CN111046185B - Method, device and terminal for extracting knowledge graph relation of text information - Google Patents
Method, device and terminal for extracting knowledge graph relation of text information Download PDFInfo
- Publication number
- CN111046185B CN111046185B CN201911291310.2A CN201911291310A CN111046185B CN 111046185 B CN111046185 B CN 111046185B CN 201911291310 A CN201911291310 A CN 201911291310A CN 111046185 B CN111046185 B CN 111046185B
- Authority
- CN
- China
- Prior art keywords
- word vector
- relation
- extracting
- extracted
- sknet
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Abstract
The invention relates to the technical field of information, in particular to a method, a device and a terminal for extracting a knowledge graph relation of text information; the method comprises the steps of obtaining text information of a knowledge graph relation to be extracted; respectively constructing word vectors of the text information by using a plurality of word vector models; transversely splicing the word vectors respectively constructed to obtain a final word vector; extracting the relation characteristics of the final word vector by adopting a multilayer SKnet network; and performing mixed pooling operation on the relationship features, and putting the mixed pooled relationship features into a classifier to obtain a knowledge graph relationship extraction label of the text information to which the relationship features belong. The word vectors generated by different word vector tools are spliced so as to increase word information; extracting word vector characteristics through a multilayer SKnet network; and the extracted features are subjected to mixed pooling operation, so that the relationship features with more abundant and accurate information can be obtained. The invention can improve the efficiency of extracting the text information relation.
Description
Technical Field
The invention relates to the technical field of information, in particular to a method, a device and a terminal for extracting a knowledge graph relation of text information.
Background
With the rapid development of the internet, various resource information is gradually enriched and even increased explosively; characters bear abundant information, and more attention needs to be paid to the research on the texts. Therefore, it is very important to extract the concerned content from the huge text, and the purpose of information extraction is to provide a powerful information acquisition tool for people. The relation extraction method based on the remote supervision deep learning is developed greatly at present, can be applied to a big data background, and can divide the relation extraction of the remote supervision deep learning into four modules: word embedding, sentence feature extraction, attention selection mechanism and classification; the sentence feature extraction is a module which needs to pay attention in relation extraction, and the quality of the sentence features determines the quality of final relation extraction, so that the introduction of an excellent feature extraction network in the sentence feature extraction is in the direction of needing effort. The high-quality feature extraction network is introduced, and the sentence feature extraction quality can be improved, so that the relation extraction quality of the knowledge graph can be greatly improved, and more accurate extraction information can be obtained.
At present, a convolutional neural network and a variant thereof are mostly adopted in the relation extraction task, for example, CNN is used for carrying out convolution on sentences for feature extraction, the obtained features are pooled, and finally the features are sent to a softmax layer to obtain the semantic relation to be extracted. However, the network used in the relation extraction task at present cannot dynamically select the size of the convolution kernel, so that the characteristics most suitable for the relation extraction task cannot be obtained; in addition, the position information is ignored by the pooling method, and the information acquisition capability is insufficient.
Disclosure of Invention
In order to improve the knowledge graph relation extraction effect and obtain a more accurate extraction effect, the invention can dynamically select the size of a convolution kernel based on an SKnet network in a relation extraction task to obtain more appropriate characteristics, and provides a new pooling method which can capture more information and retain position information to the maximum extent; the invention greatly improves the characteristic extraction capability of the network, thereby improving the relation extraction task quality. In order to improve the extraction effect of the relation of the knowledge graph; the invention provides a method, a device and a terminal for extracting a knowledge graph relation of text information, wherein the method comprises the following steps:
in a first aspect of the present invention, a method for extracting a knowledge-graph relationship of text information includes the following steps:
acquiring text information of a relation of a knowledge graph to be extracted;
respectively constructing word vectors of the text information of the relation of the knowledge graph to be extracted by using a plurality of word vector models;
transversely splicing the word vectors which are respectively constructed to obtain a final word vector of the text information of the relation of the knowledge graph to be extracted;
extracting the relation characteristics of the final word vector by adopting a multilayer SKnet network;
performing mixed pooling operation on the extracted relation characteristics to obtain more position information;
and putting the mixed and pooled relation features into a softmax classifier, and acquiring the labels to which the relation features belong.
Further, the plurality of Word vector models include a Word2vec model, a Glove model and a BERT model.
Preferably, the final word vector further comprises adding position information to the end of the word vector after the transverse splicing, wherein the position information comprises relative position information between each word and the entity in the text.
Furthermore, the extracting of the relational features of the final word vector by adopting the multilayer SKnet network comprises that each layer of SKnet network adaptively adjusts the size of the receptive field according to a knowledge map relational extraction task, sets m receptive fields with different sizes, and each receptive field respectively performs convolution operation on the input final word vector; obtaining a preliminary convolution result; adding and fusing convolution results; performing dimensionality reduction on the fused result by adopting global average pooling to obtain information of each channel, then performing dimensionality enhancement by adopting a full-connection layer in order to increase nonlinear characteristics, and processing the dimensionality enhancement result by adopting a softmax layer to obtain selection coefficients of the sizes of the receptive fields; and multiplying the receptive field size selection coefficient by the corresponding receptive field initial convolution result to obtain the relational characteristics of the final word vector.
Further, the step of performing mixed pooling on the extracted relational features comprises combining a K-maximum pooling layer with an average pooling layer; and splicing the result obtained by the K-maximum pooling and the result obtained by the average pooling to obtain the final result after the mixed pooling.
In a second aspect of the present invention, the present invention further provides a knowledge-graph relation extracting apparatus for text information, including:
the text collection module is used for acquiring text information of the relation of the knowledge graph to be extracted;
the word vector analysis module is used for respectively extracting word vectors of the text information of the relation of the knowledge graph to be extracted by adopting a plurality of word vector models;
the word vector splicing module is used for transversely splicing the word vectors extracted by adopting the various word vector models to obtain a final word vector;
the relational feature extraction module is used for extracting the extracted final word vector by adopting a multilayer SKnet network and obtaining the relational feature of the final word vector;
the mixed pooling module is used for performing mixed pooling operation on the extracted relationship characteristics;
and the softmax classifier is used for classifying the extracted relationship features after the mixing and pooling and endowing classification labels to the relationship features.
Based on the same concept as the present invention, in a third aspect of the present invention, the present invention further provides a knowledge-graph relation extraction terminal for text information, comprising a processor and a memory, wherein the memory stores a computer program capable of running on the processor, and the processor executes the program according to the above knowledge-graph relation extraction method for text information.
The invention has the beneficial effects that:
the method comprises the steps of training text information of knowledge graph relations to be extracted in a data set through different word vector tools to generate different word vectors, splicing the generated word vectors to increase word information, and adding position information at the tail end of the word vectors; extracting information of a data set to be extracted through a deep SKnet network; the information extracted by the SKnet is subjected to mixed pooling operation, so that the relation characteristics of more abundant and accurate information can be obtained; and obtaining a final relation extraction result through a softmax classifier. The invention can increase word vector information, extract more effective relation characteristics, is simple and efficient, and can achieve better knowledge map relation extraction effect. The invention generates various word vector information to combine word vectors, and increases more information compared with the current method for extracting the Chinese word vectors by relation; the convolution kernel size can be dynamically selected based on the SKnet network, more appropriate information can be obtained, and the method is superior to the traditional convolution network which only can select the convolution kernel size with a single size; compared with a maximum pooling method, the hybrid pooling method can capture more information to the maximum extent and retain position information.
Drawings
Fig. 1 is a flowchart of a method for extracting a knowledge-graph relationship of text information according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of word vector acquisition according to an embodiment of the present invention.
Fig. 3 is a flowchart of feature extraction performed by a multi-layer SKnet network according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of extracting features of a single-layer SKnet network according to an embodiment of the present invention.
FIG. 5 is a block diagram of a hybrid pooling scheme provided by an embodiment of the present invention;
fig. 6 is a device for extracting a knowledge-graph relationship of text information according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clearly and completely apparent, the technical solutions in the embodiments of the present invention are described below with reference to the accompanying drawings, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
The method for extracting the knowledge-graph relationship of the text information can be applied to the device for extracting the knowledge-graph relationship of the text information, the device for classifying the text information can be configured in a terminal for extracting the knowledge-graph relationship of the text information, and the classifying terminal can be a computer device or a server and the like.
As shown in fig. 1, in one embodiment, a method for extracting a knowledge-graph relationship of text information includes:
s1, acquiring text information of a relation of a knowledge graph to be extracted;
the text information of the knowledge graph relationship to be extracted is a text from which entity relationship information needs to be extracted, the text to be classified can be a text input by a user or a text in a designated text storage space, and the designated text storage space can store a text of a certain specific application scene, for example, a microblog text stored in the designated text storage space.
S2, respectively constructing word vectors of the text information of the knowledge graph relationship to be extracted by using a plurality of word vector models;
wherein, the word vector model refers to a word vector tool; in this embodiment, three models, namely a Word2vec model, a Glove model and a BERT model, are preferably used for processing text information to be classified. The present invention may not be limited to these three types of models for processing text information.
As shown in fig. 2, the collected text information is respectively processed as follows:
training the text information to be classified by adopting Word2vec tool to generate Word vector v word2vec ；
Processing the text information to be classified by adopting a Glove tool to generate a word vector v glove ；
Training the text information to be classified by adopting a BERT tool to generate a word vector v bert 。
Wherein v is word2vec ∈R n×d ，v glove ∈R n×d ，v bert ∈R n×d N is the number of all words, and d is the word vector dimension.
S3, transversely splicing the word vectors which are respectively constructed to obtain a final word vector of the text information to be classified;
based on the above embodiment, in order to increase the information contained in the word vectors, the present embodiment needs to transversely splice the generated word vectors to obtain the final word vector v word ＝v word2vec +v glove +v bert Word information is added.
As a preferred embodiment of the method according to the invention,the relative position between each word and each entity in the text also contains rich information, so the embodiment also adds the position information v at the tail end of the word vector position To obtain the final result v = v word +v position 。
S4, extracting the relation characteristics of the final word vector by adopting a multilayer SKnet network;
after the final word vector after splicing is obtained, the sentence needs to be subjected to feature extraction, as shown in fig. 3, where the sentence S contains S words S = { v = 1 ,v 2 ,……,v s And each word is expressed in a vector form, a multilayer SKnet network is used for extracting information of the text, and c = W sknet v i:j Wherein c is the result of extracting the characteristics of the multilayer SKnet, and W is a parameter matrix extracted by the multilayer SKnet network; v. of i:j Representing the ith through jth word vectors, all belonging to sentence S.
In order to obtain a better feature extraction effect and reduce noise influence, more than one layer of SKnet network is used for extracting sentence features, and in the multilayer SKnet network, the receptive field size suitable for a knowledge map relation extraction task can be selected from each layer, so that the noise influence is reduced, and the better feature extraction effect is achieved.
The multilayer SKnet network is actually formed by stacking a plurality of single-layer SKnet networks. Each single-layer SKnet network performs feature extraction on input data, as shown in fig. 4, which is a flow chart for performing feature extraction on a multi-layer SKnet network provided in this embodiment, each layer SKnet can adaptively adjust the size of its receptive field according to a knowledge graph relation extraction task, and set m (m +1 is smaller than the sentence length) receptive field sizes of different sizes as { w ″ 1 ＝2、w 2 ＝3、w 3 ＝4、……、w m = m +1, these receptive fields being set for the input word vector S = { v = { (v) } 1 ,v 2 ,……,v s Carry out convolution operation to get primary convolution result { U } 1 ，U 2 ，……，U m Adding and fusing all convolution results to obtain U; performing global average pooling on the result U obtained by fusion to perform dimensionality reduction to obtain a dimensionality reduction result Z, and increasing the dimensionality reduction result Z to increase the dimensionality reduction result ZMultiple nonlinearity is achieved, the full connection layer is accessed to the multiple nonlinearity, dimension increasing is conducted on the multiple nonlinearity to obtain X, and then the softmax layer is used for processing the X to obtain a receptive field size selection coefficient { X 1 ，x 2 ，……，x m }，(x 1 +x 2 +……+x m = 1), multiplying the final factor of selecting the size of the receptive field by the corresponding preliminary convolution result of the receptive field, and accumulating the multiplied results to obtain the final characteristic T, wherein T = x 1 *U 1 +x 2 *U 2 ……,x m *U m 。
And S5, performing mixed pooling operation on the extracted relation characteristics to acquire more position information.
In a preferred embodiment, after extracting features through the multi-layer SKnet network, as shown in fig. 5, a hybrid pooling operation is required; in order to reduce the loss of position information of the features, highlight strong features and enhance information, mixed pooling is performed by combining a K-maximum pooling layer and average pooling, and a result obtained by the K-maximum pooling layer and a result obtained by the average pooling are spliced to obtain a final pooling result. If K is chosen to be 2, then pooling results are obtained where 2-maximal pooling is combined with average pooling.
S6, putting the relation features into a softmax classifier, and obtaining the labels to which the relation features belong.
In a specific embodiment, if apple, qiao Busi and apple represent two entities created for a text "Qiao Busi of a relation of a knowledge graph to be extracted, the implementation needs to extract a relation between the two entities, i.e.," create "relation; in this embodiment, a large amount of text information is firstly subjected to Word vector training, for example, the NYT-Freebase data set can be adopted to respectively train the text information by using Word2vec, glove and BERT tools to generate different Word vectors v word2vec 、v glove 、v bert Transversely splicing the generated word vectors to obtain a final word vector v word ＝v word2vec +v glove +v bert Adding word information and adding position information v at the end of word vector position To obtain the final result v = v word +v position . And then, the obtained word vectors are sent to a neural network for feature extraction.
In the process of extracting the features of the SKnet network, assuming that the sizes of convolution kernels are 3, 5 and 7 respectively, preliminarily obtaining convolution results { U) with the sizes of the convolution kernels being 3, 5 and 7 1 ，U 2 ，U 3 And merging the three convolution results to obtain U = U 1 +U 2 +U 3 Performing global average pooling on the result U obtained by fusion to perform dimension reduction to obtain a dimension reduction result Z, accessing the result U to a full connection layer to perform dimension increase to obtain X in order to increase more nonlinearity, wherein X is a coefficient tensor, and then processing the X by using a softmax layer to map the coefficient value to a position between 0 and 1 to obtain 3 receptive field size selection coefficients { X } 1 ，x 2 ，x 3 }，(x 1 +x 2 +x 3 = 1), multiplying the final receiving field size selection coefficient by the corresponding receiving field initial convolution result, and accumulating the multiplied results to obtain the final characteristic T, wherein T = x 1 *U 1 +x 2 *U 2 ,x 3 *U 3 。
After the features extracted by the SKnet network are obtained, the features are put into a pooling layer, and the mixed pooling method provided by the invention is used for splicing the value of 2-max in the feature T and the average pooling value of T to obtain the content after mixed pooling, so that more information is obtained and position information is reserved. And finally, sending the obtained final relation characteristics into a final softmax layer, and obtaining a final relation label to be extracted after passing through the softmax layer.
In addition, in an embodiment, the present invention further provides a device for extracting a knowledge-graph relationship of text information, as shown in fig. 6, including:
the text collection module is used for acquiring text information of the relation of the knowledge-graph to be extracted;
the word vector analysis module is used for respectively extracting word vectors of the text information of the relation of the knowledge graph to be extracted by adopting a plurality of word vector models;
the word vector splicing module is used for transversely splicing the word vectors extracted by adopting various word vector models to obtain final word vectors;
the relational feature extraction module is used for extracting the extracted final word vector by adopting a multilayer SKnet network and obtaining the relational feature of the final word vector;
the mixed pooling module is used for performing mixed pooling operation on the extracted relationship characteristics;
and the softmax classifier is used for classifying the extracted relationship features after the mixing and pooling and endowing classification labels to the relationship features.
Based on the same concept as the present invention, in a third aspect of the present invention, the present invention further provides a knowledge-graph relation extraction terminal for text information, comprising a processor and a memory, wherein the memory stores a computer program capable of running on the processor, and the processor executes the program according to the above knowledge-graph relation extraction method for text information.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by hardware related to instructions of a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: ROM, RAM, magnetic or optical disks, and the like.
The above-mentioned embodiments, which further illustrate the objects, technical solutions and advantages of the present invention, should be understood that the above-mentioned embodiments are only preferred embodiments of the present invention, and should not be construed as limiting the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A method for extracting a knowledge graph relation of text information is characterized by comprising the following steps:
acquiring text information of a relation of a knowledge graph to be extracted;
respectively constructing word vectors of text information of the relation of the knowledge graph to be extracted by using a plurality of word vector models;
transversely splicing the word vectors which are respectively constructed to obtain a final word vector of the text information of the relation of the knowledge graph to be extracted;
extracting the relation characteristic of the final word vector by adopting a multilayer SKnet network, wherein the relation characteristic is expressed as c = W sknet v i:j ；
Wherein c is the result of extracting the characteristics of multiple layers of SKnet, W sknet Extracting a parameter matrix for the multilayer SKnet network; v. of i:j Representing the ith word vector to the jth word vector, wherein the ith word vector and the jth word vector all belong to sentences corresponding to the final word vector after splicing;
the multilayer SKnet network is formed by stacking a plurality of single-layer SKnet networks; each single-layer SKnet network performs relational feature extraction on the input final word vector;
each layer of SKnet network adaptively adjusts the size of the receptive field according to the knowledge map relation extraction task, sets m receptive fields with different sizes, and each receptive field respectively performs convolution operation on the input final word vector; obtaining a preliminary convolution result; adding and fusing convolution results; performing dimension reduction on the fused result by adopting global average pooling to obtain information of each channel, performing dimension increasing by adopting a full-connection layer in order to increase nonlinear characteristics, and processing the dimension increased result by adopting a softmax layer to obtain selection coefficients of the sizes of the receptive fields; multiplying the receptive field size selection coefficient by the corresponding receptive field initial convolution result to obtain the relational characteristics of the final word vector;
performing mixed pooling operation on the extracted relation characteristics to obtain more position information;
and putting the mixed and pooled relation features into a softmax classifier, and acquiring the labels to which the relation features belong.
2. The method for extracting knowledge-graph relationships of textual information of claim 1, wherein said plurality of Word vector models comprises using Word2vec model, glove model and BERT model.
3. The method of claim 1, wherein the final word vector further comprises adding position information to the end of the word vector after the horizontal concatenation, and the position information comprises relative position information between each word and an entity in the text.
4. The method of claim 1, wherein the performing mixed pooling of extracted relationship features comprises combining a K-max pooling layer with an average pooling layer; and splicing the result obtained by the K-maximum pooling and the result obtained by the average pooling to obtain the final result after the mixed pooling.
5. A device for extracting a knowledge-graph relationship of text information, which is used for implementing a method for extracting a knowledge-graph relationship of text information according to any one of claims 1 to 4, the device comprising:
the text collection module is used for acquiring text information of the relation of the knowledge graph to be extracted;
the word vector analysis module is used for respectively extracting word vectors of the text information of the relation of the knowledge graph to be extracted by adopting a plurality of word vector models;
the word vector splicing module is used for transversely splicing the word vectors extracted by adopting the various word vector models to obtain a final word vector;
the relational feature extraction module is used for extracting the extracted final word vector by adopting a multilayer SKnet network and obtaining the relational feature of the final word vector;
the mixed pooling module is used for performing mixed pooling operation on the extracted relation characteristics;
and the softmax classifier is used for classifying the extracted mixed pooled relational features and endowing the relational features with classification labels.
6. A terminal for extracting knowledge-graph relationships of text messages, comprising a processor and a memory, said memory having stored thereon a computer program operable on said processor, said processor when executing said program implementing the method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911291310.2A CN111046185B (en) | 2019-12-16 | 2019-12-16 | Method, device and terminal for extracting knowledge graph relation of text information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911291310.2A CN111046185B (en) | 2019-12-16 | 2019-12-16 | Method, device and terminal for extracting knowledge graph relation of text information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111046185A CN111046185A (en) | 2020-04-21 |
| CN111046185B true CN111046185B (en) | 2023-02-24 |
Family
ID=70236548
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911291310.2A Active CN111046185B (en) | 2019-12-16 | 2019-12-16 | Method, device and terminal for extracting knowledge graph relation of text information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111046185B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3905097A1 (en) * | 2020-04-30 | 2021-11-03 | Robert Bosch GmbH | Device and method for determining a knowledge graph |
| CN112307212A (en) * | 2020-11-11 | 2021-02-02 | 上海昌投网络科技有限公司 | Public opinion delivery monitoring method for advertisement delivery |
| US11829726B2 (en) | 2021-01-25 | 2023-11-28 | International Business Machines Corporation | Dual learning bridge between text and knowledge graph |
| CN112785350B (en) * | 2021-02-24 | 2023-09-19 | 深圳市慧择时代科技有限公司 | Product vector determining method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446526A (en) * | 2016-08-31 | 2017-02-22 | 北京千安哲信息技术有限公司 | Electronic medical record entity relation extraction method and apparatus |
| CN106598952A (en) * | 2016-12-23 | 2017-04-26 | 大连理工大学 | System for detecting Chinese fuzzy constraint information scope based on convolutional neural network |
| CN109376227A (en) * | 2018-10-29 | 2019-02-22 | 山东大学 | A kind of prison term prediction technique based on multitask artificial neural network |
| CN109684449A (en) * | 2018-12-20 | 2019-04-26 | 电子科技大学 | A kind of natural language characterizing semantics method based on attention mechanism |
| CN109815339A (en) * | 2019-01-02 | 2019-05-28 | 平安科技（深圳）有限公司 | Based on TextCNN Knowledge Extraction Method, device, computer equipment and storage medium |
| CN110134793A (en) * | 2019-05-28 | 2019-08-16 | 电子科技大学 | Text sentiment classification method |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7353215B2 (en) * | 2001-05-07 | 2008-04-01 | Health Discovery Corporation | Kernels and methods for selecting kernels for use in learning machines |
| RU2657173C2 (en) * | 2016-07-28 | 2018-06-08 | Общество с ограниченной ответственностью "Аби Продакшн" | Sentiment analysis at the level of aspects using methods of machine learning |
| CN108563653B (en) * | 2017-12-21 | 2020-07-31 | 清华大学 | Method and system for constructing knowledge acquisition model in knowledge graph |
| CN110458373A (en) * | 2019-08-22 | 2019-11-15 | 电子科技大学成都学院 | A kind of method of crime prediction and system of the fusion of knowledge based map |
-
2019
- 2019-12-16 CN CN201911291310.2A patent/CN111046185B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446526A (en) * | 2016-08-31 | 2017-02-22 | 北京千安哲信息技术有限公司 | Electronic medical record entity relation extraction method and apparatus |
| CN106598952A (en) * | 2016-12-23 | 2017-04-26 | 大连理工大学 | System for detecting Chinese fuzzy constraint information scope based on convolutional neural network |
| CN109376227A (en) * | 2018-10-29 | 2019-02-22 | 山东大学 | A kind of prison term prediction technique based on multitask artificial neural network |
| CN109684449A (en) * | 2018-12-20 | 2019-04-26 | 电子科技大学 | A kind of natural language characterizing semantics method based on attention mechanism |
| CN109815339A (en) * | 2019-01-02 | 2019-05-28 | 平安科技（深圳）有限公司 | Based on TextCNN Knowledge Extraction Method, device, computer equipment and storage medium |
| CN110134793A (en) * | 2019-05-28 | 2019-08-16 | 电子科技大学 | Text sentiment classification method |
Non-Patent Citations (3)
| Title |
|---|
| Selective kernel networks for weakly supervised relation;Ziyang Li等;《CAAI Trans on Intel Tech-2021》;20210408;第6卷(第2期);第224-234页 * |
| Selective Kernel Networks;Xiang Li等;《2019 CVPR》;20190620;第510-519页 * |
| 基于卷积循环神经网络的关系抽取;宋睿等;《中文信息学报》;20191030;第33卷(第10期);第64-72页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111046185A (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111046185B (en) | Method, device and terminal for extracting knowledge graph relation of text information | |
| KR102455616B1 (en) | Theme classification method based on multimodality, device, apparatus, and storage medium | |
| Pang et al. | Generalising fine-grained sketch-based image retrieval | |
| CN106156365B (en) | A kind of generation method and device of knowledge mapping | |
| CN109145712B (en) | Text information fused GIF short video emotion recognition method and system | |
| CN104735468B (en) | A kind of method and system that image is synthesized to new video based on semantic analysis | |
| CN111866610B (en) | Method and apparatus for generating information | |
| CN107145485B (en) | Method and apparatus for compressing topic models | |
| CN109815485B (en) | Method and device for identifying emotion polarity of microblog short text and storage medium | |
| CN110502742B (en) | Complex entity extraction method, device, medium and system | |
| CN109376775A (en) | The multi-modal sentiment analysis method of online news | |
| CN113626589B (en) | Multi-label text classification method based on mixed attention mechanism | |
| CN111143617A (en) | Automatic generation method and system for picture or video text description | |
| CN107832300A (en) | Towards minimally invasive medical field text snippet generation method and device | |
| CN107818173B (en) | Vector space model-based Chinese false comment filtering method | |
| JP2023535108A (en) | Video tag recommendation model training method, video tag determination method, device, electronic device, storage medium and computer program therefor | |
| CN113743079A (en) | Text similarity calculation method and device based on co-occurrence entity interaction graph | |
| CN112784580A (en) | Financial data analysis method and device based on event extraction | |
| CN114297390B (en) | Aspect category identification method and system in long tail distribution scene | |
| CN110717316A (en) | Topic segmentation method and device for subtitle dialog flow | |
| CN113516198B (en) | Cultural resource text classification method based on memory network and graphic neural network | |
| Lee et al. | Expert-guided contrastive learning for video-text retrieval | |
| CN112749556B (en) | Multi-language model training method and device, storage medium and electronic equipment | |
| CN111866609B (en) | Method and apparatus for generating video | |
| CN113535946A (en) | Text identification method, device and equipment based on deep learning and storage medium |
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 |