SlideShare a Scribd company logo
1 of 16
GRÉGOIRE BUREL, HASSAN SAIF, HARITH ALANI
Knowledge Media Institute, The Open University, Milton Keynes, UK.
ISWC’17, Vienna, Austria.
21-25 October 2017.
Semantic Wide and Deep Learning for
Detecting Crisis-Information
Categories on Social Media
Event Detection and Crisis Situations
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
2
Event detection is “the task of automatically
identifying certain clues in texts that denote a
specific event type or theme”.
- Help identifying/responding to events.
- Organise relevant information during
crises.
Twitter:
~200 million active users.
~400 million tweets a day.
Twitter usage during crises:
1. During the 2011 Japan earthquake, 177
million tweets related to the event were
sent in one day.
2. The news about the Boston bombings
first appeared on Twitter.
Crisis-Related Event Detection Tasks
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
Publications
3
Crisis-related event detection is often divided into three main tasks (Olteanu et al.
2015):
Crisis Related /
Unrelated
Crisis
Type
Information
Categories
Task 1
Identify the
different types
of crises the
message is
related to.
Differentiate the
type of information
contained in the
message.
e.g., shooting,
explosion, building
collapse, fires, floods,
meteorite fall, etc.
e.g., affected individuals,
infrastructures and
utilities, donations and
volunteer, caution and
advice, etc.
Granularity
Differentiate the
posts that are
related or unrelated
to crises.
Task 2 Task 3
Crisis-Related Event Detection Tasks
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
Publications
4
Crisis-related event detection is often divided into three main tasks (Olteanu et al.
2015):
Crisis Related /
Unrelated
Crisis
Type
Information
Categories
Task 1
Identify the
different types
of crises the
message is
related to.
Differentiate the
type of information
contained in the
message.
e.g., shooting,
explosion, building
collapse, fires, floods,
meteorite fall, etc.
e.g., affected individuals,
infrastructures and
utilities, donations and
volunteer, caution and
advice, etc.
Granularity
Differentiate the
posts that are
related or unrelated
to crises.
Task 2 Task 3
‘Traditional’ ML vs. Deep Learning
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
5
Deep Learning
- Artificial neural networks.
- Minimum feature engineering
- Word embeddings (Bengio et
al., 2013).
‘Traditional’ ML
- Standard classifiers (e.g., SVM,
J48…).
- Feature engineering (e.g.,
lemmatisation, TF-IDF…).
- Bag of words.
Text vs. Semantics – Document Contextualisation
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
6
Obama attends vigil for Boston Marathon bombing victims
Politician /
Person
Sports Event / Social
Event / Event
Disaster / Event
Incorporating Semantics into ML Classification Methods for
contextualising documents:
- Approach 1: Traditional ML Classifiers
- Approach 2: Deep Learning
CNN for Sentence Classification (Kim et al., 2014)
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
7
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
8
CNN for Sentence Classification Dual-CNN (Semantic Channel)
CNN for Sentence Classification (Kim et al., 2014)
+ Competitive results for text classification tasks.
+ No or Little Feature Engineering required.
+ Relatively good at taking local textual relations
within short documents.
- No ‘native’ semantic context.
Dual-CNN (Burel et al., 2017)
+ Text CNN
+ Aligned Semantic channel
- Concept extraction.
- Semantics vocabulary (4000) <<
Words vocabulary (60000)
Wide and Deep Learning (Cheng et al., 2016)
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
9
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
10
Wide and Deep Learning Sem-CNN (W-D-CNN)
Wide and Deep Learning (Cheng et al., 2016)
+ Efficiently Deal with ‘sparse’ and ‘dense’
inputs.
- Not very efficient for modelling text relations.
- No ‘native’ semantic context.
Sem-CNN (W-D-CNN)
+ Text CNN / Wide and Deep Models
+ Deep Shallow Word Embeddings
+ Wide Deep Semantics
- Requires semantic extraction.
Wide and Deep Semantic CNN (Sem-CNN)
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
11
Sem-CNN – Experimental Setup
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
Dataset - T26 (28,000 annotated tweets)
- 12 Crisis types (shooting, explosion, building collapse, fires, floods,
meteorite fall, haze, bombing, typhoon, crash, earthquake, and
derailment).
- 6 Information categories (affected individuals, infrastructures and
utilities, donations and volunteer, caution and advice, sympathy and
emotional support, and other useful information)
Semantic Extraction -
- Extracted Entities/Concepts: 65% dataset coverage.
Concept Vectors Initialisation
- Concept Labels: Obama → dbo:Obama
- Concept Abstracts: Obama → dbo:Obama → ‘Barack Hussein
Obama II; born August 4, 1961) is an American politician…’
Sem-CNN – Experimental Setup
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
13
Dataset versions
- Full Dataset: 28,000 tweets.
- Balanced Dataset (BD1): 9100 tweets (32.6%).
- Semantically Balanced Dataset (>2 entities/concepts, BD2): 1194 tweets
(4.3%).
Baselines
- SVM (TF-IDF): Linear SVM using the words’ TF-IDF vectors extracted
from our dataset.
- SVM (Word2Vec): Linear SVM using the Google pre-trained 300-
dimensional word embeddings.
Evaluation
- 5-folds cross validation.
- Sem-CNN: 300-dim embeddings, Fn = 128 convolutional filter of sizes
Fs = [3,4,5], 0.5 dropout and ADAM.
- Evaluation Measures: P, R and F1.
?
Results
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
14
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
Publications
15
+ - Sem-CNN significantly outperforms the baselines
(p < 0.001)
- More semantics leads to better results.
- Sem-CNN appears to perform better than Dual-
CNN (up to +4% F1) with F1 up to 64%.
- Abstract outperform the Concept vectors but it is
not always significant (i.e., on the full dataset).
- Consider more complex deep learning models
such as Recurrent Neural Networks (RNN) or
Hierarchical Attention Networks (HAN).
- Initialise with different embeddings (e.g., Twitter)
and perform parameter optimisation.
- Investigate other methods for integrating
semantics (e.g., extended concept graphs).
-
Results and Future Work
CREES
Crisis Event Extraction
Service
?
Questions
@
Email: g.burel@open.ac.uk
Twitter: @evhart
CREES: https://github.com/evhart/crees
COMRADES: http://comrades-project.eu
Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
16

More Related Content

What's hot

FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MINING
FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MININGFAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MINING
FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MININGijnlc
 
The mimetic virus A vector for cyberterrorism
The mimetic virus A vector for cyberterrorismThe mimetic virus A vector for cyberterrorism
The mimetic virus A vector for cyberterrorismNicholas Ayres
 
A potency relation for worms and next generation attack tools
A potency relation for worms and next generation attack toolsA potency relation for worms and next generation attack tools
A potency relation for worms and next generation attack toolsUltraUploader
 
Expelling Information of Events from Critical Public Space using Social Senso...
Expelling Information of Events from Critical Public Space using Social Senso...Expelling Information of Events from Critical Public Space using Social Senso...
Expelling Information of Events from Critical Public Space using Social Senso...ijtsrd
 
友人関係と感染症伝搬をネットワークで理解する
友人関係と感染症伝搬をネットワークで理解する友人関係と感染症伝搬をネットワークで理解する
友人関係と感染症伝搬をネットワークで理解するtm1966
 
Bayesian Model Fusion for Forecasting Civil Unrest
Bayesian Model Fusion for Forecasting Civil UnrestBayesian Model Fusion for Forecasting Civil Unrest
Bayesian Model Fusion for Forecasting Civil UnrestParang Saraf
 
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatraman
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatramanOdsc 2018 detection_classification_of_fake_news_using_cnn_venkatraman
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatramanvenkatramanJ4
 
Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets
 Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets
Twitter Is the Megaphone of Cross-platform Messaging on the White HelmetsSameera Horawalavithana
 
Unfollowing on twitter
Unfollowing on twitterUnfollowing on twitter
Unfollowing on twittermor
 
Stanford Info Seminar: Unfollowing and Emotion on Twitter
Stanford Info Seminar: Unfollowing and Emotion on TwitterStanford Info Seminar: Unfollowing and Emotion on Twitter
Stanford Info Seminar: Unfollowing and Emotion on Twittermor
 
IRJET- Fake News Detection
IRJET- Fake News DetectionIRJET- Fake News Detection
IRJET- Fake News DetectionIRJET Journal
 
Hybrid sentiment and network analysis of social opinion polarization icoict
Hybrid sentiment and network analysis of social opinion polarization   icoictHybrid sentiment and network analysis of social opinion polarization   icoict
Hybrid sentiment and network analysis of social opinion polarization icoictAndry Alamsyah
 
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...COMRADES project
 
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...IRJET- Identification of Prevalent News from Twitter and Traditional Media us...
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...IRJET Journal
 
IRJET - Fake News Detection using Machine Learning
IRJET -  	  Fake News Detection using Machine LearningIRJET -  	  Fake News Detection using Machine Learning
IRJET - Fake News Detection using Machine LearningIRJET Journal
 
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...caijjournal
 
Prediction of Reaction towards Textual Posts in Social Networks
Prediction of Reaction towards Textual Posts in Social NetworksPrediction of Reaction towards Textual Posts in Social Networks
Prediction of Reaction towards Textual Posts in Social NetworksMohamed El-Geish
 
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...Parang Saraf
 

What's hot (20)

FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MINING
FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MININGFAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MINING
FAKE NEWS DETECTION WITH SEMANTIC FEATURES AND TEXT MINING
 
The mimetic virus A vector for cyberterrorism
The mimetic virus A vector for cyberterrorismThe mimetic virus A vector for cyberterrorism
The mimetic virus A vector for cyberterrorism
 
A potency relation for worms and next generation attack tools
A potency relation for worms and next generation attack toolsA potency relation for worms and next generation attack tools
A potency relation for worms and next generation attack tools
 
Expelling Information of Events from Critical Public Space using Social Senso...
Expelling Information of Events from Critical Public Space using Social Senso...Expelling Information of Events from Critical Public Space using Social Senso...
Expelling Information of Events from Critical Public Space using Social Senso...
 
友人関係と感染症伝搬をネットワークで理解する
友人関係と感染症伝搬をネットワークで理解する友人関係と感染症伝搬をネットワークで理解する
友人関係と感染症伝搬をネットワークで理解する
 
Bayesian Model Fusion for Forecasting Civil Unrest
Bayesian Model Fusion for Forecasting Civil UnrestBayesian Model Fusion for Forecasting Civil Unrest
Bayesian Model Fusion for Forecasting Civil Unrest
 
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatraman
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatramanOdsc 2018 detection_classification_of_fake_news_using_cnn_venkatraman
Odsc 2018 detection_classification_of_fake_news_using_cnn_venkatraman
 
Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets
 Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets
Twitter Is the Megaphone of Cross-platform Messaging on the White Helmets
 
Unfollowing on twitter
Unfollowing on twitterUnfollowing on twitter
Unfollowing on twitter
 
Stanford Info Seminar: Unfollowing and Emotion on Twitter
Stanford Info Seminar: Unfollowing and Emotion on TwitterStanford Info Seminar: Unfollowing and Emotion on Twitter
Stanford Info Seminar: Unfollowing and Emotion on Twitter
 
IRJET- Fake News Detection
IRJET- Fake News DetectionIRJET- Fake News Detection
IRJET- Fake News Detection
 
Hybrid sentiment and network analysis of social opinion polarization icoict
Hybrid sentiment and network analysis of social opinion polarization   icoictHybrid sentiment and network analysis of social opinion polarization   icoict
Hybrid sentiment and network analysis of social opinion polarization icoict
 
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...
SemEval-2017 Task 8: RumourEval: Determining rumour veracity and support for ...
 
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...IRJET- Identification of Prevalent News from Twitter and Traditional Media us...
IRJET- Identification of Prevalent News from Twitter and Traditional Media us...
 
IRJET - Fake News Detection using Machine Learning
IRJET -  	  Fake News Detection using Machine LearningIRJET -  	  Fake News Detection using Machine Learning
IRJET - Fake News Detection using Machine Learning
 
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...
A RELIABLE ARTIFICIAL INTELLIGENCE MODEL FOR FALSE NEWS DETECTION MADE BY PUB...
 
DIE 20130724
DIE 20130724DIE 20130724
DIE 20130724
 
Prediction of Reaction towards Textual Posts in Social Networks
Prediction of Reaction towards Textual Posts in Social NetworksPrediction of Reaction towards Textual Posts in Social Networks
Prediction of Reaction towards Textual Posts in Social Networks
 
Tcat
TcatTcat
Tcat
 
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...
EMBERS at 4 years: Experiences operating an Open Source Indicators Forecastin...
 

Similar to Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media

Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...
Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...
Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...Gregoire Burel
 
Helping Crisis Responders Find the Informative Needle in the Tweet Haystack
Helping Crisis Responders Find the Informative Needle in the Tweet HaystackHelping Crisis Responders Find the Informative Needle in the Tweet Haystack
Helping Crisis Responders Find the Informative Needle in the Tweet HaystackCOMRADES project
 
Information Contagion through Social Media: Towards a Realistic Model of the ...
Information Contagion through Social Media: Towards a Realistic Model of the ...Information Contagion through Social Media: Towards a Realistic Model of the ...
Information Contagion through Social Media: Towards a Realistic Model of the ...Axel Bruns
 
Multimodal Combination.pdf
Multimodal Combination.pdfMultimodal Combination.pdf
Multimodal Combination.pdfclientmentailai
 
Detection and resolution of rumours in social media
Detection and resolution of rumours in social mediaDetection and resolution of rumours in social media
Detection and resolution of rumours in social mediaObedullahFahad
 
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...Automated Emerging Cyber Threat Identification and Profiling Based on Natural...
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...Shakas Technologies
 
Enhancing prediction of user stance for social networks rumors
Enhancing prediction of user stance for social networks rumorsEnhancing prediction of user stance for social networks rumors
Enhancing prediction of user stance for social networks rumorsIJECEIAES
 
Situation Awareness with IT for Regional Disaster Mangement Workshop
Situation Awareness with IT for Regional Disaster Mangement WorkshopSituation Awareness with IT for Regional Disaster Mangement Workshop
Situation Awareness with IT for Regional Disaster Mangement WorkshopChamindra de Silva
 
Classification of Disastrous Tweets on Twitter using BERT Model
Classification of Disastrous Tweets on Twitter using BERT ModelClassification of Disastrous Tweets on Twitter using BERT Model
Classification of Disastrous Tweets on Twitter using BERT ModelIRJET Journal
 
Disaster data informatics for situation awareness
Disaster data informatics for situation awareness Disaster data informatics for situation awareness
Disaster data informatics for situation awareness Ashutosh Jadhav
 
Crisis Information Processing - with the power of A.I.
Crisis Information Processing - with the power of A.I.Crisis Information Processing - with the power of A.I.
Crisis Information Processing - with the power of A.I.The Open University
 
Enhancing Soft Power: using cyberspace to enhance Soft Power
Enhancing Soft Power: using cyberspace to enhance Soft PowerEnhancing Soft Power: using cyberspace to enhance Soft Power
Enhancing Soft Power: using cyberspace to enhance Soft PowerAmit Sheth
 
An Online Social Network for Emergency Management
An Online Social Network for Emergency ManagementAn Online Social Network for Emergency Management
An Online Social Network for Emergency Managementguestc9f21b
 
DAY 2 Economy, Politics ETC....pptx
DAY 2 Economy, Politics ETC....pptxDAY 2 Economy, Politics ETC....pptx
DAY 2 Economy, Politics ETC....pptxLOUIEDELMUNDO3
 
Sputnik Education Reform Movement
Sputnik Education Reform MovementSputnik Education Reform Movement
Sputnik Education Reform MovementJennifer Reither
 
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...Connie White
 
Acg Terr Sand2004 2130w
Acg Terr Sand2004 2130wAcg Terr Sand2004 2130w
Acg Terr Sand2004 2130wNKHAYDEN
 
MASINT and Global War on Terror
MASINT and Global War on TerrorMASINT and Global War on Terror
MASINT and Global War on TerrorTpeisi Nesby
 

Similar to Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media (20)

Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...
Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...
Crisis Event Extraction Service (CREES) – Automatic Detection and Classificat...
 
Helping Crisis Responders Find the Informative Needle in the Tweet Haystack
Helping Crisis Responders Find the Informative Needle in the Tweet HaystackHelping Crisis Responders Find the Informative Needle in the Tweet Haystack
Helping Crisis Responders Find the Informative Needle in the Tweet Haystack
 
Information Contagion through Social Media: Towards a Realistic Model of the ...
Information Contagion through Social Media: Towards a Realistic Model of the ...Information Contagion through Social Media: Towards a Realistic Model of the ...
Information Contagion through Social Media: Towards a Realistic Model of the ...
 
Multimodal Combination.pdf
Multimodal Combination.pdfMultimodal Combination.pdf
Multimodal Combination.pdf
 
Detection and resolution of rumours in social media
Detection and resolution of rumours in social mediaDetection and resolution of rumours in social media
Detection and resolution of rumours in social media
 
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...Automated Emerging Cyber Threat Identification and Profiling Based on Natural...
Automated Emerging Cyber Threat Identification and Profiling Based on Natural...
 
Enhancing prediction of user stance for social networks rumors
Enhancing prediction of user stance for social networks rumorsEnhancing prediction of user stance for social networks rumors
Enhancing prediction of user stance for social networks rumors
 
Akram.pptx
Akram.pptxAkram.pptx
Akram.pptx
 
Situation Awareness with IT for Regional Disaster Mangement Workshop
Situation Awareness with IT for Regional Disaster Mangement WorkshopSituation Awareness with IT for Regional Disaster Mangement Workshop
Situation Awareness with IT for Regional Disaster Mangement Workshop
 
Classification of Disastrous Tweets on Twitter using BERT Model
Classification of Disastrous Tweets on Twitter using BERT ModelClassification of Disastrous Tweets on Twitter using BERT Model
Classification of Disastrous Tweets on Twitter using BERT Model
 
Disaster data informatics for situation awareness
Disaster data informatics for situation awareness Disaster data informatics for situation awareness
Disaster data informatics for situation awareness
 
Crisis Information Processing - with the power of A.I.
Crisis Information Processing - with the power of A.I.Crisis Information Processing - with the power of A.I.
Crisis Information Processing - with the power of A.I.
 
Enhancing Soft Power: using cyberspace to enhance Soft Power
Enhancing Soft Power: using cyberspace to enhance Soft PowerEnhancing Soft Power: using cyberspace to enhance Soft Power
Enhancing Soft Power: using cyberspace to enhance Soft Power
 
s00146-014-0549-4.pdf
s00146-014-0549-4.pdfs00146-014-0549-4.pdf
s00146-014-0549-4.pdf
 
An Online Social Network for Emergency Management
An Online Social Network for Emergency ManagementAn Online Social Network for Emergency Management
An Online Social Network for Emergency Management
 
DAY 2 Economy, Politics ETC....pptx
DAY 2 Economy, Politics ETC....pptxDAY 2 Economy, Politics ETC....pptx
DAY 2 Economy, Politics ETC....pptx
 
Sputnik Education Reform Movement
Sputnik Education Reform MovementSputnik Education Reform Movement
Sputnik Education Reform Movement
 
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...
A Framework to Identify Best Practices: Social Media and Web 2.0 Technologies...
 
Acg Terr Sand2004 2130w
Acg Terr Sand2004 2130wAcg Terr Sand2004 2130w
Acg Terr Sand2004 2130w
 
MASINT and Global War on Terror
MASINT and Global War on TerrorMASINT and Global War on Terror
MASINT and Global War on Terror
 

More from Gregoire Burel

Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...
Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...
Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...Gregoire Burel
 
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...Gregoire Burel
 
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...Gregoire Burel
 
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...Gregoire Burel
 
DoRES — A Three-tier Ontology for Modelling Crises in the Digital Age
DoRES — A Three-tier Ontology for Modelling Crises in the Digital AgeDoRES — A Three-tier Ontology for Modelling Crises in the Digital Age
DoRES — A Three-tier Ontology for Modelling Crises in the Digital AgeGregoire Burel
 
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUse
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUseMonitoring, Discussing and Publishing Energy Consumption Data using EnergyUse
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUseGregoire Burel
 
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...Gregoire Burel
 
Structural Normalisation Methods for Improving Best Answer Identification in ...
Structural Normalisation Methods for Improving Best Answer Identification in ...Structural Normalisation Methods for Improving Best Answer Identification in ...
Structural Normalisation Methods for Improving Best Answer Identification in ...Gregoire Burel
 
Quantising Contribution Effort in Online Communities
Quantising Contribution Effort in Online CommunitiesQuantising Contribution Effort in Online Communities
Quantising Contribution Effort in Online CommunitiesGregoire Burel
 
Predicting Answering Behaviour in Online Question Answering Communities
Predicting Answering Behaviour in Online Question Answering CommunitiesPredicting Answering Behaviour in Online Question Answering Communities
Predicting Answering Behaviour in Online Question Answering CommunitiesGregoire Burel
 
A Question of Complexity - Measuring the Maturity of Online Enquiry Communities
A Question of Complexity - Measuring the Maturity of Online Enquiry CommunitiesA Question of Complexity - Measuring the Maturity of Online Enquiry Communities
A Question of Complexity - Measuring the Maturity of Online Enquiry CommunitiesGregoire Burel
 
Automatic Identification of Best Answers in Online Enquiry Communities
Automatic Identification of Best Answers in Online Enquiry CommunitiesAutomatic Identification of Best Answers in Online Enquiry Communities
Automatic Identification of Best Answers in Online Enquiry CommunitiesGregoire Burel
 
Asterid: Linked Data Asterisms
Asterid: Linked Data AsterismsAsterid: Linked Data Asterisms
Asterid: Linked Data AsterismsGregoire Burel
 
Veracity ESWC2010 Poster
Veracity ESWC2010 PosterVeracity ESWC2010 Poster
Veracity ESWC2010 PosterGregoire Burel
 
Attention-Streams Recommendations
Attention-Streams RecommendationsAttention-Streams Recommendations
Attention-Streams RecommendationsGregoire Burel
 
Sparks O3 Browser: Augmenting the Web with Semantic Overlays
Sparks O3 Browser: Augmenting the Web with Semantic OverlaysSparks O3 Browser: Augmenting the Web with Semantic Overlays
Sparks O3 Browser: Augmenting the Web with Semantic OverlaysGregoire Burel
 

More from Gregoire Burel (16)

Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...
Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...
Monitoring, Understanding, and Influencing the Co-Spread of COVID-19 Misinfor...
 
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...
Monitoring, Understanding and Influencing the Co-Spread of COVID-19 Misinform...
 
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...
Monitoring and Understanding the Co-Spread of COVID-19 Misinformation and Fac...
 
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...
Co-Spread of Misinformation and Fact-Checking Content during the Covid-19 Pan...
 
DoRES — A Three-tier Ontology for Modelling Crises in the Digital Age
DoRES — A Three-tier Ontology for Modelling Crises in the Digital AgeDoRES — A Three-tier Ontology for Modelling Crises in the Digital Age
DoRES — A Three-tier Ontology for Modelling Crises in the Digital Age
 
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUse
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUseMonitoring, Discussing and Publishing Energy Consumption Data using EnergyUse
Monitoring, Discussing and Publishing Energy Consumption Data using EnergyUse
 
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...
EnergyUse - A Collective Semantic Platform for Monitoring and Discussing Ener...
 
Structural Normalisation Methods for Improving Best Answer Identification in ...
Structural Normalisation Methods for Improving Best Answer Identification in ...Structural Normalisation Methods for Improving Best Answer Identification in ...
Structural Normalisation Methods for Improving Best Answer Identification in ...
 
Quantising Contribution Effort in Online Communities
Quantising Contribution Effort in Online CommunitiesQuantising Contribution Effort in Online Communities
Quantising Contribution Effort in Online Communities
 
Predicting Answering Behaviour in Online Question Answering Communities
Predicting Answering Behaviour in Online Question Answering CommunitiesPredicting Answering Behaviour in Online Question Answering Communities
Predicting Answering Behaviour in Online Question Answering Communities
 
A Question of Complexity - Measuring the Maturity of Online Enquiry Communities
A Question of Complexity - Measuring the Maturity of Online Enquiry CommunitiesA Question of Complexity - Measuring the Maturity of Online Enquiry Communities
A Question of Complexity - Measuring the Maturity of Online Enquiry Communities
 
Automatic Identification of Best Answers in Online Enquiry Communities
Automatic Identification of Best Answers in Online Enquiry CommunitiesAutomatic Identification of Best Answers in Online Enquiry Communities
Automatic Identification of Best Answers in Online Enquiry Communities
 
Asterid: Linked Data Asterisms
Asterid: Linked Data AsterismsAsterid: Linked Data Asterisms
Asterid: Linked Data Asterisms
 
Veracity ESWC2010 Poster
Veracity ESWC2010 PosterVeracity ESWC2010 Poster
Veracity ESWC2010 Poster
 
Attention-Streams Recommendations
Attention-Streams RecommendationsAttention-Streams Recommendations
Attention-Streams Recommendations
 
Sparks O3 Browser: Augmenting the Web with Semantic Overlays
Sparks O3 Browser: Augmenting the Web with Semantic OverlaysSparks O3 Browser: Augmenting the Web with Semantic Overlays
Sparks O3 Browser: Augmenting the Web with Semantic Overlays
 

Recently uploaded

Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024AyushiRastogi48
 
FREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naFREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naJASISJULIANOELYNV
 
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 GenuineCall Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuinethapagita
 
Behavioral Disorder: Schizophrenia & it's Case Study.pdf
Behavioral Disorder: Schizophrenia & it's Case Study.pdfBehavioral Disorder: Schizophrenia & it's Case Study.pdf
Behavioral Disorder: Schizophrenia & it's Case Study.pdfSELF-EXPLANATORY
 
Speech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxSpeech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxpriyankatabhane
 
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)riyaescorts54
 
The dark energy paradox leads to a new structure of spacetime.pptx
The dark energy paradox leads to a new structure of spacetime.pptxThe dark energy paradox leads to a new structure of spacetime.pptx
The dark energy paradox leads to a new structure of spacetime.pptxEran Akiva Sinbar
 
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubai
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In DubaiDubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubai
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubaikojalkojal131
 
Neurodevelopmental disorders according to the dsm 5 tr
Neurodevelopmental disorders according to the dsm 5 trNeurodevelopmental disorders according to the dsm 5 tr
Neurodevelopmental disorders according to the dsm 5 trssuser06f238
 
Pests of soyabean_Binomics_IdentificationDr.UPR.pdf
Pests of soyabean_Binomics_IdentificationDr.UPR.pdfPests of soyabean_Binomics_IdentificationDr.UPR.pdf
Pests of soyabean_Binomics_IdentificationDr.UPR.pdfPirithiRaju
 
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRCall Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRlizamodels9
 
Citronella presentation SlideShare mani upadhyay
Citronella presentation SlideShare mani upadhyayCitronella presentation SlideShare mani upadhyay
Citronella presentation SlideShare mani upadhyayupadhyaymani499
 
Bioteknologi kelas 10 kumer smapsa .pptx
Bioteknologi kelas 10 kumer smapsa .pptxBioteknologi kelas 10 kumer smapsa .pptx
Bioteknologi kelas 10 kumer smapsa .pptx023NiWayanAnggiSriWa
 
User Guide: Magellan MX™ Weather Station
User Guide: Magellan MX™ Weather StationUser Guide: Magellan MX™ Weather Station
User Guide: Magellan MX™ Weather StationColumbia Weather Systems
 
Pests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPirithiRaju
 
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPirithiRaju
 
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingBase editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingNetHelix
 
Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...navyadasi1992
 
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptx
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptxSTOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptx
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptxMurugaveni B
 

Recently uploaded (20)

Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024Vision and reflection on Mining Software Repositories research in 2024
Vision and reflection on Mining Software Repositories research in 2024
 
FREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by naFREE NURSING BUNDLE FOR NURSES.PDF by na
FREE NURSING BUNDLE FOR NURSES.PDF by na
 
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 GenuineCall Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
 
Hot Sexy call girls in Moti Nagar,🔝 9953056974 🔝 escort Service
Hot Sexy call girls in  Moti Nagar,🔝 9953056974 🔝 escort ServiceHot Sexy call girls in  Moti Nagar,🔝 9953056974 🔝 escort Service
Hot Sexy call girls in Moti Nagar,🔝 9953056974 🔝 escort Service
 
Behavioral Disorder: Schizophrenia & it's Case Study.pdf
Behavioral Disorder: Schizophrenia & it's Case Study.pdfBehavioral Disorder: Schizophrenia & it's Case Study.pdf
Behavioral Disorder: Schizophrenia & it's Case Study.pdf
 
Speech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptxSpeech, hearing, noise, intelligibility.pptx
Speech, hearing, noise, intelligibility.pptx
 
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
(9818099198) Call Girls In Noida Sector 14 (NOIDA ESCORTS)
 
The dark energy paradox leads to a new structure of spacetime.pptx
The dark energy paradox leads to a new structure of spacetime.pptxThe dark energy paradox leads to a new structure of spacetime.pptx
The dark energy paradox leads to a new structure of spacetime.pptx
 
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubai
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In DubaiDubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubai
Dubai Calls Girl Lisa O525547819 Lexi Call Girls In Dubai
 
Neurodevelopmental disorders according to the dsm 5 tr
Neurodevelopmental disorders according to the dsm 5 trNeurodevelopmental disorders according to the dsm 5 tr
Neurodevelopmental disorders according to the dsm 5 tr
 
Pests of soyabean_Binomics_IdentificationDr.UPR.pdf
Pests of soyabean_Binomics_IdentificationDr.UPR.pdfPests of soyabean_Binomics_IdentificationDr.UPR.pdf
Pests of soyabean_Binomics_IdentificationDr.UPR.pdf
 
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCRCall Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
Call Girls In Nihal Vihar Delhi ❤️8860477959 Looking Escorts In 24/7 Delhi NCR
 
Citronella presentation SlideShare mani upadhyay
Citronella presentation SlideShare mani upadhyayCitronella presentation SlideShare mani upadhyay
Citronella presentation SlideShare mani upadhyay
 
Bioteknologi kelas 10 kumer smapsa .pptx
Bioteknologi kelas 10 kumer smapsa .pptxBioteknologi kelas 10 kumer smapsa .pptx
Bioteknologi kelas 10 kumer smapsa .pptx
 
User Guide: Magellan MX™ Weather Station
User Guide: Magellan MX™ Weather StationUser Guide: Magellan MX™ Weather Station
User Guide: Magellan MX™ Weather Station
 
Pests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdfPests of safflower_Binomics_Identification_Dr.UPR.pdf
Pests of safflower_Binomics_Identification_Dr.UPR.pdf
 
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdfPests of Blackgram, greengram, cowpea_Dr.UPR.pdf
Pests of Blackgram, greengram, cowpea_Dr.UPR.pdf
 
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editingBase editing, prime editing, Cas13 & RNA editing and organelle base editing
Base editing, prime editing, Cas13 & RNA editing and organelle base editing
 
Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...Radiation physics in Dental Radiology...
Radiation physics in Dental Radiology...
 
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptx
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptxSTOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptx
STOPPED FLOW METHOD & APPLICATION MURUGAVENI B.pptx
 

Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media

  • 1. GRÉGOIRE BUREL, HASSAN SAIF, HARITH ALANI Knowledge Media Institute, The Open University, Milton Keynes, UK. ISWC’17, Vienna, Austria. 21-25 October 2017. Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media
  • 2. Event Detection and Crisis Situations Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 2 Event detection is “the task of automatically identifying certain clues in texts that denote a specific event type or theme”. - Help identifying/responding to events. - Organise relevant information during crises. Twitter: ~200 million active users. ~400 million tweets a day. Twitter usage during crises: 1. During the 2011 Japan earthquake, 177 million tweets related to the event were sent in one day. 2. The news about the Boston bombings first appeared on Twitter.
  • 3. Crisis-Related Event Detection Tasks Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media Publications 3 Crisis-related event detection is often divided into three main tasks (Olteanu et al. 2015): Crisis Related / Unrelated Crisis Type Information Categories Task 1 Identify the different types of crises the message is related to. Differentiate the type of information contained in the message. e.g., shooting, explosion, building collapse, fires, floods, meteorite fall, etc. e.g., affected individuals, infrastructures and utilities, donations and volunteer, caution and advice, etc. Granularity Differentiate the posts that are related or unrelated to crises. Task 2 Task 3
  • 4. Crisis-Related Event Detection Tasks Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media Publications 4 Crisis-related event detection is often divided into three main tasks (Olteanu et al. 2015): Crisis Related / Unrelated Crisis Type Information Categories Task 1 Identify the different types of crises the message is related to. Differentiate the type of information contained in the message. e.g., shooting, explosion, building collapse, fires, floods, meteorite fall, etc. e.g., affected individuals, infrastructures and utilities, donations and volunteer, caution and advice, etc. Granularity Differentiate the posts that are related or unrelated to crises. Task 2 Task 3
  • 5. ‘Traditional’ ML vs. Deep Learning Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 5 Deep Learning - Artificial neural networks. - Minimum feature engineering - Word embeddings (Bengio et al., 2013). ‘Traditional’ ML - Standard classifiers (e.g., SVM, J48…). - Feature engineering (e.g., lemmatisation, TF-IDF…). - Bag of words.
  • 6. Text vs. Semantics – Document Contextualisation Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 6 Obama attends vigil for Boston Marathon bombing victims Politician / Person Sports Event / Social Event / Event Disaster / Event Incorporating Semantics into ML Classification Methods for contextualising documents: - Approach 1: Traditional ML Classifiers - Approach 2: Deep Learning
  • 7. CNN for Sentence Classification (Kim et al., 2014) Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 7
  • 8. Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 8 CNN for Sentence Classification Dual-CNN (Semantic Channel) CNN for Sentence Classification (Kim et al., 2014) + Competitive results for text classification tasks. + No or Little Feature Engineering required. + Relatively good at taking local textual relations within short documents. - No ‘native’ semantic context. Dual-CNN (Burel et al., 2017) + Text CNN + Aligned Semantic channel - Concept extraction. - Semantics vocabulary (4000) << Words vocabulary (60000)
  • 9. Wide and Deep Learning (Cheng et al., 2016) Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 9
  • 10. Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 10 Wide and Deep Learning Sem-CNN (W-D-CNN) Wide and Deep Learning (Cheng et al., 2016) + Efficiently Deal with ‘sparse’ and ‘dense’ inputs. - Not very efficient for modelling text relations. - No ‘native’ semantic context. Sem-CNN (W-D-CNN) + Text CNN / Wide and Deep Models + Deep Shallow Word Embeddings + Wide Deep Semantics - Requires semantic extraction.
  • 11. Wide and Deep Semantic CNN (Sem-CNN) Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 11
  • 12. Sem-CNN – Experimental Setup Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media Dataset - T26 (28,000 annotated tweets) - 12 Crisis types (shooting, explosion, building collapse, fires, floods, meteorite fall, haze, bombing, typhoon, crash, earthquake, and derailment). - 6 Information categories (affected individuals, infrastructures and utilities, donations and volunteer, caution and advice, sympathy and emotional support, and other useful information) Semantic Extraction - - Extracted Entities/Concepts: 65% dataset coverage. Concept Vectors Initialisation - Concept Labels: Obama → dbo:Obama - Concept Abstracts: Obama → dbo:Obama → ‘Barack Hussein Obama II; born August 4, 1961) is an American politician…’
  • 13. Sem-CNN – Experimental Setup Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 13 Dataset versions - Full Dataset: 28,000 tweets. - Balanced Dataset (BD1): 9100 tweets (32.6%). - Semantically Balanced Dataset (>2 entities/concepts, BD2): 1194 tweets (4.3%). Baselines - SVM (TF-IDF): Linear SVM using the words’ TF-IDF vectors extracted from our dataset. - SVM (Word2Vec): Linear SVM using the Google pre-trained 300- dimensional word embeddings. Evaluation - 5-folds cross validation. - Sem-CNN: 300-dim embeddings, Fn = 128 convolutional filter of sizes Fs = [3,4,5], 0.5 dropout and ADAM. - Evaluation Measures: P, R and F1. ?
  • 14. Results Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 14
  • 15. Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media Publications 15 + - Sem-CNN significantly outperforms the baselines (p < 0.001) - More semantics leads to better results. - Sem-CNN appears to perform better than Dual- CNN (up to +4% F1) with F1 up to 64%. - Abstract outperform the Concept vectors but it is not always significant (i.e., on the full dataset). - Consider more complex deep learning models such as Recurrent Neural Networks (RNN) or Hierarchical Attention Networks (HAN). - Initialise with different embeddings (e.g., Twitter) and perform parameter optimisation. - Investigate other methods for integrating semantics (e.g., extended concept graphs). - Results and Future Work CREES Crisis Event Extraction Service ?
  • 16. Questions @ Email: g.burel@open.ac.uk Twitter: @evhart CREES: https://github.com/evhart/crees COMRADES: http://comrades-project.eu Semantic Wide and Deep Learning for Detecting Crisis-Information Categories on Social Media 16