(Semi-)Automatic analysis of online contents

Institute for Web Science and Technologies · University of Koblenz-Landau, Germany

(Semi-)Automatic Analysis of Online Contents

Steffen Staab@ststaab

Web and Internet Science Group · ECS · University of Southampton, UK &

(Semi-)Automatic analysis of online content 2/68Steffen Staab

Content analysis

Text++

Content


Is it difficult?

„Nach dem Auspacken der LPS-105 präsentiert sich dem Betrachter ein stabiles Laufwerk, das genauso geringe Außenmaße besitzt wie die Maxtor.“

Unpacking the LPS 105 reveals a sturdy disk drive which is of the same small size as the Maxtor.

Text++

Content


„Content“ analysis: What is in online content?

....

Entailment

Summaries

Arguments

Discourse

OpinionsSentiments

Facts – who, what, when?

Syntax

Semantics

Pragmatics

Knowledge

Text++

Content

CLing


Purpose

Technical objectives• Search• data & knowledge

bases:• facts• arguments• ...

Applications• Google Search• Watson • „Watson 2“

Social science and humanities objectives

• Form hypotheses• Find indications• Recognize trends• ...


Objective oriented content analysis

....

Entailment

Summaries

Arguments

Discourse

OpinionsSentiments

Facts – who, what, when?

Syntax

Semantics

Pragmatics

Knowledge

Text++

Semantic Web

Trend hypotheses

Selection of facts, function,

trust

CLing


SEMANTIC WEB ANNOTATION


CREAM – Creating Metadata (Handschuh et al 2002, 2003)

Document Viewer / EditorOntology

Guidance & Fact Browser

Concepts

Instances of Concepts

Attribute Instances = instance of a property to a datatype instance

Relationship Instances =instance of a property

to a class instance


CREAM – Creating Metadata (Handschuh et al 2002, 2003)

Open world - Target ontologies now could be:• Schema.org

(3 Trillion facts collected by Google; 10,000 of concepts)

• Wikidata1,148,230 concepts (2 weeks ago)


Annotating facts with Cream

+++Open (wrt ontologies)Flexible

Semi-automatic: SCREAM

---

Effort for annotation(minimize # of clicks)Thick ClientTech Readiness Level ~5A lot of effort to prepare tool

for a taskLimited accuracy


Technology Readiness LevelsTRL 1: Beobachtung und Beschreibung des

Funktionsprinzips (8-15 Jahre zur Marktreife)TRL 2: Beschreibung der Anwendung einer TechnologieTRL 3: Nachweis der Funktionstüchtigkeit einer Technologie

(5-13 Jahre zur Marktreife)TRL 4: Versuchsaufbau im LaborTRL 5: Versuchsaufbau in EinsatzumgebungTRL 6: Prototyp in EinsatzumgebungTRL 7: Prototyp im Einsatz (1-5 Jahre)TRL 8: Qualifiziertes System mit Nachweis der

Funktionstüchtigkeit im EinsatzbereichTRL 9: Qualifiziertes System mit Nachweis des

erfolgreichen Einsatzes


CLUSTERING OF TEXTDATA

http://topicmodels.west.uni-koblenz.deWith Christoph Kling


Text Mining Documents

Documents are PDFs, emails, tweets,

Flickr photo tags, Word companions,…

Documents consist of bag of words metadata

- author(s) - timestamp- geolocation- publisher- booktitle- device...

Chinese food

Vegan

food

Break-fast

dimsumduckeggs

...

vegantofu...

eggsham...

Objective:Cluster, categorize,

& explain


Latent Dirichlet Allocation (LDA)


Latent Dirichlet Allocation (LDA)

Document-topic distributions

Topic-word distributions

K topicsM documentsEach doc m from M has length Nm


Use Metadata to Help Topic Prediction

Improve topic detection→ Morning times may help to improve the breakfast topic Describe dependencies: metadata ↔ topics

→ breakfast topic happens during morning hours Chinese

food

Vegan

food

Break-

fast

dimsumduckeggs

...

vegantofu...

eggsham...


Use Metadata to Help Topic Prediction

Improve topic detection→ Morning times may help to improve the breakfast topic Describe dependencies: metadata ↔ topics

→ breakfast topic happens during morning hours

Usage Autocompletion

→ From words to words Prediction of search queries

→ From metadata to words→ From words to metadata

Chinese food

Vegan

food

Break-

fast

dimsumduckeggs

...

vegantofu...

eggsham...


Dataset

Linux Kernel Mailinglist3,400,000 emails with timestamps and mailinglist ID


Nominal

Ordinal

Cyclic

Spherical

Networked

Structures of Metadata Spaces Kern Desk Mail

Spatial Model is not used in this application(but might be)!


Topics


Topics


Topics

Professional topics:

Hobbyist topics:


Topics

Metadata weighting:


126,408 Online Fetish Users: First 8 Topics


Sociodemographics of Fetish dataset


Influence of Sociodemographics on Favorite Fetishes


Other applications of (extended) LDA

Sentiment and Topics(Naveed et al ICWSM 2013)

Topics and spatial knowledge(Kling et al WSDM 2014)

Modelling of power(Kling et al ICWSM 2015)


BELIEVABILITY AND TRUST IN ONLINE NEWS

With Christoph Kling, Jerome KunegisCollaboraiton with Jutta Milde, Karin Stengel, Ines VogelOngoing work in KOMEPOL


Targets


Example article at Spiegel.de


Requirements

Scalability:• # Documents• # Annotators• # Annotations per

annotater

Tool:• Administration• Crowdsourcing• Semi-automatic


Separating article management and coding


Text-Upload


Managing projects


Article


Defining a Coding-Job


Highlighting using Keywords and Clustering


Article coding


Preparing a code book (1)


Preparing a code book (2)


CONCLUSION


Lessons Learned

New targets• Require new modeling of

gaps

Challenges• Technology Readiness

Levels• Many tools – no „good“ tool

(„done is better than perfect“?)

• Reproducability

ToDos• Eclipse/Protege of

annotation• modular• extensible• open

• Optimizing the processes


No tool to rule them all

....

Entailment

Summaries

Arguments

Discourse

OpinionsSentiments

Facts – who, when, where, what?

Syntax

Semantics

Pragmatics

Knowledge

Text++

Semantic Web

Trend-hypothesen

Faktenauswahl, Funktion, Vertrauen

Gap

Gap

CLing


THANK YOU FOR YOUR ATTENTION!


C. C. Kling, J. Kunegis, S. Sizov, and S. Staab. “Detecting non-gaussian geographical topics in tagged photo collections.” In: Seventh ACM International Conference on Web Search and Data Mining, WSDM 2014, New York, NY, USA, February 24-28, 2014.

I. C. Vogel, J. Milde, K. Stengel, S. Staab, C. C. Kling, and J. Kunegis. “Glaubwürdigkeit und Vertrauen von Online-News.” In: Datenschutz und Datensicherheit 39.5 (2015), pp. 312–316.

S. Handschuh, S. Staab. CREAM – CREAting Metadata for the Semantic Web. Computer Networks. 42(5): 579-598, Elsevier 2003.

S. Handschuh, S. Staab, F. Ciravegna. S-CREAM – Semi-automatic CREAtion of Metadata.In: Proc. of the European Conference on Knowledge Acquisition and Management – EKAW-2002 . Madrid, Spain, October 1-4, 2002. LNCS/LNAI 2473, Springer, 2002, pp. 358-372.

C. Kling. Probabilistic Models for Context in Social Media. Novel Approaches and Inference Schemes. Submitted as PhD thesis, Institute for Web Science and Technologies, University of Koblenz-Landau, to be defended Nov/Dec 2016

Nasir Naveed, Thomas Gottron, Steffen Staab:Feature Sentiment Diversification of User Generated Reviews: The FREuD Approach. ICWSM 2013

Christoph Carl Kling, Jérôme Kunegis, Heinrich Hartmann, Markus Strohmaier, Steffen Staab:Voting Behaviour and Power in Online Democracy: A Study of LiquidFeedback in Germany's Pirate Party. ICWSM 2015: 208-217

Bibliography


URLs

http://topicmodels.west.uni-koblenz.dehttp://komepol.west.uni-koblenz.de

http://www.slideshare.net/steffenstaab