ESWC 2014 Tutorial Part 4
1 like334 views
This document discusses using semantics and ontologies to analyze and understand social media data. It covers: 1) Representing social media content, users, and relations using semantic web technologies like SIOC. 2) Using semantics to understand topics and their evolution in social media conversations by extracting concepts and analyzing them over time and across groups. 3) Analyzing sentiment in social media by incorporating semantics into machine learning models, which increases accuracy of sentiment analysis. 4) Modeling user behavior and roles in online communities using an ontology, inferring roles over time, and analyzing how role compositions correlate with community characteristics.
ESWC 2014 Tutorial Part 4
- 1. Social Web: Where are the Semantics? ESWC 2014 Miriam Fernández, Victor Rodríguez, Andrés García-Silva, Oscar Corcho Ontology Engineering Group, UPM, Spain Knowledge Media Institute, The Open University
- 2. Outline 2 • Part 1: Understanding Social Media – Theory: background & applications described in this tutorial – Hands on: data extraction from Twitter and Facebook • Part 2: Using semantics to represent data from SNS – Theory: Using SW to represent content, users and relations – Hands on: applying and extending SIOC • Part 3: Using semantics to understand social media conversations – Theory: Using semantics to understand topics in social media – Hands on: using LDA to extract topics from social media • Part 4: Using semantics to understand user behaviour
- 3. Implicit vs. Explicit Semantics • Implicit Semantics – Implicit, also called statistical semantics, focus on extracting word sense by studying the patterns of human word usage in massive collections of text or other human generated data – It does not rely on an explicit formalisation/conceptualisation of knowledge • Explicit Semantics – Explicit semantics focus on the analysis of content by using the support of explicit conceptualisations in the form of ontologies and knowledge bases ESWC 2014 Social Web: Where are the Semantics? 3
- 4. Explicit Semantics Structured Unstructured From the Web of human generated content The Web of unstructured text (Posts / Documents) and Links To the Web of machine understandable content The Web of Objects and Relations
- 5. • The annotators extract entities (classes / individuals) and relations from the text and link them to object URIs Obtaining explicit semantics from social media content
- 6. Using Semantics to Analyse Topic Evolution • LDA topics are identified by a set of keywords – Difficult to assess their meaning and evolution • Use explicit semantics to characterise topics as concrete entities 6
- 7. ! ! Using Semantics to Analyse Topic Evolution ESWC 2014 Social Web: Where are the Semantics? 7 • Analyse concepts appearance – Within a group – Across groups – Over time • Type filtering • Interlinking with other datasets (data.open.ac.uk)
- 8. Using Semantics To Analyse Sentiment • Sentiment analysis on social media – Offers a fast and cheap access to publics’ feelings towards brands, business, people, etc. – Comes with additional challenges – Current approaches • Lexical-based • Machine Learning – Explicit semantics are often neglected ESWC 2014 Social Web: Where are the Semantics? 8
- 9. Using Semantics to Analyse Sentiment • Add semantics as additional features into the training set • Results – Incorporating semantics increases accuracy by 6.5% for negative sentiment and by 4.8% for positive sentiment – The use of explicit semantics is more appropriate when the datasets being analysed are large and cover a wide range of topics Saif, Hassan, He, Yulan, Alani, Harith (2012). Semantic sentiment analysis of twitter. In: 11th International Semantic Web Conference (ISWC 2012)
- 10. “Words that occur in similar context tend to have similar meaning” Wittgenstein (1953) Using Semantics To Analyse Sentiment • SentiCircles – Integrates implicit and explicit semantics to analyse sentiment – Outperforms other lexicon labeling methods and overtakes the state-of-the- art SentiStrength approach in accuracy, with a marginal drop in F-measure ESWC 2014 Social Web: Where are the Semantics? 10 Saif, Hassan, Fernandez, Miriam, He, Yulan, Alani, Harith (2014). SentiCircles for Tweet-level Sentiment Analysis (ESWC 2014) -> conference presentation on the 27, 14:00!!
- 11. Using Semantics To Analyse Sentiment ESWC 2014 Social Web: Where are the Semantics? 11
- 12. Using Semantics to Analyse User Behaviour • Goal – Monitor and capture member activities – Analyse emerging behaviour over time – Understand the correlation of behaviour with community evolution • Approach – Identify behavioural features and behaviour roles – Create an ontology to model behavioural roles and behaviour features – Use semantic rules to infer user roles in online communities – Study role composition patterns ESWC 2014 Social Web: Where are the Semantics? 12 Angeletou, S., Rowe, M. and Alani, H. (2011) Modelling and Analysis of User Behaviour in Online Communities, 10th International Semantic Web Conference (ISWC 2011), Bonn, Germany Rowe, Matthew; Fernandez, Miriam; Angeletou, Sofia and Alani, Harith (2013). Community analysis through semantic rules and role composition derivation. Journal of Web Semantics: Science, Services and Agents on the World Wide Web, 18(1) pp. 31–47
- 13. Behavioural roles and features ESWC 2014 Social Web: Where are the Semantics? 13 Table 1. Roles and the feature-to-level mappings Role Feature Level Elitist In-Degree Ratio low Bi-directional Threads Ratio high Bi-directional Neighbours Ratio low Grunt Bi-directional Threads Ratio med Bi-directional Neighbours Ratio med Average Posts per Thread low STD of Posts per Thread low Joining Conversationalist Thread Initiation Ratio low Average Posts per Thread high STD of Posts per Thread high Popular Initiator In-Degree Ratio high Thread Initiation Ratio high Popular Participants In-Degree Ratio high Thread Initiation Ratio low Average Posts per Thread med STD of Posts per Thread med Supporter In-Degree Ratio med Bi-directional Threads Ratio med Bi-directional Neighbours Ratio med Taciturn Bi-directional Threads Ratio low Bi-directional Neighbours Ratio low Average Posts per Thread low STD of Posts per Thread low Ignored Posts Replied Ratio low Jeffrey Chan, Conor Hayes, and Elizabeth Daly. Decomposing discussion forums using common user roles. In Proc. Web Science Conf. (WebSci10), Raleigh, NC: US, 2010.
- 14. Modelling user features and interactions ESWC 2014 Social Web: Where are the Semantics? 14 http://purl.org/net/oubo/0.3• OUBO: The OU Behaviour Ontology
- 15. Encoding Rules in Ontologies with SPIN ESWC 2014 Social Web: Where are the Semantics? 15
- 16. Apply rules to infer user roles over time ESWC 2014 Social Web: Where are the Semantics? 16 1.- Construct features for community users at a given time step 2.- Derive bings using equal frequency binning Popularity-low cutoff = 0.5 Initiation-high cutoff = 0.4 3.- Use skeleton rule base to construct rules using bin levels Popularity=low, Initiation=high ->roleA Popularity<0.5, Initiation > 0.4 -> roleA 4.- Apply rules to infer user roles and community composition 5.- Repeat 1-4 following time steps
- 17. Analyse the role composition of the community ESWC 2014 Social Web: Where are the Semantics? 17 • Investigate the correlation between the role composition and the students’ performance
- 18. Analyse the role composition of the community • Allow Policy Makers to focus on a smaller set of users, with whom they may want to engage more closely ESWC 2014 Social Web: Where are the Semantics? 18
- 19. Analyse the role composition of the community • Development of models to predict community health based on role compositions and evolution of user behaviour – Health Indicators • Churn Rate: proportion of users who leave the network in a given time segment • User Count: number of users who posted at least once • Seeds / Non seeds: proportion of posts that get responses vs. those that don’t • Clustering coefficient: measures the cohesion within the network – Results • Accurate detection of community health is possible using role composition information • There is no “one size fits all” model ESWC 2014 Social Web: Where are the Semantics? 19 Rowe, M. and Alani, H. (2012) What Makes Communities Tick? Community Health Analysis using Role compositions. International Conference on Social Computing, 2012 0.0 0.2 0.4 0.6 0.8 1.0 0.00.20.40.60.81.0 Churn Rate FPR TPR 0.0 0.2 0.4 0.6 0.8 1.0 0.00.20.40.60.81.0 User Count FPR TPR 0.0 0.2 0.4 0.6 0.8 1.0 0.00.20.40.60.81.0 Seeds / Non−seeds Prop FPR TPR 0.0 0.2 0.4 0.6 0.8 1.0 0.00.20.40.60.81.0 Clustering Coefficient FPR TPR
- 20. Challenges: How would you address them? • Scalability – Communities exceed millions of users – Infrastructures must support hundreds of millions discussion threads • Growth (real-time analysis) – Speed of new incoming data / stream processing • Concept vs. keyword based data acquisition/pre-processing – How to filter certain tags? – Which new topics emerge? – How topics evolve over time? – Authorship in social media, who copies who? • Multilingualism – We all speak different languages • Understanding the user and acting accordingly – We all have different personalities, behaviours and preferences ESWC 2014 Social Web: Where are the Semantics? 20