CN110232158A - Burst occurred events of public safety detection method based on multi-modal data - Google Patents
Burst occurred events of public safety detection method based on multi-modal data Download PDFInfo
- Publication number
- CN110232158A CN110232158A CN201910371529.7A CN201910371529A CN110232158A CN 110232158 A CN110232158 A CN 110232158A CN 201910371529 A CN201910371529 A CN 201910371529A CN 110232158 A CN110232158 A CN 110232158A
- Authority
- CN
- China
- Prior art keywords
- data
- public safety
- sparse
- event
- text
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
Abstract
The present invention discloses a kind of burst occurred events of public safety detection method based on multi-modal data.This approach includes the following steps, step 1: establishing the vocabulary of burst occurred events of public safety;Step 2: crawler obtains social media data, obtains data attribute feature;Step 3: based on extract information, using it is multi-modal it is sparse from code machine learn uniform characteristics, every data is then structured as triple;Step 4: the method clustered using increment generates event cluster;Step 5: the burst occurred events of public safety vocabulary based on foundation screens obtained event cluster, extracts burst occurred events of public safety.Multi-modal information is considered in the method for the invention characteristic extraction procedure, social media data can be understood from multiple angles.In addition, social media data flow can be handled by carrying out event detection using incremental clustering algorithm, burst occurred events of public safety contained in social media is detected, provide certain support for further public opinion analysis.
Description
Technical field
The invention belongs to computer application technologies, are related to pattern-recognition, social network analysis, event detection, especially
It is a kind of burst occurred events of public safety detection method based on multi-modal data.
Background technique
China is in the critical period of social transformation, it is various it is systematic, acute structural imbalance, regional development is uneven
The conditions of the current stage of the social developments such as weighing apparatus, which are concentrated, to be occurred, and the security situation that this faces China is increasingly serious.The height of internet
Speed development, accelerates information flow speed, has widened communication space, enhance information interchange and interaction between netizen.Microblogging,
Push away special one kind social media have user volume is big, information publication rapidly, message is propagated and renewal speed is fast, communication space is wide
The features such as, people can pay close attention to the public accident of generation on this platform, comment on, exchange, so that event is fermented, carriage
Feelings range expands.Possible occurred events of public safety on network is detected early, is the pass of public administration for relevant department's Analysis of Policy Making
Key, and the basis supporting the national economic development, maintaining state security and social stability.Therefore, towards the burst of social platform
Occurred events of public safety detection is significant to assuring the safety for life and property of the people and maintaining social stability.
Traditional event detecting method generally only considers a kind of modal information of text, however can be comprising big in social media
The multi-modal informations such as spirogram piece, text.In view of this, the invention proposes a kind of public peaces of the burst based on multi-modal data
Total event detection method.The method of the invention considers that time, place, text, picture etc. are multi-modal in characteristic extraction procedure
Information, therefore social media data can be understood with carrying out various dimensions.In addition, carrying out event detection using incremental clustering algorithm
Ever-increasing social media data flow can be handled, using the burst public safety vocabulary of building, social matchmaker is effectively detected out
Burst occurred events of public safety contained in body provides certain support for further public opinion analysis.
Summary of the invention
The purpose of the present invention is examining by a kind of burst occurred events of public safety detection method based on multi-modal data
The multi-modal attribute that data are considered during survey event, social media data are understood from multiple angles.And by being built
Burst occurred events of public safety vocabulary detects the burst occurred events of public safety contained in social media, for further public opinion point
Analysis provides certain support.
In order to achieve the above objectives, the present invention the following technical schemes are provided:
Step 1: establishing the vocabulary of burst occurred events of public safety.
Step 2: crawler collect largely containing pairs of text, image social media data, extract every data when
Between, site attribute and text feature, characteristics of image.
Step 3: by WkIn Text eigenvector TkWith picture feature vector IkIt inputs sparse from code machine group by two
At it is multi-modal sparse from code machine, the uniform characteristics of text and image can be expressed simultaneously by obtaining, GkAs kth data
Content characteristic, every data is structured as<time, place, content>ternary feature group.
Step 4: the triple of every information being successively subjected to similarity-rough set with existing event cluster, if similar be higher than
The data is then incorporated to the highest event cluster of similarity by threshold value, otherwise based on the data, generates a new thing
Part cluster;
Step 5: obtained event cluster is screened, the keyword in the cluster is obtained, the burst public safety with construction
Vocabulary compares, if it is public that the event cluster is denoted as burst containing a large amount of burst public safety words in the keyword of the cluster
Security incident altogether.
Further, in step 1 specifically includes the following steps:
Step 1.1: construction burst occurred events of public safety dictionary.Master State Plan for Rapid Response to Public Emergencies will be public
The subevents such as security event classification such as bloods and droughts, the attack of terrorism altogether select Harbin Institute of Technology's social computing to grind with information retrieval center
" synonym woods (extended edition) " out is studied carefully to public safety word established in Master State Plan for Rapid Response to Public Emergencies
Word in table carries out synonymous extension, obtains the vocabulary of burst occurred events of public safety.
Further, in step 2 specifically includes the following steps:
Step 2.1: crawler collects the largely social media data W={ W containing pairs of text, image1,W2,…Wk,…,
WN, wherein N is acquired social media number of data.
Step 2.2: extracting WkTime, site attribute, be denoted as Ti respectivelyk、Lk。
Step 2.3: extracting WkText feature, characteristics of image, be denoted as T respectivelyk, Ik。
Further, in step 3 specifically includes the following steps:
Step 3.1: it is multi-modal it is sparse from code machine by two it is sparse formed from code machine, two hidden layers loss letter
The uniform characteristics to same data Chinese word information and pictorial information learnt under several constraints.If neuron activation functions
ForIndicate the output valve of i-th of neuron in l layers in a-th of sparse self-encoding encoder.
Step 3.2: setting text data set as T={ T1,T2,…Tk,…,TN, by text TkIt is input to as input data
First sparse from code machine, exports and isWherein sparse self-encoding encoder second layer neuron number is J.If image data
Integrate as I={ I1,I2,…Ik,…,IN, by picture IkSecond sparse from code machine, output is input to as input data
ForWherein sparse self-encoding encoder second layer neuron number is J.
Step 3.3: usingRespectively indicate kth group text, picture input Tk、IkIn the case of, hidden layer
(l=2) output valve of j-th of neuron.The then average activation value of two sparse j-th of neurons of hidden layer from code machine
It is respectively as follows:It is sparse from the sparse of code machine for guarantee two
Property, allows p '1,j=p '2,j=p, p are sparsity parameter, usually take 0.05, be j-th of neuron average activation value it is close
In p.
Step 3.4: to realize sparsity limitation, definition two is sparse to be respectively as follows: from the loss function of code machine Wherein
First item is the loss function of text/image reconstruct, Section 2 is penalty term, constrains p '1,j、p′2,jThe dramatically different feelings with p
Condition is to realize the sparse limitation to sparse self-encoding encoder.For guarantee text and image information relevance, then add relevance damage
Lose function
Step 3.5: in loss function JT、JI、JcorrConstraint under, it is more when iterative algorithm is restrained or after to maximum times
The sparse automatic coding machine of mode in same data text, pictorial information study to unified content feature G, G be J tie up to
Amount.
Step 3.6: every data is expressed as F by the information obtained according to above stepk=< Tik, Lk, Gk> triple
Feature vector.
Further, in step 4 specifically includes the following steps:
Step 4.1: setting existing event set { e1,e2,…,ei,…,em, each element e in event setiIn containing can generation
The triple of the n data of the table event, i.e.,
Step 4.2: one new data W of every additionk, the triple feature vector for representing the data is obtained by step 3
Fk, withTime, place, content similarity calculating are carried out, remembers FkWith eiSimilarity beTake maximum similarityk,iWith pre-set threshold value ratio
Compared with if more than threshold value, then the data being incorporated to the maximum e of similarityi, otherwise using the data as new events cluster, the cluster
It is denoted as em+1。
Further, in steps of 5 specifically includes the following steps:
Step 5.1: event cluster being filtered, specially deletion data bulk is lower than threshold θ1Cluster.
Step 5.2: extracting the keyword of event cluster, the vocabulary of itself and burst occurred events of public safety is compared, if keyword
In containing having more than θ2Word in a vocabulary then remembers that the event cluster is public safety dependent event cluster.
The beneficial effects of the present invention are: in view of the multi-modal attribute in network media data, divide from multiple angles
Analyse the data on network.In addition, ever-increasing social media data can be handled by carrying out event detection using incremental clustering algorithm
Stream, using the burst public safety vocabulary of building, is effectively detected out burst public safety thing contained in social media
Part provides certain support for further public opinion analysis.
Detailed description of the invention
In order to keep the purpose of the present invention, technical scheme and beneficial effects clearer, the present invention provides following attached drawing and carries out
Illustrate:
Fig. 1 is the method flow diagram that invention carries out burst occurred events of public safety detection based on multi-modal data.
Fig. 2 is the multi-modal sparse schematic diagram from code machine model in the present invention.
Fig. 3 is part burst occurred events of public safety vocabulary signal.
Specific embodiment
With reference to the accompanying drawings and examples to a kind of burst public safety thing based on multi-modal data provided by the invention
Part detection method is described further.
As shown in Figure 1, figure is the burst occurred events of public safety detection method flow chart based on multi-modal data.Specific packet
Include following steps.
S1: the vocabulary of burst occurred events of public safety is established.
S11: referring to Master State Plan for Rapid Response to Public Emergencies by occurred events of public safety classification as bloods and droughts,
The subevents such as the attack of terrorism." synonym woods (extended edition) " for selecting Harbin Institute of Technology's social computing and information retrieval center to develop
Synonymous extension is carried out to the word in public safety vocabulary established in Master State Plan for Rapid Response to Public Emergencies, is obtained
The vocabulary of burst occurred events of public safety is taken, part vocabulary schematic diagram is as shown in Fig. 2.
S2: collecting social media data, extract every data essential attribute, extract data and the text feature for being included,
Characteristics of image.
S21: crawler collects the largely social media data containing pairs of text, image.
S22: W is extractedkTime, site attribute, be denoted as Ti respectivelyk、Lk。
S23: W is extractedkText feature, characteristics of image, be denoted as T respectivelyk, Ik.The calculation of SIFT local feature can be used in picture
Method, Gabor local feature algorithm etc. extract feature, and CNN convolutional neural networks global characteristics algorithm can also be used and extract global spy
Sign.TF-IDF feature can be used in text.
S3: by WkIn Text eigenvector TkWith picture feature vector IkInput sparse is formed by two from code machine
It is multi-modal sparse from code machine, the unified content feature that can express text and image simultaneously is obtained, then by every data knot
Structure turns to<time, place, content>ternary feature group.
S31: it is multi-modal it is sparse from code machine by two it is sparse formed from code machine, two hidden layers are in loss function
The lower uniform characteristics to same data Chinese word information and pictorial information learnt of constraint.If neuron activation functions areIndicate the output valve of i-th of neuron in l layers in a-th of sparse self-encoding encoder, a takes 1,2 in this example.Activate letter
Number f (z) uses sigmoid function.
S32: text data set is T={ T1,T2,…Tk,…,TN, by text TkFirst is input to as input data
It is sparse from code machine, export and beIllustratively, text it is sparse from code machine input and output layer neuron number be 500,
Hidden layer neuron number is 1024.
Image data integrates as I={ I1,I2,…Ik,…,IN, by picture IkAs input data be input to second it is sparse
From in code machine, exports and beIllustratively, text it is sparse from code machine input and output layer neuron number be 500, imply
Layer neuron number is 1024.
S33: it usesRespectively indicate kth group text, picture input Tk、IkIn the case of, hidden layer (l=
2) output valve of j-th of neuron.The then average activation value difference of two sparse j-th of neurons of hidden layer from code machine
Are as follows:To guarantee two sparse sparsities from code machine, allow
p′1,j=p '2,j=p, p are sparsity parameter, usually take 0.05, are the average activation value of j-th of neuron close to p.
S34: to realize sparsity limitation, definition two is sparse to be respectively as follows: from the loss function of code machine Wherein
First item is the loss function of text/image reconstruct, Section 2 is penalty term, constrains p '1,j、p′2,jThe dramatically different feelings with p
Condition is to realize the sparse limitation to sparse self-encoding encoder.For guarantee text and image information relevance, then add relevance damage
Lose function
S35: in loss function JT、JI、JcorrConstraint under, it is multi-modal when iterative algorithm is restrained or after to maximum times
Sparse automatic coding machine in same data text, pictorial information study to unified content feature G, G be 1024 tie up to
Amount.
S36: every data is expressed as F by the information obtained according to above stepk=< Tik, Lk, Gk> triple feature
Vector.
S4: the triple of every information is successively subjected to similarity-rough set with existing event cluster, if similar be higher than threshold
Value, then be incorporated to the highest event cluster of similarity for the data, otherwise based on the data, generates a new event
Cluster.
S41: existing event set { e is set1,e2,…,ei,…,em, each element e in event setiIn containing capable of representing this
The triple of the n data of event, i.e.,
S42: one new data W of every additionk, the triple feature vector F for representing the data is obtained by step 3k, withTime, place, content similarity calculating are carried out, remembers FkWith eiSimilarity be
WhereinFor time similarity, place similarity, the mean value of content similarity.Time
The calculating of time inverse function fashion can be used in similarity, and place similarity is calculated using Haversine equation mode, interior
Hold similarity to calculate using COS distance.
Take maximum similarityk,iThe data is then incorporated to phase if more than threshold value with pre-set threshold value comparison
Like the maximum e of degreei, otherwise using the data as new events cluster, remember that the cluster is em+1。
Step 5: obtained event cluster is screened, the keyword in the cluster is obtained, the burst public safety with construction
Vocabulary compares, if it is public that the event cluster is denoted as burst containing a large amount of burst public safety words in the keyword of the cluster
Security incident altogether.
S51: being filtered event cluster, deletes data bulk and is lower than threshold θ1=(1*e-4) * N cluster.
S52: extracting 20 keywords to each event cluster, the vocabulary of itself and burst occurred events of public safety is compared, if closing
Containing the word having more than in 10 vocabularys in keyword, then remember that the event cluster is public safety dependent event cluster.
Above embodiments are only to illustrate the present invention, and not limit the technical scheme described by the invention.Therefore, one
The technical solution and its improvement for not departing from the spirit and scope of the present invention are cut, should all be covered in scope of the presently claimed invention
It is interior.
Claims (6)
1. a kind of burst occurred events of public safety detection method based on multi-modal data, includes the following steps:
Step 1: establishing the vocabulary of burst occurred events of public safety.
Step 2: crawler collects the largely social media data containing pairs of text, image, extracts time, the place of every data
Attribute and text feature, characteristics of image.
Step 3: by WkIn Text eigenvector TkWith picture feature vector IkInput by two it is sparse form from code machine it is more
Mode is sparse from code machine, obtains the uniform characteristics that can express text and image simultaneously, GkThe content of as kth data is special
Sign, is structured as<time, place, content>ternary feature group for every data.
Step 4: the triple of every data is successively subjected to similarity-rough set with existing event cluster, if similar be higher than threshold value,
The data is then incorporated to the highest event cluster of similarity, otherwise based on the data, generates a new event cluster.
Step 5: obtained event cluster is screened, the keyword in the cluster is obtained, the burst public safety vocabulary with construction
It compares, if the event cluster is denoted as the public peace of burst containing a large amount of burst public safety words in the keyword of the cluster
Total event.
2. event detecting method as claimed in claim 1, which is characterized in that specifically include in step 1:
Step 1.1: construction burst occurred events of public safety dictionary.Master State Plan for Rapid Response to Public Emergencies is by public safety
The subevents such as event category such as bloods and droughts, the attack of terrorism select Harbin Institute of Technology's social computing and information retrieval center to develop
" synonym woods (extended edition) " is in public safety vocabulary established in Master State Plan for Rapid Response to Public Emergencies
Word carries out synonymous extension, obtains the vocabulary of burst occurred events of public safety.
3. event detecting method as claimed in claim 1, which is characterized in that specifically include in step 2:
Step 2.1: crawler collects the largely social media data W={ W containing pairs of text, image1,W2,…Wk,…,WN,
Middle N is acquired social media number of data.
Step 2.2: extracting WkTime, site attribute, be denoted as Ti respectivelyk、Lk。
Step 2.3: extracting WkText feature, characteristics of image, be denoted as T respectivelyk, Ik。
4. event detecting method as claimed in claim 1, which is characterized in that specifically include in step 3:
Step 3.1: it is multi-modal it is sparse from code machine by two it is sparse formed from code machine, pact of two hidden layers in loss function
The uniform characteristics to same data Chinese word information and pictorial information learnt under beam.If neuron activation functions are f (z),Indicate the output valve of i-th of neuron in l layers in a-th of sparse self-encoding encoder.
Step 3.2: setting text data set as T={ T1,T2,…Tk,…,TN, by text TkFirst is input to as input data
It is a sparse from code machine, it exports and isWherein sparse self-encoding encoder second layer neuron number is J.If image data integrates as I
={ I1,I2,…Ik,…,IN, by picture IkAs input data be input to second it is sparse from code machine, export and beIts
In sparse self-encoding encoder second layer neuron number be J.
Step 3.3: usingRespectively indicate kth group text, picture input Tk、IkIn the case of, hidden layer (l=2)
The output valve of j-th of neuron.Then the average activation value of two sparse j-th of neurons of hidden layer from code machine is respectively as follows: To guarantee two sparse sparsities from code machine, p ' is allowed1,j=
p′2,j=p, p are sparsity parameter, usually take 0.05, are the average activation value of j-th of neuron close to p.
Step 3.4: to realize sparsity limitation, definition two is sparse to be respectively as follows: from the loss function of code machine Wherein
First item is the loss function of text/image reconstruct, Section 2 is penalty term, constrains p '1,j、p′2,jThe dramatically different situation with p
To realize the sparse limitation to sparse self-encoding encoder.For guarantee text and image information relevance, then add relevance loss
Function
Step 3.5: in loss function JT、JI、JcorrConstraint under, it is multi-modal dilute when iterative algorithm is restrained or after to maximum times
Automatic coding machine is dredged to the text in same data, pictorial information study to unified content feature G, G is J dimensional vector.
Step 3.6: every data is expressed as F by the information obtained according to above stepk=< Tik, Lk, Gk> triple feature to
Amount.
5. event detecting method as claimed in claim 1, which is characterized in that specifically include in step 4:
Step 4.1: setting existing event set { e1,e2,…,ei,…,em, each element e in event setiIn containing capable of representing this
The triple of the n data of event, i.e.,
Step 4.2: one new data W of every additionk, the triple feature vector F for representing the data is obtained by step 3k, withTime, place, content similarity calculating are carried out, remembers FkWith eiSimilarity be Take maximum similarityk,iWith pre-set threshold value comparison,
If more than threshold value, then the data is incorporated to the maximum e of similarityi, otherwise using the data as new events cluster, which is denoted as
em+1。
6. event detecting method as claimed in claim 1, which is characterized in that specifically include in steps of 5:
Step 5.1: event cluster being filtered, specially deletion data bulk is lower than threshold θ1Cluster.
Step 5.2: extracting the keyword of event cluster, the vocabulary of itself and burst occurred events of public safety is compared, if containing in keyword
Have more than θ2Word in a vocabulary then remembers that the event cluster is public safety dependent event cluster.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910371529.7A CN110232158A (en) | 2019-05-06 | 2019-05-06 | Burst occurred events of public safety detection method based on multi-modal data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910371529.7A CN110232158A (en) | 2019-05-06 | 2019-05-06 | Burst occurred events of public safety detection method based on multi-modal data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110232158A true CN110232158A (en) | 2019-09-13 |
Family
ID=67861116
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910371529.7A Pending CN110232158A (en) | 2019-05-06 | 2019-05-06 | Burst occurred events of public safety detection method based on multi-modal data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232158A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259851A (en) * | 2020-01-23 | 2020-06-09 | 清华大学 | Multi-mode event detection method and device |
CN111651559A (en) * | 2020-05-29 | 2020-09-11 | 辽宁工程技术大学 | Social network user relationship extraction method based on event extraction |
CN111932427A (en) * | 2020-09-24 | 2020-11-13 | 北京泰策科技有限公司 | Method and system for detecting emergent public security incident based on multi-mode data |
CN112527960A (en) * | 2020-12-17 | 2021-03-19 | 华东师范大学 | Emergency detection method based on keyword clustering |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091054A (en) * | 2014-06-26 | 2014-10-08 | 中国科学院自动化研究所 | Mass disturbance warning method and system applied to short texts |
CN104765878A (en) * | 2015-04-27 | 2015-07-08 | 合肥工业大学 | Sparse coding algorithm suitable for multi-modal information and application thereof |
CN106557558A (en) * | 2016-11-09 | 2017-04-05 | 中国工商银行股份有限公司 | A kind of data analysing method and device |
CN107229712A (en) * | 2017-05-27 | 2017-10-03 | 中南大学 | A kind of space-time clustering method towards occurred events of public safety acquisition of information |
CN107291886A (en) * | 2017-06-21 | 2017-10-24 | 广西科技大学 | A kind of microblog topic detecting method and system based on incremental clustering algorithm |
CN107992904A (en) * | 2017-12-22 | 2018-05-04 | 重庆邮电大学 | Forest Ecology man-machine interaction method based on Multi-source Information Fusion |
CN108182441A (en) * | 2017-12-29 | 2018-06-19 | 华中科技大学 | Parallel multichannel convolutive neural network, construction method and image characteristic extracting method |
EP3364364A1 (en) * | 2017-02-21 | 2018-08-22 | Privately Sàrl | Method to detect incidents from social network use |
CN108537233A (en) * | 2018-03-15 | 2018-09-14 | 南京师范大学 | A kind of pathology brain image sorting technique based on the sparse self-encoding encoder of depth stack |
CN108549647A (en) * | 2018-01-17 | 2018-09-18 | 中移在线服务有限公司 | The method without accident in mark language material active predicting movement customer service field is realized based on SinglePass algorithms |
US20180314982A1 (en) * | 2017-04-28 | 2018-11-01 | At&T Intellectual Property I, L.P. | Bridging heterogeneous domains with parallel transport and sparse coding for machine learning models |
CN109145974A (en) * | 2018-08-13 | 2019-01-04 | 广东工业大学 | One kind being based on the matched multi-level image Feature fusion of picture and text |
-
2019
- 2019-05-06 CN CN201910371529.7A patent/CN110232158A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091054A (en) * | 2014-06-26 | 2014-10-08 | 中国科学院自动化研究所 | Mass disturbance warning method and system applied to short texts |
CN104765878A (en) * | 2015-04-27 | 2015-07-08 | 合肥工业大学 | Sparse coding algorithm suitable for multi-modal information and application thereof |
CN106557558A (en) * | 2016-11-09 | 2017-04-05 | 中国工商银行股份有限公司 | A kind of data analysing method and device |
EP3364364A1 (en) * | 2017-02-21 | 2018-08-22 | Privately Sàrl | Method to detect incidents from social network use |
US20180314982A1 (en) * | 2017-04-28 | 2018-11-01 | At&T Intellectual Property I, L.P. | Bridging heterogeneous domains with parallel transport and sparse coding for machine learning models |
CN107229712A (en) * | 2017-05-27 | 2017-10-03 | 中南大学 | A kind of space-time clustering method towards occurred events of public safety acquisition of information |
CN107291886A (en) * | 2017-06-21 | 2017-10-24 | 广西科技大学 | A kind of microblog topic detecting method and system based on incremental clustering algorithm |
CN107992904A (en) * | 2017-12-22 | 2018-05-04 | 重庆邮电大学 | Forest Ecology man-machine interaction method based on Multi-source Information Fusion |
CN108182441A (en) * | 2017-12-29 | 2018-06-19 | 华中科技大学 | Parallel multichannel convolutive neural network, construction method and image characteristic extracting method |
CN108549647A (en) * | 2018-01-17 | 2018-09-18 | 中移在线服务有限公司 | The method without accident in mark language material active predicting movement customer service field is realized based on SinglePass algorithms |
CN108537233A (en) * | 2018-03-15 | 2018-09-14 | 南京师范大学 | A kind of pathology brain image sorting technique based on the sparse self-encoding encoder of depth stack |
CN109145974A (en) * | 2018-08-13 | 2019-01-04 | 广东工业大学 | One kind being based on the matched multi-level image Feature fusion of picture and text |
Non-Patent Citations (2)
Title |
---|
YAQING WANG ET AL: "《KDD "18: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining》", 19 July 2018 * |
钱胜胜等: "多媒体社会事件分析的研究与展望", 《南京信息工程大学学报(自然科学版)》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259851A (en) * | 2020-01-23 | 2020-06-09 | 清华大学 | Multi-mode event detection method and device |
CN111651559A (en) * | 2020-05-29 | 2020-09-11 | 辽宁工程技术大学 | Social network user relationship extraction method based on event extraction |
CN111651559B (en) * | 2020-05-29 | 2023-05-26 | 辽宁工程技术大学 | Social network user relation extraction method based on event extraction |
CN111932427A (en) * | 2020-09-24 | 2020-11-13 | 北京泰策科技有限公司 | Method and system for detecting emergent public security incident based on multi-mode data |
CN111932427B (en) * | 2020-09-24 | 2021-01-26 | 北京泰策科技有限公司 | Method and system for detecting emergent public security incident based on multi-mode data |
CN112527960A (en) * | 2020-12-17 | 2021-03-19 | 华东师范大学 | Emergency detection method based on keyword clustering |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110232158A (en) | Burst occurred events of public safety detection method based on multi-modal data | |
CN110598005B (en) | Public safety event-oriented multi-source heterogeneous data knowledge graph construction method | |
CN108399230A (en) | A kind of Chinese financial and economic news file classification method based on convolutional neural networks | |
CN111159395A (en) | Chart neural network-based rumor standpoint detection method and device and electronic equipment | |
CN108492200A (en) | A kind of user property estimating method and device based on convolutional neural networks | |
CN105955951B (en) | A kind of method and device of message screening | |
CN111143576A (en) | Event-oriented dynamic knowledge graph construction method and device | |
CN110555455A (en) | Online transaction fraud detection method based on entity relationship | |
CN107070943A (en) | Industry internet intrusion detection method based on traffic characteristic figure and perception Hash | |
CN112001565A (en) | Earthquake disaster loss prediction and evaluation method and system based on Softmax regression model | |
CN109471944A (en) | Training method, device and the readable storage medium storing program for executing of textual classification model | |
CN109657947A (en) | A kind of method for detecting abnormality towards enterprises ' industry classification | |
WO2014094332A1 (en) | Method for creating knowledge base engine for emergency management of sudden event and method for querying in knowledge base engine | |
Fan et al. | A system analytics framework for detecting infrastructure-related topics in disasters using social sensing | |
CN109359137B (en) | User growth portrait construction method based on feature screening and semi-supervised learning | |
CN111460818A (en) | Web page text classification method based on enhanced capsule network and storage medium | |
Chen et al. | A deep learning method for judicial decision support | |
CN110019820A (en) | Main suit and present illness history symptom Timing Coincidence Detection method in a kind of case history | |
CN109857869A (en) | A kind of hot topic prediction technique based on Ap increment cluster and network primitive | |
CN112131453A (en) | Method, device and storage medium for detecting network bad short text based on BERT | |
CN115017887A (en) | Chinese rumor detection method based on graph convolution | |
CN112084240B (en) | Intelligent identification and linkage treatment method and system for group renting | |
CN113806528A (en) | Topic detection method and device based on BERT model and storage medium | |
CN108519993A (en) | The social networks focus incident detection method calculated based on multiple data stream | |
CN115277159B (en) | Industrial Internet security situation assessment method based on improved random forest |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190913 |
|
WD01 | Invention patent application deemed withdrawn after publication |