per-B", "/Per-I", "/Loc-B", and "/Loc-I" correspondingly, and then converting the named entity tags into the one-hot form.
Preferably, the window size for windowing the word vector is 5.
Preferably, in neural network training, one-tenth words are extracted from the sample data and excluded from the neural network training, but are used as evaluation criteria for the neural network.
Compared with the prior art, the present invention targets the following beneficial effects:
Word vectors without retraining the neural network may be extracted incrementally, prediction may be carried out with the neural network, and disambiguation may be performed with a
step 4: re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in a conditional random field model, utilizing an EM
algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: in the recognition process, first, converting the text to be recognized into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, converting the word into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc., windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network;
then, re-windowing a prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining a final named entity tag of the text to be recognized.
Preferably, the parameters of the Word2Vec tool are as follows: length of word vector: 200, number of iterations: 25, initial step length: 0.025, minimum step length:
0.0001, and a CBOW
model is selected.
Preferably, the parameters of the neural network are as follows: number of hidden layers: 2, number of hidden nodes: 150, step length: 0.01, batchSize: 40, activation function: sigmoid function.
Preferably, the tags in the sample corpora are converted into an one-hot form with the following method: converting the tags "/o", "In", and "/p" in the sample corpora into named entity tags "/Org-B", "Org-I", "/Per-B", "/Per-I", "/Loc-B", and "/Loc-I" correspondingly, and then converting the named entity tags into the one-hot form.
Preferably, the window size for windowing the word vector is 5.
Preferably, in neural network training, one-tenth words are extracted from the sample data and excluded from the neural network training, but are used as evaluation criteria for the neural network.
Compared with the prior art, the present invention targets the following beneficial effects:
Word vectors without retraining the neural network may be extracted incrementally, prediction may be carried out with the neural network, and disambiguation may be performed with a
3 probability model, so that the method achieves better practicability, accuracy and precision in named entity recognition of network text. In the task of named entity recognition of network text, the present invention provides an incremental word vector learning method without changing the structure of a neural network according to a characteristic that network words and new words exist, and employs a probability disambiguation method to deal with the problems that network texts are non-standard in grammatical structure and contain a lot of wrongly written or mispronounced characters. Thus, the method provided in the present invention attains high accuracy in network text named entity recognition tasks.
BRIEF DESCRIPTION OF DRAWINGS
Fig. 1 is a flow chart of training a network text named entity recognition device based on neural network probability disambiguation according to the present invention;
Fig. 2 is a flow chart of converting a word into word features according to the present invention;
Fig. 3 is a schematic diagram of the text processing and neural network architecture according to the present invention.
DETAILED DESCRIPTION OF THE EMBODIMENTS
Hereunder the present invention will be further detailed in embodiments, with reference to the accompanying drawings. It should be appreciated that those embodiments are provided only for describing the present invention, and shall not be deemed as constituting any limitation to the scope of the present invention. After reading the present invention, modifications to the present invention in various equivalent forms made by those skilled in the art shall be deemed as falling into the protected scope as defined by the attached claims in this application.
A network text named entity recognition method based on neural network probability disambiguation, performing word segmentation on an untagged corpus, utilizing Word2Vec to extract a word vector, converting sample corpora into a word feature matrix and windowing, building a deep neural network for training, adding a softmax function into an output layer of the neural network, and performing normalization, to acquire a probability matrix of named entity category corresponding to each word; re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
Specifically, the method comprises the following steps:
step I: Acquiring untagged network text by means of a web crawler, downloading corpora with named entity tags as sample corpora from a corpus base, and performing word segmentation on the untagged corpus with a natural language tool;
step 2: Performing word vector space training on the segmented untagged corpus and the sample corpora with a Word2Vec tool;
step 3: Converting the text in the sample corpora to a word vector that represents word features
BRIEF DESCRIPTION OF DRAWINGS
Fig. 1 is a flow chart of training a network text named entity recognition device based on neural network probability disambiguation according to the present invention;
Fig. 2 is a flow chart of converting a word into word features according to the present invention;
Fig. 3 is a schematic diagram of the text processing and neural network architecture according to the present invention.
DETAILED DESCRIPTION OF THE EMBODIMENTS
Hereunder the present invention will be further detailed in embodiments, with reference to the accompanying drawings. It should be appreciated that those embodiments are provided only for describing the present invention, and shall not be deemed as constituting any limitation to the scope of the present invention. After reading the present invention, modifications to the present invention in various equivalent forms made by those skilled in the art shall be deemed as falling into the protected scope as defined by the attached claims in this application.
A network text named entity recognition method based on neural network probability disambiguation, performing word segmentation on an untagged corpus, utilizing Word2Vec to extract a word vector, converting sample corpora into a word feature matrix and windowing, building a deep neural network for training, adding a softmax function into an output layer of the neural network, and performing normalization, to acquire a probability matrix of named entity category corresponding to each word; re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
Specifically, the method comprises the following steps:
step I: Acquiring untagged network text by means of a web crawler, downloading corpora with named entity tags as sample corpora from a corpus base, and performing word segmentation on the untagged corpus with a natural language tool;
step 2: Performing word vector space training on the segmented untagged corpus and the sample corpora with a Word2Vec tool;
step 3: Converting the text in the sample corpora to a word vector that represents word features
4 according to a trained Word2Vec model, and taking the word vector as an input to a neural network; converting the tags in the sample corpora into an one-hot form and taking them as outputs of the neural network. In view that a named entity may be divided into several words in a text processing task, the tagging is performed in an JOB pattern, in order to ensure that the recognized named entity has integrality.
Which named entity category a word belongs to should not be judged merely on the basis of the word itself, but should be further judged according to the context information of the word.
Therefore, a concept of "window" is introduced in the building of the neural network, i.e., in the judgment of a word, both the word and the characteristic information of content in fixed length thereof are taken as inputs to the neural network; thus, the input to the neural network is no longer the length d of a word feature vector, but is a two-dimensional matrix composed by multiplying the window w by the length d of word feature vector instead.
An output layer of the neural network is normalized with a softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or a named entity. The structure, depth, number of nodes, step length, activation function, initial value parameters in the neural network are adjusted, and an activation function is selected to train the neural network.
step 4: Re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in a conditional random field model, utilizing an EM
algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: In the recognition process, first, converting the text to be recognized into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, converting the word into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc.
(1) matching the word to be converted in a trained word vector space;
(2) converting the word to be converted directly to a corresponding word vector, if the word is matched in the word vector space;
(3) if the Word2Vec model doesn't contain a corresponding word, backing up the word vector space to prevent degradation of the accuracy of the neural network model incurred by deviation of a word space created in incremental learning, loading the Word2Vec model, acquiring a sentence where the mismatched word exists, inputting the sentence into the Word2Vec model and performing increment training, acquiring the word vector of the word, and utilizing the backup word vector space to performing backtracking of the model;
windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of word vector as an input to the neural network;
then, re-windowing a prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining a final named entity tag of the text to be recognized.
Example Network text is acquired by means of a web crawler from Sogou News website (http://news.sogou.com/), corpora with named entity tags are downloaded from Datatang corpus base (http://www.datatang.com/) as sample corpora, word segmentation is performed on the acquired network text with a natural language tool, word vector space training is performed on the segmented corpus and sample corpora with gensim package in python by Word2Vec model, utilizing the following parameters: length of word vector: 200, number of iterations: 25, initial step length: 0.025, and minimum step length: 0.0001, and a CBOW model is selected.
The text in the sample corpora is converted into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, the word is converted into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc., as the features of the word. The tags "/o", "In", and "/p" in the sample corpora acquired from Datatang are converted into named entity tags "/Org-B", "/Org-I", "/Per-B", "/Per-I", "/Loc-B", and "/Loc-I", etc.
correspondingly, and then the named entity tags are converted into the one-hot form as outputs of the neural network.
The window size is set to 5, i.e., in the consideration of the named entity category of the current word, the word features of the word and two words before the word and two words after the word are used as inputs to the neural network; the input to the neural network is a batchSize*1000 vector; one-tenth words are extracted from the sample data and excluded from the neural network training, but are used as evaluation criteria for the neural network; the output layer of the neural network is normalized with a softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or named entity; the maximum value of probability is taken as the final categorization result temporarily. The parameters in the neural network, such as structure, depth, number of nodes, step length, activation function, and initial value, etc., are adjusted to ensure the neural network attain high accuracy; the final parameters are as follows:
number of hidden layers: 2, number of hidden nodes: 150, step length: 0.01, batchSize: 40, activation function:
sigmoid; thus, a good categorization effect can be attained, the accuracy may be as high as 99.83%, and the F values of the most representative personal names, place names, and organization names may be 93.4%, 84.2%, and 80.4% respectively.
The step of taking the maximum probability value of the prediction matrix outputted from the neural network as the final categorization result is removed, the probability matrix is re-windowed directly, the context prediction information of the word to be tagged is used as a point of correlation with the actual category of the word to be tagged in a conditional random field model, an EM algorithm is used to calculate expected values at all sides of the conditional random field according to the training corpora, and a corresponding conditional random field model is trained; after disambiguation with the conditional random field, the F values of personal names, place names, and organization names can be improved to 94.8%, 85.0%, and 82.0%
respectively.
It is seen from the embodiment described above: compared with the conventional supervised named entity recognition method, the text named entity recognition method based on neural network probability disambiguation provided in the present invention employs a word vector conversion method that can be used to extract word features incrementally without causing deviation of the word vector space; thus, the neural network can be applied to network text that contains a lot of new words and wrongly written or mispronounced characters.
Moreover, in the present invention, the probability matrix outputted from the neural network is re-windowed, and context disambiguation is performed with a conditional random field model, so as to deal with the phenomenon that the network text involves a lot of wrongly written or mispronounced characters and non-standard grammatical structures successfully.
While the present invention is described above in some preferred embodiments, it should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and those improvements and modifications should be deemed as falling in the scope of protection of the present invention.
Which named entity category a word belongs to should not be judged merely on the basis of the word itself, but should be further judged according to the context information of the word.
Therefore, a concept of "window" is introduced in the building of the neural network, i.e., in the judgment of a word, both the word and the characteristic information of content in fixed length thereof are taken as inputs to the neural network; thus, the input to the neural network is no longer the length d of a word feature vector, but is a two-dimensional matrix composed by multiplying the window w by the length d of word feature vector instead.
An output layer of the neural network is normalized with a softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or a named entity. The structure, depth, number of nodes, step length, activation function, initial value parameters in the neural network are adjusted, and an activation function is selected to train the neural network.
step 4: Re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in a conditional random field model, utilizing an EM
algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: In the recognition process, first, converting the text to be recognized into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, converting the word into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc.
(1) matching the word to be converted in a trained word vector space;
(2) converting the word to be converted directly to a corresponding word vector, if the word is matched in the word vector space;
(3) if the Word2Vec model doesn't contain a corresponding word, backing up the word vector space to prevent degradation of the accuracy of the neural network model incurred by deviation of a word space created in incremental learning, loading the Word2Vec model, acquiring a sentence where the mismatched word exists, inputting the sentence into the Word2Vec model and performing increment training, acquiring the word vector of the word, and utilizing the backup word vector space to performing backtracking of the model;
windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of word vector as an input to the neural network;
then, re-windowing a prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining a final named entity tag of the text to be recognized.
Example Network text is acquired by means of a web crawler from Sogou News website (http://news.sogou.com/), corpora with named entity tags are downloaded from Datatang corpus base (http://www.datatang.com/) as sample corpora, word segmentation is performed on the acquired network text with a natural language tool, word vector space training is performed on the segmented corpus and sample corpora with gensim package in python by Word2Vec model, utilizing the following parameters: length of word vector: 200, number of iterations: 25, initial step length: 0.025, and minimum step length: 0.0001, and a CBOW model is selected.
The text in the sample corpora is converted into a word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding training word, the word is converted into a word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, etc., as the features of the word. The tags "/o", "In", and "/p" in the sample corpora acquired from Datatang are converted into named entity tags "/Org-B", "/Org-I", "/Per-B", "/Per-I", "/Loc-B", and "/Loc-I", etc.
correspondingly, and then the named entity tags are converted into the one-hot form as outputs of the neural network.
The window size is set to 5, i.e., in the consideration of the named entity category of the current word, the word features of the word and two words before the word and two words after the word are used as inputs to the neural network; the input to the neural network is a batchSize*1000 vector; one-tenth words are extracted from the sample data and excluded from the neural network training, but are used as evaluation criteria for the neural network; the output layer of the neural network is normalized with a softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or named entity; the maximum value of probability is taken as the final categorization result temporarily. The parameters in the neural network, such as structure, depth, number of nodes, step length, activation function, and initial value, etc., are adjusted to ensure the neural network attain high accuracy; the final parameters are as follows:
number of hidden layers: 2, number of hidden nodes: 150, step length: 0.01, batchSize: 40, activation function:
sigmoid; thus, a good categorization effect can be attained, the accuracy may be as high as 99.83%, and the F values of the most representative personal names, place names, and organization names may be 93.4%, 84.2%, and 80.4% respectively.
The step of taking the maximum probability value of the prediction matrix outputted from the neural network as the final categorization result is removed, the probability matrix is re-windowed directly, the context prediction information of the word to be tagged is used as a point of correlation with the actual category of the word to be tagged in a conditional random field model, an EM algorithm is used to calculate expected values at all sides of the conditional random field according to the training corpora, and a corresponding conditional random field model is trained; after disambiguation with the conditional random field, the F values of personal names, place names, and organization names can be improved to 94.8%, 85.0%, and 82.0%
respectively.
It is seen from the embodiment described above: compared with the conventional supervised named entity recognition method, the text named entity recognition method based on neural network probability disambiguation provided in the present invention employs a word vector conversion method that can be used to extract word features incrementally without causing deviation of the word vector space; thus, the neural network can be applied to network text that contains a lot of new words and wrongly written or mispronounced characters.
Moreover, in the present invention, the probability matrix outputted from the neural network is re-windowed, and context disambiguation is performed with a conditional random field model, so as to deal with the phenomenon that the network text involves a lot of wrongly written or mispronounced characters and non-standard grammatical structures successfully.
While the present invention is described above in some preferred embodiments, it should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and those improvements and modifications should be deemed as falling in the scope of protection of the present invention.
Claims (7)
1. A method for recognizing network text named entity based on neural network probability disambiguation, comprising: performing word segmentation on an untagged corpus, utilizing Word2Vec to extract a word vector, converting sample corpora into a word feature matrix, windowing, building a deep neural network for training, adding a softmax function into an output layer of the neural network, and perfonning normalization, to acquire a probability matrix of named entity category corresponding to each word;
re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
re-windowing the probability matrix, and utilizing a conditional random field model for disambiguation to obtain a final named entity tag.
2. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, comprising the following steps:
step 1: acquiring the untagged corpus by means of a web crawler, acquiring sample corpora with named entity tags from a corpus base, and perfonning word segmentation on the untagged corpus with a natural language tool;
step 2: performing word vector space training on the segmented untagged corpus and the sample corpora by the Word2Vec tool;
step 3: converting the text in the sample corpora into the word vector representing word features according to the trained Word2Vec model, windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network; converting the tags in the sample corpora into a one-hot fonn and taking them as outputs of the neural network; performing nonnalization on an output layer of the neural network with the softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or a named entity, adjusting the structure, depth, number of nodes, step length, activation function, and initial value parameters in the neural network, and selecting an activation function to train the neural network;
step 4: re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in the conditional random field model, utilizing an expectation-maximization (EM) algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: in the recognition process, first, converting the text to be recognized into the word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding word, converting the word into the word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, windowing the word vector, and Date Recue/Date Received 2020-09-10 taking the two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network; then, re-windowing the prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining the final named entity tag of the text to be recognized.
step 1: acquiring the untagged corpus by means of a web crawler, acquiring sample corpora with named entity tags from a corpus base, and perfonning word segmentation on the untagged corpus with a natural language tool;
step 2: performing word vector space training on the segmented untagged corpus and the sample corpora by the Word2Vec tool;
step 3: converting the text in the sample corpora into the word vector representing word features according to the trained Word2Vec model, windowing the word vector, and taking a two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network; converting the tags in the sample corpora into a one-hot fonn and taking them as outputs of the neural network; performing nonnalization on an output layer of the neural network with the softmax function, so that a categorization result produced by the neural network becomes a probability of whether the word belongs to an unnamed entity or a named entity, adjusting the structure, depth, number of nodes, step length, activation function, and initial value parameters in the neural network, and selecting an activation function to train the neural network;
step 4: re-windowing a prediction matrix outputted from the neural network, taking context prediction information of the word to be tagged as a point of correlation with an actual category of the word to be tagged in the conditional random field model, utilizing an expectation-maximization (EM) algorithm to calculate expected values at all sides according to training corpora, and training a corresponding conditional random field model;
step 5: in the recognition process, first, converting the text to be recognized into the word vector that represents word features according to the trained Word2Vec model, and, if the Word2Vec model doesn't contain a corresponding word, converting the word into the word vector by means of incremental learning, word vector acquisition, and word vector space backtracking, windowing the word vector, and Date Recue/Date Received 2020-09-10 taking the two-dimensional matrix composed by multiplying the window w by the length d of the word vector as an input to the neural network; then, re-windowing the prediction matrix obtained from the neural network, performing disambiguation on the prediction matrix in the trained conditional random field model, and obtaining the final named entity tag of the text to be recognized.
3. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, wherein, the parameters of the Word2Vec tool are as follows: length of word vector: 200, number of iterations: 25, initial step length: 0.025, minimum step length: 0.0001, and a continuous bag-of-words (CBOW) model is selected.
4. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, wherein, the parameters of the neural network are as follows: number of hidden layers: 2, number of hidden nodes:
150, step length: 0.01, batch size: 40, activation function: sigmoid function.
150, step length: 0.01, batch size: 40, activation function: sigmoid function.
5. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, wherein, the tags in the sample corpora are converted into a one-hot form with the following method: converting the tags "/n", and "/p" in the sample corpora into named entity tags "/Org-B", "/Org-I", "/Per-B", "/Per-I", "/Loc-B", and "/Loc-I" correspondingly, and then converting the named entity tags into the one-hot form.
6. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, wherein, the window size for windowing the word vector is 5.
7. The method for recognizing network text named entity based on neural network probability disambiguation according to claim 1, wherein, in neural network training, one-tenth of the words are extracted from the sample data and excluded from the neural network training, but are used as evaluation criteria for the neural network.
Date Recue/Date Received 2020-09-10
Date Recue/Date Received 2020-09-10
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710390409.2 | 2017-05-27 | ||
| CN201710390409.2A CN107203511B (en) | 2017-05-27 | 2017-05-27 | Network text named entity identification method based on neural network probability disambiguation |
| PCT/CN2017/089135 WO2018218705A1 (en) | 2017-05-27 | 2017-06-20 | Method for recognizing network text named entity based on neural network probability disambiguation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA3039280A1 CA3039280A1 (en) | 2018-12-06 |
| CA3039280C true CA3039280C (en) | 2021-07-20 |
Family
ID=59905476
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA3039280A Active CA3039280C (en) | 2017-05-27 | 2017-06-20 | Method for recognizing network text named entity based on neural network probability disambiguation |
Country Status (5)
| Country | Link |
|---|---|
| CN (1) | CN107203511B (en) |
| AU (1) | AU2017416649A1 (en) |
| CA (1) | CA3039280C (en) |
| RU (1) | RU2722571C1 (en) |
| WO (1) | WO2018218705A1 (en) |
Families Citing this family (67)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107203511B (en) * | 2017-05-27 | 2020-07-17 | 中国矿业大学 | Network text named entity identification method based on neural network probability disambiguation |
| CN107665252B (en) * | 2017-09-27 | 2020-08-25 | 深圳证券信息有限公司 | Method and device for creating knowledge graph |
| CN107832289A (en) * | 2017-10-12 | 2018-03-23 | 北京知道未来信息技术有限公司 | A kind of name entity recognition method based on LSTM CNN |
| CN107908614A (en) * | 2017-10-12 | 2018-04-13 | 北京知道未来信息技术有限公司 | A kind of name entity recognition method based on Bi LSTM |
| CN107885721A (en) * | 2017-10-12 | 2018-04-06 | 北京知道未来信息技术有限公司 | A kind of name entity recognition method based on LSTM |
| CN107967251A (en) * | 2017-10-12 | 2018-04-27 | 北京知道未来信息技术有限公司 | A kind of name entity recognition method based on Bi-LSTM-CNN |
| CN107797989A (en) * | 2017-10-16 | 2018-03-13 | 平安科技（深圳）有限公司 | Enterprise name recognition methods, electronic equipment and computer-readable recording medium |
| CN107943788B (en) * | 2017-11-17 | 2021-04-06 | 平安科技（深圳）有限公司 | Enterprise abbreviation generation method and device and storage medium |
| CN110019648B (en) * | 2017-12-05 | 2021-02-02 | 深圳市腾讯计算机系统有限公司 | Method and device for training data and storage medium |
| CN108052504B (en) * | 2017-12-26 | 2020-11-20 | 浙江讯飞智能科技有限公司 | Structure analysis method and system for mathematic subjective question answer result |
| CN108121702B (en) * | 2017-12-26 | 2020-11-24 | 浙江讯飞智能科技有限公司 | Method and system for evaluating and reading mathematical subjective questions |
| CN108280062A (en) * | 2018-01-19 | 2018-07-13 | 北京邮电大学 | Entity based on deep learning and entity-relationship recognition method and device |
| CN108563626B (en) * | 2018-01-22 | 2022-01-25 | 北京颐圣智能科技有限公司 | Medical text named entity recognition method and device |
| CN108388559B (en) * | 2018-02-26 | 2021-11-19 | 中译语通科技股份有限公司 | Named entity identification method and system under geographic space application and computer program |
| CN108763192B (en) * | 2018-04-18 | 2022-04-19 | 达而观信息科技（上海）有限公司 | Entity relation extraction method and device for text processing |
| CN108805196B (en) * | 2018-06-05 | 2022-02-18 | 西安交通大学 | Automatic incremental learning method for image recognition |
| RU2699687C1 (en) * | 2018-06-18 | 2019-09-09 | Общество с ограниченной ответственностью "Аби Продакшн" | Detecting text fields using neural networks |
| CN109062983A (en) * | 2018-07-02 | 2018-12-21 | 北京妙医佳信息技术有限公司 | Name entity recognition method and system for medical health knowledge mapping |
| CN109241520B (en) * | 2018-07-18 | 2023-05-23 | 五邑大学 | Sentence trunk analysis method and system based on multi-layer error feedback neural network for word segmentation and named entity recognition |
| CN109255119B (en) * | 2018-07-18 | 2023-04-25 | 五邑大学 | Sentence trunk analysis method and system of multi-task deep neural network based on word segmentation and named entity recognition |
| CN109299458B (en) * | 2018-09-12 | 2023-03-28 | 广州多益网络股份有限公司 | Entity identification method, device, equipment and storage medium |
| CN109446514B (en) * | 2018-09-18 | 2024-08-20 | 平安科技（深圳）有限公司 | News entity identification model construction method and device and computer equipment |
| CN109657238B (en) * | 2018-12-10 | 2023-10-13 | 宁波深擎信息科技有限公司 | Knowledge graph-based context identification completion method, system, terminal and medium |
| CN109710927B (en) * | 2018-12-12 | 2022-12-20 | 东软集团股份有限公司 | Named entity identification method and device, readable storage medium and electronic equipment |
| CN109670177A (en) * | 2018-12-20 | 2019-04-23 | 翼健(上海)信息科技有限公司 | One kind realizing the semantic normalized control method of medicine and control device based on LSTM |
| CN109858025B (en) * | 2019-01-07 | 2023-06-13 | 鼎富智能科技有限公司 | Word segmentation method and system for address standardized corpus |
| CN109767817B (en) * | 2019-01-16 | 2023-05-30 | 南通大学 | Drug potential adverse reaction discovery method based on neural network language model |
| CN111563380A (en) * | 2019-01-25 | 2020-08-21 | 浙江大学 | Named entity identification method and device |
| CN109800437B (en) * | 2019-01-31 | 2023-11-14 | 北京工业大学 | Named entity recognition method based on feature fusion |
| CN109992629B (en) * | 2019-02-28 | 2021-08-06 | 中国科学院计算技术研究所 | Neural network relation extraction method and system fusing entity type constraints |
| CN109858041B (en) * | 2019-03-07 | 2023-02-17 | 北京百分点科技集团股份有限公司 | Named entity recognition method combining semi-supervised learning with user-defined dictionary |
| CN109933801B (en) * | 2019-03-25 | 2022-03-29 | 北京理工大学 | Bidirectional LSTM named entity identification method based on predicted position attention |
| CN111858838A (en) * | 2019-04-04 | 2020-10-30 | 拉扎斯网络科技（上海）有限公司 | Menu calibration method and device, electronic equipment and nonvolatile storage medium |
| CN110083778A (en) * | 2019-04-08 | 2019-08-02 | 清华大学 | The figure convolutional neural networks construction method and device of study separation characterization |
| CN110334110A (en) * | 2019-05-28 | 2019-10-15 | 平安科技（深圳）有限公司 | Natural language classification method, device, computer equipment and storage medium |
| CN110245242B (en) * | 2019-06-20 | 2022-01-18 | 北京百度网讯科技有限公司 | Medical knowledge graph construction method and device and terminal |
| CN110298043B (en) * | 2019-07-03 | 2023-04-07 | 吉林大学 | Vehicle named entity identification method and system |
| CN110750992B (en) * | 2019-10-09 | 2023-07-04 | 吉林大学 | Named entity recognition method, named entity recognition device, electronic equipment and named entity recognition medium |
| CN110781646B (en) * | 2019-10-15 | 2023-08-22 | 泰康保险集团股份有限公司 | Name standardization method, device, medium and electronic equipment |
| CN111008271B (en) * | 2019-11-20 | 2022-06-24 | 佰聆数据股份有限公司 | Neural network-based key information extraction method and system |
| CN110993081B (en) * | 2019-12-03 | 2023-08-11 | 济南大学 | Doctor online recommendation method and system |
| CN111091003B (en) * | 2019-12-05 | 2023-10-10 | 电子科技大学广东电子信息工程研究院 | Parallel extraction method based on knowledge graph query |
| CN111209748B (en) * | 2019-12-16 | 2023-10-24 | 合肥讯飞数码科技有限公司 | Error word recognition method, related device and readable storage medium |
| CN113139382A (en) * | 2020-01-20 | 2021-07-20 | 北京国双科技有限公司 | Named entity identification method and device |
| CN111368545B (en) * | 2020-02-28 | 2024-04-30 | 北京明略软件系统有限公司 | Named entity recognition method and device based on multitask learning |
| CN111477320B (en) * | 2020-03-11 | 2023-05-30 | 北京大学第三医院（北京大学第三临床医学院） | Treatment effect prediction model construction system, treatment effect prediction system and terminal |
| CN111523323B (en) * | 2020-04-26 | 2022-08-12 | 梁华智能科技(上海)有限公司 | Disambiguation processing method and system for Chinese word segmentation |
| CN111581957B (en) * | 2020-05-06 | 2022-04-12 | 浙江大学 | Nested entity detection method based on pyramid hierarchical network |
| CN111476022B (en) * | 2020-05-15 | 2023-07-07 | 湖南工商大学 | Character embedding and mixed LSTM entity identification method, system and medium for entity characteristics |
| CN111859937B (en) * | 2020-07-20 | 2024-07-30 | 上海汽车集团股份有限公司 | Entity identification method and device |
| CN112199953B (en) * | 2020-08-24 | 2024-06-28 | 广州九四智能科技有限公司 | Method and device for extracting information in telephone call and computer equipment |
| RU2760637C1 (en) * | 2020-08-31 | 2021-11-29 | Публичное Акционерное Общество "Сбербанк России" (Пао Сбербанк) | Method and system for retrieving named entities |
| CN112101041B (en) * | 2020-09-08 | 2022-02-15 | 平安科技（深圳）有限公司 | Entity relationship extraction method, device, equipment and medium based on semantic similarity |
| CN112765983A (en) * | 2020-12-14 | 2021-05-07 | 四川长虹电器股份有限公司 | Entity disambiguation method based on neural network combined with knowledge description |
| CN112487816B (en) * | 2020-12-14 | 2024-02-13 | 安徽大学 | Named entity identification method based on network classification |
| CN112905742B (en) * | 2021-02-20 | 2022-07-29 | 厦门吉比特网络技术股份有限公司 | Method and device for recognizing new vocabulary based on semantic model neural network |
| CN113343690B (en) * | 2021-06-22 | 2024-03-12 | 北京语言大学 | Text readability automatic evaluation method and device |
| CN114218924A (en) * | 2021-07-27 | 2022-03-22 | 广东电力信息科技有限公司 | Text intention and entity combined identification method based on BERT model |
| CN114519355A (en) * | 2021-08-25 | 2022-05-20 | 浙江万里学院 | Medicine named entity recognition and entity standardization method |
| CN113849597B (en) * | 2021-08-31 | 2024-04-30 | 艾迪恩(山东)科技有限公司 | Illegal advertisement word detection method based on named entity recognition |
| CN113934815A (en) * | 2021-09-18 | 2022-01-14 | 有米科技股份有限公司 | Advertisement and pattern characteristic information identification method and device based on neural network |
| CN114036948B (en) * | 2021-10-26 | 2024-05-31 | 天津大学 | Named entity identification method based on uncertainty quantification |
| CN114048749B (en) * | 2021-11-19 | 2024-02-02 | 北京第一因科技有限公司 | Chinese named entity recognition method suitable for multiple fields |
| CN114510943B (en) * | 2022-02-18 | 2024-05-28 | 北京大学 | Incremental named entity recognition method based on pseudo sample replay |
| WO2023204724A1 (en) * | 2022-04-20 | 2023-10-26 | Общество С Ограниченной Ответственностью "Дентонс Юроп" (Ооо "Дентонс Юроп") | Method for analyzing a legal document |
| CN115587594B (en) * | 2022-09-20 | 2023-06-30 | 广东财经大学 | Unstructured text data extraction model training method and system for network security |
| CN115905456B (en) * | 2023-01-06 | 2023-06-02 | 浪潮电子信息产业股份有限公司 | Data identification method, system, equipment and computer readable storage medium |
Family Cites Families (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7502971B2 (en) * | 2005-10-12 | 2009-03-10 | Hewlett-Packard Development Company, L.P. | Determining a recurrent problem of a computer resource using signatures |
| US8583416B2 (en) * | 2007-12-27 | 2013-11-12 | Fluential, Llc | Robust information extraction from utterances |
| RU2399959C2 (en) * | 2008-10-29 | 2010-09-20 | Закрытое акционерное общество "Авикомп Сервисез" | Method for automatic text processing in natural language through semantic indexation, method for automatic processing collection of texts in natural language through semantic indexation and computer readable media |
| US8239349B2 (en) * | 2010-10-07 | 2012-08-07 | Hewlett-Packard Development Company, L.P. | Extracting data |
| CN103455581B (en) * | 2013-08-26 | 2016-05-04 | 北京理工大学 | This information filtering method of Massive short documents based on semantic extension |
| CN105404632B (en) * | 2014-09-15 | 2020-07-31 | 深港产学研基地 | System and method for carrying out serialized annotation on biomedical text based on deep neural network |
| CN104809176B (en) * | 2015-04-13 | 2018-08-07 | 中央民族大学 | Tibetan language entity relation extraction method |
| CN105740349B (en) * | 2016-01-25 | 2019-03-08 | 重庆邮电大学 | A kind of sensibility classification method of combination Doc2vec and convolutional neural networks |
| CN105868184B (en) * | 2016-05-10 | 2018-06-08 | 大连理工大学 | A kind of Chinese personal name recognition method based on Recognition with Recurrent Neural Network |
| CN106202032B (en) * | 2016-06-24 | 2018-08-28 | 广州数说故事信息科技有限公司 | A kind of sentiment analysis method and its system towards microblogging short text |
| CN106202044A (en) * | 2016-07-07 | 2016-12-07 | 武汉理工大学 | A kind of entity relation extraction method based on deep neural network |
| CN107203511B (en) * | 2017-05-27 | 2020-07-17 | 中国矿业大学 | Network text named entity identification method based on neural network probability disambiguation |
-
2017
- 2017-05-27 CN CN201710390409.2A patent/CN107203511B/en active Active
- 2017-06-20 AU AU2017416649A patent/AU2017416649A1/en not_active Abandoned
- 2017-06-20 CA CA3039280A patent/CA3039280C/en active Active
- 2017-06-20 RU RU2019117529A patent/RU2722571C1/en active
- 2017-06-20 WO PCT/CN2017/089135 patent/WO2018218705A1/en active Application Filing
Also Published As
| Publication number | Publication date |
|---|---|
| WO2018218705A1 (en) | 2018-12-06 |
| CA3039280A1 (en) | 2018-12-06 |
| CN107203511B (en) | 2020-07-17 |
| CN107203511A (en) | 2017-09-26 |
| RU2722571C1 (en) | 2020-06-01 |
| AU2017416649A1 (en) | 2019-05-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA3039280C (en) | Method for recognizing network text named entity based on neural network probability disambiguation | |
| CN109493977B (en) | Text data processing method and device, electronic equipment and computer readable medium | |
| CN110472003B (en) | Social network text emotion fine-grained classification method based on graph convolution network | |
| Dashtipour et al. | Exploiting deep learning for Persian sentiment analysis | |
| CN110796160A (en) | Text classification method, device and storage medium | |
| CN111159405B (en) | Irony detection method based on background knowledge | |
| CN115309915B (en) | Knowledge graph construction method, device, equipment and storage medium | |
| CN108763192B (en) | Entity relation extraction method and device for text processing | |
| CN112528654A (en) | Natural language processing method and device and electronic equipment | |
| CN114519356A (en) | Target word detection method and device, electronic equipment and storage medium | |
| CN117764084A (en) | Short text emotion analysis method based on multi-head attention mechanism and multi-model fusion | |
| CN114936274A (en) | Model training method, dialogue generating device, dialogue training equipment and storage medium | |
| Mercan et al. | Abstractive text summarization for resumes with cutting edge NLP transformers and LSTM | |
| CN111241273A (en) | Text data classification method and device, electronic equipment and computer readable medium | |
| CN110309355A (en) | Generation method, device, equipment and the storage medium of content tab | |
| Rajani Shree et al. | POS tagger model for Kannada text with CRF++ and deep learning approaches | |
| CN112818124A (en) | Entity relationship extraction method based on attention neural network | |
| Li et al. | A recurrent neural network language model based on word embedding | |
| CN115796141A (en) | Text data enhancement method and device, electronic equipment and storage medium | |
| Hung | College admissions counseling using intelligent question answering system | |
| CN113886530A (en) | Semantic phrase extraction method and related device | |
| Garrido et al. | Information extraction on weather forecasts with semantic technologies | |
| CN113704472A (en) | Hate and offensive statement identification method and system based on topic memory network | |
| Meng et al. | Design of Intelligent Recognition Model for English Translation Based on Deep Machine Learning | |
| Prajapati et al. | Empirical Analysis of Humor Detection Using Deep Learning and Machine Learning on Kaggle Corpus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request |
Effective date: 20190403 |