CN107102989B - Entity disambiguation method based on word vector and convolutional neural network - Google Patents

Entity disambiguation method based on word vector and convolutional neural network Download PDF

Info

Publication number
CN107102989B
CN107102989B CN201710373502.2A CN201710373502A CN107102989B CN 107102989 B CN107102989 B CN 107102989B CN 201710373502 A CN201710373502 A CN 201710373502A CN 107102989 B CN107102989 B CN 107102989B
Authority
CN
China
Prior art keywords
entity
disambiguated
word
neural network
word vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710373502.2A
Other languages
Chinese (zh)
Other versions
CN107102989A (en
Inventor
张雷
高扬
唐驰
谢俊元
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University
Original Assignee
Nanjing University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University filed Critical Nanjing University
Priority to CN201710373502.2A priority Critical patent/CN107102989B/en
Publication of CN107102989A publication Critical patent/CN107102989A/en
Application granted granted Critical
Publication of CN107102989B publication Critical patent/CN107102989B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/237—Lexical tools
    • G06F40/247—Thesauruses; Synonyms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • 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)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Machine Translation (AREA)

Abstract

The invention provides an entity disambiguation method based on word vectors and a convolutional neural network. The method constructs semantic feature vectors respectively aiming at the context of an entity to be disambiguated and candidate entity abstract information in a knowledge base by depending on word vectors and a convolutional neural network trained by word2 vec. And calculating the cosine similarity of the characteristic vectors in the entity classification stage, and taking the candidate entity with the maximum similarity as the final target entity of the entity to be disambiguated. By the method, the semantic representation capability of the entity is greatly improved, and the accuracy of subsequent disambiguation is further improved.

Description

Entity disambiguation method based on word vector and convolutional neural network
Technical Field
The invention belongs to the technical field of internet information, and particularly relates to an entity disambiguation method, in particular to an entity disambiguation method based on word vectors and a convolutional neural network.
Background
With the popularization of mobile internet, microblogs, blogs, posts, forums, various news websites, government work websites and the like greatly facilitate the life of people. Most of the data on the platforms exist in an unstructured or semi-structured form, so that a large amount of ambiguity exists in the data. If these ambiguous entities can be accurately disambiguated, great convenience is brought to later utilization.
Most of the existing mainstream entity disambiguation algorithm bottom models are based on bag-of-words models, and the bag-of-words models have inherent limitations, so that the algorithms cannot fully utilize semantic information of contexts, and the entity disambiguation effect has a great improvement space. Word embedding is a hot spot of machine learning in recent years, and the core idea of word embedding is to construct a distributed representation for each word, so that a gap between words is avoided. The convolutional neural network is a branch of a neural network model, and can effectively capture local features and then carry out global modeling. If the convolutional neural network can be used for modeling word embedding, semantic features more effective than the bag-of-words model can be obtained. And based on the thought of local perception and weight sharing, the parameters in the convolutional neural network model are greatly reduced, the training speed is high, and the core of Alphago of Google is two convolutional neural networks.
The invention combines the word vector and the convolutional neural network, constructs semantic representation respectively aiming at the context of the entity to be disambiguated and the entity abstract information of the knowledge base, trains the convolutional neural network and predicts. The semantic description capability of the entity context is greatly improved.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides an entity disambiguation method based on word vectors and a convolutional neural network aiming at the current situation that the existing entity disambiguation method is difficult to utilize context semantic information, and aims to capture the context semantic information to help entity disambiguation.
The technical scheme is as follows:
an entity disambiguation method based on word vector and convolutional neural network comprises the following steps:
step 1: according to a text set which is collected by an application scene and contains entities to be disambiguated, preprocessing the text set, and determining each entity to be disambiguated and context characteristics thereof in the text set;
step 2: constructing a knowledge base of the entities to be disambiguated according to the domain knowledge, searching the knowledge base, and determining a candidate entity set of each entity to be disambiguated and description characteristics of each candidate entity in the set;
and step 3: taking word vectors of nouns in a fixed-size window taking an entity to be disambiguated as a center to form a word vector matrix as a context semantic feature of the entity to be disambiguated; taking word vectors of the first 20 nouns with larger weights after calculating TF & IDF of the abstract information of each entity in the knowledge base to form a word vector matrix as semantic features of the knowledge base entities;
and 4, step 4: combining known unambiguous entities in the text with target entities and candidate entities in a knowledge base to form a training set, inputting the training set into a convolutional neural network model for training, and adjusting parameters in the model;
and 5: inputting a sample consisting of each entity to be disambiguated and the knowledge base candidate entity set into the convolutional neural network model obtained in the step (4) to respectively obtain semantic feature vectors of each knowledge base entity in the entity to be disambiguated and the knowledge base candidate entity set;
step 6: calculating the cosine similarity of the entity to be disambiguated and each entity in the knowledge base candidate entity set based on the semantic feature vector; and taking the candidate entity with the maximum similarity as the final target entity of the entity to be disambiguated.
The preprocessing in the step 1 is to use Chinese word segmentation program ICTCCLAS of Chinese academy of sciences to label and segment words of the text set, then filter out stop words according to a stop word list, and create a name dictionary for proper nouns and entity names which are difficult to recognize.
And in the step 2, a Chinese word segmentation program ICTCCLAS is called to perform part-of-speech tagging and word segmentation on entity descriptions in the knowledge base, and stop words are filtered according to the stop word list.
The step 3 of forming a word vector matrix by word vectors of nouns in a fixed-size window with the entity to be disambiguated as the center specifically comprises the following steps:
1) calling a Google deep learning program word2vec to train a Wikipedia corpus so as to obtain a word vector table L, wherein the length of a word vector is 200 dimensions, and each dimension is a real number;
2) context to be disambiguated entity ee＝{w1,w2,…,wi,…,wKEach noun w iniInquiring the word vector table L to obtain the word vector v of each nouni；
3) Constructing a context word vector matrix [ v ] of the entity e to be disambiguated according to the word vector of the context word of the entity e to be disambiguated1,v2,v3,…vi,…,vK]；
4) And (6) ending.
The step 3 of taking the word vectors of the first 20 nouns with larger weights after calculating the TF · IDF of the summary information of each entity in the knowledge base to form the word vector matrix specifically comprises:
1) for candidate entity set E ═ E1,e2,…,enEach candidate entity e iniEach noun w in the description of (1)iInquiring the word vector table L to obtain the word vector v of each nouni；
2) Constructing a word vector matrix of entity description according to the word vector of each noun in the description characteristics;
3) and (6) ending.
The convolutional neural network learning training of the step 4 specifically comprises the following steps:
1) each semantic feature to be disambiguated and the semantic features of the candidate entity set are used as a training sample and input into the neural network model;
2) convolving semantic features to be disambiguated, setting the number of convolution kernel feature maps to be 200, and setting the size of the convolution kernel feature maps to be [2, 200], namely setting a matrix with the length of 2 and the width of 200;
3) pooling the convolution result of each convolution kernel by using 1-max to obtain the characteristic of each convolution kernel;
4)200 convolution kernel features form an intermediate result, the intermediate result is input into a full connection layer, the size of the full connection layer is 50, and a 50-dimensional semantic feature vector is finally obtained;
5) the semantic features of the candidate entity set are added and averaged, then the sum and the average are input into a full connection layer, the size of the full connection layer is also 50, and finally a 50-dimensional semantic feature vector is obtained;
6) loss function Loss of each training sample in neural networkeIs defined as:
Losse＝max(0,1-sim(e,e)+sim(e,e′))
wherein: e.g. of the typeRepresenting a target entity of an entity e to be disambiguated, and e' representing any other candidate entity in the candidate entity set, which means that the difference between the semantic feature vector similarity of the target entity and any other candidate entity is maximized;
the global Loss function is defined as Loss ∑ Losse；
7) Parameters in the neural network are initialized by uniformly distributed U (-0.01, 0.01);
8) the activation functions in the neural network all adopt tanh hyperbolic tangent activation functions;
9) parameters in the neural network are updated by adopting random gradient descent;
10) and (6) ending.
The step 6 of entity classification specifically comprises the following steps:
1) reading a semantic feature vector a of an entity e to be disambiguated from a file system;
2) reading a candidate entity set E ═ E from a file system1,e2,…,enSet of semantic feature vectors in { B ═ B } ═ B1,b2,…,bn}；
3) Traversing the candidate entity set, and calculating the cosine similarity of each feature vector in E and E
Figure BDA0001303391840000043
Figure BDA0001303391840000041
4) Selecting the entity with the largest similarity
Figure BDA0001303391840000042
As a final prediction result;
5) and (6) ending.
Has the advantages that: the entity disambiguation method based on the word vector and the convolutional neural network constructs semantic representations for the entity to be disambiguated and the candidate entity of the knowledge base respectively. And training the neural network model by utilizing the training set, inputting the entity to be disambiguated into the trained neural network model when the entity is disambiguated, and outputting the most similar candidate entity of the entity to be disambiguated as a final target entity.
Drawings
For a more clear description of the invention, reference will now be made to the accompanying drawings, which form a part hereof and in which:
FIG. 1 is a block diagram of an entity disambiguation method based on word vector convolutional neural networks of the present invention.
Fig. 2 is a block diagram of a convolutional neural network model.
FIG. 3 is a flow chart of an entity classification phase.
Detailed Description
The present invention will be further described with reference to the accompanying drawings.
The flow chart of the entity disambiguation method based on the word vector and the convolutional neural network is shown in FIG. 1.
Step 0 is the initial state of the entity disambiguation method of the invention;
in the entity identification phase (steps 1-6):
step 1, collecting a text set containing an entity to be disambiguated according to an application scene;
step 2, constructing a knowledge base of the entity to be disambiguated according to the domain knowledge;
step 3, calling a Chinese word segmentation program ICTCCLAS of the Chinese academy of sciences to perform part-of-speech tagging and word segmentation on the text set, then filtering out stop words according to a stop word list, and creating a noun dictionary for some proper nouns and entity names which are difficult to recognize;
step 4, calling a Chinese word segmentation program ICTCCLAS to perform part-of-speech tagging and word segmentation on entity descriptions in a knowledge base, and filtering stop words according to a stop word list;
step 5, determining each concerned entity to be disambiguated and the context characteristics thereof according to the application scene;
step 6, generating candidate entities, searching a knowledge base, comparing whether the name of the entity to be disambiguated in the text is the same as the name of the entity in the knowledge base, if so, taking the entities as the candidate entities of the name of the entity to be disambiguated in the text, and determining the candidate entity set of each entity to be disambiguated and the description characteristics of each candidate entity in the set;
in the entity semantic representation phase (step 7-10):
step 7, a word vector matrix is formed by the word vectors of nouns in a fixed-size window taking the entity to be disambiguated as the center; (after the part of speech tagging and word segmentation processing is carried out on the text set, the word with/n marks is carried out), and the window size is 10;
1) calling a Google deep learning program word2vec to train a Wikipedia corpus so as to obtain a word vector table L, wherein the length of a word vector is 200 dimensions, and each dimension is a real number;
2) context to be disambiguated entity ee＝{w1,w2,…,wi,…,wKEach noun w iniInquiring the word vector table L to obtain the word vector v of each nouni；
3) Constructing a context word vector matrix [ v ] of the entity e to be disambiguated according to the word vector of the context word of the entity e to be disambiguated1,v2,v3,…vi,…,vK]；
4) And (6) ending.
Step 8, word vectors of the first 20 nouns with larger weights are obtained from the summary information of each entity in the knowledge base after TF & IDF calculation to form a word vector matrix; if the number of the nouns is less than 20, taking all the existing nouns;
1) for candidate entity set E ═ E1,e2,…,enEach candidate entity e iniEach noun w in the description of (1)iInquiring the word vector table L to obtain the word vector v of each nouni；
2) Constructing a word vector matrix of entity description according to the word vector of each noun in the description characteristics;
3) and (6) ending.
Step 9, the word vector matrix in the step 7 is used as the context semantic feature of the entity to be disambiguated;
step 10, the word vector matrix in the step 8 is used as the semantic feature of the knowledge base entity;
in the neural network learning training phase (step 11-12):
step 11, combining known unambiguous entities in the text with knowledge base entities to form a training set;
step 12, inputting the training set in the step 11 into a convolutional neural network model for training, and adjusting parameters in the model;
1) each semantic representation to be disambiguated and the semantic features of the candidate entity set are used as a training sample and input into the neural network model;
2) convolving semantic features to be disambiguated, setting the number of convolution kernel feature maps to be 200, and setting the size of the convolution kernel feature maps to be [2, 200], namely setting a matrix with the length of 2 and the width of 200;
3) pooling the convolution result of each convolution kernel by using 1-max to obtain the characteristic of each convolution kernel;
4)200 convolution kernel features form an intermediate result, the intermediate result is input into a full connection layer, the size of the full connection layer is 50, and a 50-dimensional semantic feature vector is finally obtained;
5) the semantic features of the candidate entity set are added and averaged, then the sum and the average are input into a full connection layer, the size of the full connection layer is also 50, and finally a 50-dimensional semantic feature vector is obtained;
6) loss function Loss of each training sample in neural networkeIs defined as:
Losse＝max(0,1-sim(e,e)+sim(e,e′))
wherein: e.g. of the typeRepresenting a target entity of an entity e to be disambiguated, and e' representing any other candidate entity in the candidate entity set, which means that the difference between the semantic feature vector similarity of the target entity and any other candidate entity is maximized;
the global Loss function is defined as Loss ∑ Losse；
7) Parameters in the neural network are initialized by uniformly distributed U (-0.01, 0.01);
8) the activation functions in the neural network all adopt tanh hyperbolic tangent activation functions;
9) parameters in the neural network are updated by adopting random gradient descent;
10) and (6) ending.
In the entity classification phase (steps 13-14):
step 13, reading a sample set of the entity to be disambiguated and the knowledge base candidate entity in the text;
step 14, traversing the sample set read in step 13, inputting each sample into the convolutional neural network model obtained by training in step 12, and outputting a classification result;
step 15 is the end step of the entity disambiguation method based on word vector, convolutional neural network of the present invention;
FIG. 2 is a detailed overview of the neural network structure of step 12 in the training phase for neural network learning of FIG. 1, including the following components:
a word vector matrix: taking a word vector matrix of the context of the entity to be disambiguated and a word vector matrix of the entity description characteristics of the knowledge base as the input of the convolutional neural network;
and (3) rolling layers: carrying out convolution on the context word vector matrix of the entity to be disambiguated through 200 different convolution kernels to obtain the characteristics of each convolution kernel;
1-max pooling layer: performing 1-max pooling on the output characteristics of the convolutional layer to obtain a 200-dimensional intermediate result;
full connection layer: connecting a full-connection layer with the size of 50 to the intermediate result, and connecting a full-connection layer with the size of 50 to the word vector sum average of the candidate entities in the knowledge base so as to obtain two 50-dimensional semantic feature vectors;
and (3) similarity calculation: calculating cosine similarity of the two semantic feature vectors;
FIG. 3 is a detailed flow description of step 14 in the entity classification phase of FIG. 1:
step 16 is the start state diagram of FIG. 3;
step 17, reading a trained neural network model in the file system;
step 18, reading a sample set of the entity to be disambiguated and the knowledge base candidate entity in the text;
step 19, inputting the sample set into a convolutional neural network model to obtain semantic feature vectors, traversing the knowledge base candidate entity set, and calculating the cosine similarity of the semantic feature vectors of the entity to be disambiguated and each candidate entity;
step 20, outputting the entity with the highest similarity as a final target entity;
step 21 is the ending state diagram of FIG. 3;
specifically, the method comprises the following steps: 1) reading a semantic feature vector a of an entity e to be disambiguated from a file system;
2) reading a candidate entity set E ═ E from a file system1,e2,…,enSet of semantic feature vectors in { B ═ B } ═ B1,b2,…,bn}；
3) Traversing the candidate entity set, and calculating the cosine similarity of each feature vector in E and E
Figure BDA0001303391840000073
Figure BDA0001303391840000071
4) Selecting the entity with the largest similarity
Figure BDA0001303391840000072
As a final prediction result;
5) and (6) ending.
In summary, the present invention constructs a word vector matrix by comprehensively using the word vector and the convolutional neural network, and respectively constructing the word vector matrix for the context of the entity to be disambiguated and the abstract information of the candidate entity in the knowledge base, and inputting the word vector matrix into the convolutional neural network model. And training a convolutional neural network model, and adjusting parameters in the model. In the prediction phase, the most similar entity is output as the target entity. The problem that the semantic representation capability is insufficient due to the existence of a vocabulary gap in a traditional bag-of-words model is solved, and the accuracy of entity disambiguation is further improved.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (3)

1. An entity disambiguation method based on word vectors and convolutional neural networks is characterized in that: the method comprises the following steps:
step 1: according to a text set which is collected by an application scene and contains entities to be disambiguated, preprocessing the text set, and determining each entity to be disambiguated and context characteristics thereof in the text set;
step 2: constructing a knowledge base of the entities to be disambiguated according to the domain knowledge, searching the knowledge base, and determining a candidate entity set of each entity to be disambiguated and description characteristics of each candidate entity in the set;
and step 3: taking word vectors of nouns in a fixed-size window taking an entity to be disambiguated as a center to form a word vector matrix as a context semantic feature of the entity to be disambiguated; taking word vectors of the first 20 nouns with larger weights after calculating TF & IDF of the abstract information of each entity in the knowledge base to form a word vector matrix as semantic features of the knowledge base entities;
the method for forming the word vector matrix by using the word vectors of the nouns in the fixed-size window with the entity to be disambiguated as the center specifically comprises the following steps:
1) calling a Google deep learning program word2vec to train a Wikipedia corpus so as to obtain a word vector table L, wherein the length of a word vector is 200 dimensions, and each dimension is a real number;
2) context to be disambiguated entity ee＝{w1,w2,…,wi,…,wKEach noun w iniInquiring the word vector table L to obtain the word vector v of each nouni；
3) Constructing a context word vector matrix [ v ] of the entity e to be disambiguated according to the word vector of the context word of the entity e to be disambiguated1,v2,v3,…vi,…,vK]；
4) Finishing;
the specific method for forming the word vector matrix by taking the word vectors of the first 20 nouns with larger weights after calculating TF & IDF of the abstract information of each entity in the knowledge base is as follows:
1) for candidate entity set E ═ E1,e2,…,enEach candidate entity e iniEach noun w in the description of (1)iInquiring the word vector table L to obtain the word vector v of each nouni；
2) Constructing a word vector matrix of entity description according to the word vector of each noun in the description characteristics;
3) finishing;
and 4, step 4: combining known unambiguous entities in the text with target entities and candidate entities in a knowledge base to form a training set, inputting the training set into a convolutional neural network model for training, and adjusting parameters in the model; the method specifically comprises the following steps:
1) each semantic feature to be disambiguated and the semantic features of the candidate entity set are used as a training sample and input into the neural network model;
2) convolving semantic features to be disambiguated, setting the number of convolution kernel feature maps to be 200, and setting the size of the convolution kernel feature maps to be [2, 200], namely setting a matrix with the length of 2 and the width of 200;
3) pooling the convolution result of each convolution kernel by using 1-max to obtain the characteristic of each convolution kernel;
4)200 convolution kernel features form an intermediate result, the intermediate result is input into a full connection layer, the size of the full connection layer is 50, and a 50-dimensional semantic feature vector is finally obtained;
5) the semantic features of the candidate entity set are added and averaged, then the sum and the average are input into a full connection layer, the size of the full connection layer is also 50, and finally a 50-dimensional semantic feature vector is obtained;
6) loss function Loss of each training sample in neural networkeIs defined as:
Losse＝max(0,1-sim(e,e)+sim(e,e′))
wherein: e.g. of the typeRepresenting a target entity of an entity e to be disambiguated, and e' representing any other candidate entity in the candidate entity set, which means that the difference between the semantic feature vector similarity of the target entity and any other candidate entity is maximized;
the global Loss function is defined as Loss ∑ Losse；
7) Parameters in the neural network are initialized by uniformly distributed U (-0.01, 0.01);
8) the activation functions in the neural network all adopt tanh hyperbolic tangent activation functions;
9) parameters in the neural network are updated by adopting random gradient descent;
10) finishing;
and 5: inputting a sample consisting of each entity to be disambiguated and the knowledge base candidate entity set into the convolutional neural network model obtained in the step (4) to respectively obtain semantic feature vectors of each knowledge base entity in the entity to be disambiguated and the knowledge base candidate entity set;
step 6: calculating the cosine similarity of the entity to be disambiguated and each entity in the knowledge base candidate entity set based on the semantic feature vector; taking the candidate entity with the maximum similarity as the final target entity of the entity to be disambiguated; the method specifically comprises the following steps:
1) reading a semantic feature vector a of an entity e to be disambiguated from a file system;
2) reading a candidate entity set E ═ E from a file system1,e2,…,enSet of semantic feature vectors in { B ═ B } ═ B1,b2,…,bn}；
3) Traversing the candidate entity set, and calculating the cosine similarity l of each feature vector in E and Ei,
Figure FDA0002602389770000021
4) Selecting the entity with the largest similarity
Figure FDA0002602389770000022
As a final prediction result;
5) and (6) ending.
2. The entity disambiguation method of claim 1, characterized in that: the preprocessing in the step 1 is to use Chinese word segmentation program ICTCCLAS of Chinese academy of sciences to perform part-of-speech tagging and word segmentation on a text set, then filter out stop words according to a stop word list, and create a noun dictionary for proper nouns and entity names which are difficult to recognize.
3. The entity disambiguation method of claim 1, characterized in that: and in the step 2, a Chinese word segmentation program ICTCCLAS is called to perform part-of-speech tagging and word segmentation on entity descriptions in the knowledge base, and stop words are filtered according to the stop word list.
CN201710373502.2A 2017-05-24 2017-05-24 Entity disambiguation method based on word vector and convolutional neural network Active CN107102989B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710373502.2A CN107102989B (en) 2017-05-24 2017-05-24 Entity disambiguation method based on word vector and convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710373502.2A CN107102989B (en) 2017-05-24 2017-05-24 Entity disambiguation method based on word vector and convolutional neural network

Publications (2)

Publication Number Publication Date
CN107102989A CN107102989A (en) 2017-08-29
CN107102989B true CN107102989B (en) 2020-09-29

Family

ID=59670296

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710373502.2A Active CN107102989B (en) 2017-05-24 2017-05-24 Entity disambiguation method based on word vector and convolutional neural network

Country Status (1)

Country Link
CN (1) CN107102989B (en)

Families Citing this family (55)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107562729B (en) * 2017-09-14 2020-12-08 云南大学 Party building text representation method based on neural network and theme enhancement
CN107730002B (en) * 2017-10-13 2020-06-02 国网湖南省电力公司 Intelligent fuzzy comparison method for remote control parameters of communication gateway machine
CN107729509B (en) * 2017-10-23 2020-07-07 中国电子科技集团公司第二十八研究所 Discourse similarity determination method based on recessive high-dimensional distributed feature representation
CN110019792A (en) * 2017-10-30 2019-07-16 阿里巴巴集团控股有限公司 File classification method and device and sorter model training method
CN108280061B (en) * 2018-01-17 2021-10-26 北京百度网讯科技有限公司 Text processing method and device based on ambiguous entity words
CN108304552B (en) * 2018-02-01 2021-01-08 浙江大学 Named entity linking method based on knowledge base feature extraction
CN108335731A (en) * 2018-02-09 2018-07-27 辽宁工程技术大学 A kind of invalid diet's recommendation method based on computer vision
CN108399230A (en) * 2018-02-13 2018-08-14 上海大学 A kind of Chinese financial and economic news file classification method based on convolutional neural networks
CN108446269B (en) * 2018-03-05 2021-11-23 昆明理工大学 Word sense disambiguation method and device based on word vector
CN108573047A (en) * 2018-04-18 2018-09-25 广东工业大学 A kind of training method and device of Module of Automatic Chinese Documents Classification
CN108563766A (en) * 2018-04-19 2018-09-21 天津科技大学 The method and device of food retrieval
CN108959242B (en) * 2018-05-08 2021-07-27 中国科学院信息工程研究所 Target entity identification method and device based on part-of-speech characteristics of Chinese characters
CN108647785A (en) * 2018-05-17 2018-10-12 普强信息技术（北京）有限公司 A kind of neural network method for automatic modeling, device and storage medium
CN108647191B (en) * 2018-05-17 2021-06-25 南京大学 Sentiment dictionary construction method based on supervised sentiment text and word vector
CN108804595B (en) * 2018-05-28 2021-07-27 中山大学 Short text representation method based on word2vec
CN110555208B (en) * 2018-06-04 2021-11-19 北京三快在线科技有限公司 Ambiguity elimination method and device in information query and electronic equipment
CN108805290B (en) * 2018-06-28 2021-03-12 国信优易数据股份有限公司 Entity category determination method and device
CN108921213B (en) * 2018-06-28 2021-06-22 国信优易数据股份有限公司 Entity classification model training method and device
CN109101579B (en) * 2018-07-19 2021-11-23 深圳追一科技有限公司 Customer service robot knowledge base ambiguity detection method
CN108920467B (en) * 2018-08-01 2021-04-27 北京三快在线科技有限公司 Method and device for learning word meaning of polysemous word and search result display method
CN109325108B (en) * 2018-08-13 2022-05-27 北京百度网讯科技有限公司 Query processing method, device, server and storage medium
CN109241294A (en) * 2018-08-29 2019-01-18 国信优易数据有限公司 A kind of entity link method and device
CN109214007A (en) * 2018-09-19 2019-01-15 哈尔滨理工大学 A kind of Chinese sentence meaning of a word based on convolutional neural networks disappears qi method
CN109299462B (en) * 2018-09-20 2022-11-29 武汉理工大学 Short text similarity calculation method based on multi-dimensional convolution characteristics
CN109614615B (en) * 2018-12-04 2022-04-22 联想(北京)有限公司 Entity matching method and device and electronic equipment
CN109740728B (en) * 2018-12-10 2019-11-01 杭州世平信息科技有限公司 A kind of measurement of penalty calculation method based on a variety of neural network ensembles
CN109635114A (en) * 2018-12-17 2019-04-16 北京百度网讯科技有限公司 Method and apparatus for handling information
CN109933788B (en) * 2019-02-14 2023-05-23 北京百度网讯科技有限公司 Type determining method, device, equipment and medium
CN110263324B (en) * 2019-05-16 2021-02-12 华为技术有限公司 Text processing method, model training method and device
CN110598846B (en) * 2019-08-15 2022-05-03 北京航空航天大学 Hierarchical recurrent neural network decoder and decoding method
CN110705292B (en) * 2019-08-22 2022-11-29 成都信息工程大学 Entity name extraction method based on knowledge base and deep learning
CN110569506A (en) * 2019-09-05 2019-12-13 清华大学 Medical named entity recognition method based on medical dictionary
CN110705295B (en) * 2019-09-11 2021-08-24 北京航空航天大学 Entity name disambiguation method based on keyword extraction
CN110674304A (en) * 2019-10-09 2020-01-10 北京明略软件系统有限公司 Entity disambiguation method and device, readable storage medium and electronic equipment
CN110826331B (en) * 2019-10-28 2023-04-18 南京师范大学 Intelligent construction method of place name labeling corpus based on interactive and iterative learning
CN110852106B (en) * 2019-11-06 2024-05-03 腾讯科技（深圳）有限公司 Named entity processing method and device based on artificial intelligence and electronic equipment
CN110852108B (en) * 2019-11-11 2022-03-29 中山大学 Joint training method, apparatus and medium for entity recognition and entity disambiguation
CN113010633B (en) * 2019-12-20 2023-01-31 海信视像科技股份有限公司 Information interaction method and equipment
CN111241298B (en) * 2020-01-08 2023-10-10 腾讯科技（深圳）有限公司 Information processing method, apparatus, and computer-readable storage medium
CN111241824B (en) * 2020-01-09 2020-11-24 中国搜索信息科技股份有限公司 Method for identifying Chinese metaphor information
CN111310481B (en) * 2020-01-19 2021-05-18 百度在线网络技术（北京）有限公司 Speech translation method, device, computer equipment and storage medium
CN111597804B (en) * 2020-05-15 2023-03-10 腾讯科技（深圳）有限公司 Method and related device for training entity recognition model
CN111709243B (en) * 2020-06-19 2023-07-07 南京优慧信安科技有限公司 Knowledge extraction method and device based on deep learning
CN112069826B (en) * 2020-07-15 2021-12-07 浙江工业大学 Vertical domain entity disambiguation method fusing topic model and convolutional neural network
CN112100356A (en) * 2020-09-17 2020-12-18 武汉纺织大学 Knowledge base question-answer entity linking method and system based on similarity
CN112257443B (en) * 2020-09-30 2024-04-02 华泰证券股份有限公司 MRC-based company entity disambiguation method combined with knowledge base
CN112464669B (en) * 2020-12-07 2024-02-09 宁波深擎信息科技有限公司 Stock entity word disambiguation method, computer device, and storage medium
CN112966117A (en) * 2020-12-28 2021-06-15 成都数之联科技有限公司 Entity linking method
CN112580351B (en) * 2020-12-31 2022-04-19 成都信息工程大学 Machine-generated text detection method based on self-information loss compensation
CN113761218B (en) * 2021-04-27 2024-05-10 腾讯科技（深圳）有限公司 Method, device, equipment and storage medium for entity linking
CN113283236B (en) * 2021-05-31 2022-07-19 北京邮电大学 Entity disambiguation method in complex Chinese text
CN113361283B (en) * 2021-06-28 2024-09-24 东南大学 Paired entity joint disambiguation method for Web form
CN113704416B (en) * 2021-10-26 2022-03-04 深圳市北科瑞声科技股份有限公司 Word sense disambiguation method and device, electronic equipment and computer-readable storage medium
CN114298028B (en) * 2021-12-13 2024-09-03 盈嘉互联（北京）科技有限公司 BIM semantic disambiguation method and system
CN116976324A (en) * 2022-04-21 2023-10-31 北京沃东天骏信息技术有限公司 Disambiguation method and disambiguation device for product words

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104572892A (en) * 2014-12-24 2015-04-29 中国科学院自动化研究所 Text classification method based on cyclic convolution network
CN106295796A (en) * 2016-07-22 2017-01-04 浙江大学 Entity link method based on degree of depth study
CN106547735A (en) * 2016-10-25 2017-03-29 复旦大学 The structure and using method of the dynamic word or word vector based on the context-aware of deep learning
CN106570170A (en) * 2016-11-09 2017-04-19 武汉泰迪智慧科技有限公司 Text classification and naming entity recognition integrated method and system based on depth cyclic neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104572892A (en) * 2014-12-24 2015-04-29 中国科学院自动化研究所 Text classification method based on cyclic convolution network
CN106295796A (en) * 2016-07-22 2017-01-04 浙江大学 Entity link method based on degree of depth study
CN106547735A (en) * 2016-10-25 2017-03-29 复旦大学 The structure and using method of the dynamic word or word vector based on the context-aware of deep learning
CN106570170A (en) * 2016-11-09 2017-04-19 武汉泰迪智慧科技有限公司 Text classification and naming entity recognition integrated method and system based on depth cyclic neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于中文维基百科的命名实体消歧方法;杜婧君等;《杭州电子科技大学学报》;20121231;第32卷(第6期);第57-60页 *

Also Published As

Publication number Publication date
CN107102989A (en) 2017-08-29

Similar Documents

Publication Publication Date Title
CN107102989B (en) Entity disambiguation method based on word vector and convolutional neural network
WO2021114745A1 (en) Named entity recognition method employing affix perception for use in social media
Kumar et al. Identifying clickbait: A multi-strategy approach using neural networks
CN104615767B (en) Training method, search processing method and the device of searching order model
CN106570141B (en) Approximate repeated image detection method
CN104391942B (en) Short essay eigen extended method based on semantic collection of illustrative plates
CN112084331A (en) Text processing method, text processing device, model training method, model training device, computer equipment and storage medium
CN107085581A (en) Short text classification method and device
CN107480143A (en) Dialogue topic dividing method and system based on context dependence
CN106095749A (en) A kind of text key word extracting method based on degree of depth study
Tuan et al. Multimodal fusion with BERT and attention mechanism for fake news detection
CN110347790B (en) Text duplicate checking method, device and equipment based on attention mechanism and storage medium
CN112434533B (en) Entity disambiguation method, entity disambiguation device, electronic device, and computer-readable storage medium
CN110222328B (en) Method, device and equipment for labeling participles and parts of speech based on neural network and storage medium
Zhang et al. Relation classification via BiLSTM-CNN
CN106933787A (en) Adjudicate the computational methods of document similarity, search device and computer equipment
CN108304377B (en) Extraction method of long-tail words and related device
CN110569405A (en) method for extracting government affair official document ontology concept based on BERT
CN105760363B (en) Word sense disambiguation method and device for text file
CN105740448B (en) More microblogging timing abstract methods towards topic
CN117251551B (en) Natural language processing system and method based on large language model
JP7181999B2 (en) SEARCH METHOD AND SEARCH DEVICE, STORAGE MEDIUM
CN110390104B (en) Irregular text transcription method and system for voice dialogue platform
Le Huy et al. Keyphrase extraction model: a new design and application on tourism information
CN113076744A (en) Cultural relic knowledge relation extraction method based on convolutional neural network

Legal Events

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