Department of Computational Perception
Department of
Computational Perception
Johannes Kepler Universität Linz


Home  –  Mission  –  Teaching  –  People  –  Research  –  Media  –  Awards  –  Impressum


WEB & SOCIAL MEDIA

<< back to overview

We are actively pursuing research related to Web and Social Media Mining and Analysis, exploiting resources such as web pages, blogs, YouTube, Facebook, Twitter, and Last.fm. Many of the following topics are strongly connected to music and multimedia applications. However, we are also open to more general topics, involving web crawling, information retrieval, and information extraction.
Topics might also be worked on during or as an extension to our classes on Learning from User-generated Data, Multimedia Data Mining, and Social Media Mining and Analysis.

Analysis and Prediction of User Traits and Behavior

This group of topics focuses on analyzing connections between social media data and characteristics of the users as well as predicting these characteristics by applying machine learning techniques. Features extracted from social media data may include metadata about the user (e.g., number of tweets per day or number of friends) or aspects extracted via natural language processing. Also shared multimedia material (e.g., images and video) or sensor data (e.g., from smart phones) can be analyzed and incorporated. User characteristics can be of very different nature, for instance, behavioral or consumption patterns, taste (music, movies, web pages, etc.), age, gender, location, or even personality traits. For analysis and prediction, methods from statistics and machine learning, (e.g., correlation analysis and classification, respectively), should be investigated and their performane thoroughly evaluated.

References: Contact: Markus Schedl, Peter Knees


Trend and Popularity Prediction

Usage of social media has experienced an incredible increase during the last couple of years. Nowadays, people create, share, consume, and comment on all kinds of multimedia items through online social networks and platforms. Predicting whether a particular item will become popular or not is a hot topic both in academia and industry. Possible topics for practical projects and Master's theses in this context include elaborating new techniques to harvest a variety of data sources (e.g., multimedia content descriptors, microblogs, social network structure, or consumption histories), researching models that describe the popularity of items (songs, videos, persons, scientific publications, etc.), designing computational features that will serve as predictors for future popularity in machine learning algorithms, and evaluating them in comprehensive experiments. Another related topic is the analysis and prediction of "popularity flows", i.e., where trends emerge and how they spread temporally and spatially.

References: Contact: Markus Schedl


Opinion Mining and Sentiment Analysis

"Sentiment analysis or opinion mining refers to the application of natural language processing, computational linguistics, and text analytics to identify and extract subjective information in source materials. Generally speaking, sentiment analysis aims to determine the attitude of a speaker or a writer with respect to some topic or the overall contextual polarity of a document. The attitude may be his or her judgement or evaluation (see appraisal theory), affective state (that is to say, the emotional state of the author when writing), or the intended emotional communication (that is to say, the emotional effect the author wishes to have on the reader)." [Wikipedia]

Examples for projects:
  • Comment and posting analysis
  • Automatic rating prediction
  • Controversy detection / detecting bipolar sentiments
  • Spam-comment filtering
  • Affect prediction from multimedia content (e.g., speech or video)
Contact: Marcin Skowron, Markus Schedl, Peter Knees


Recommendation and Discovery

Exploiting the web and social media, one can identify many potential sources for discovery and recommendation of similar items (music pieces, movies, news, etc.). While topics related to all kinds of multimedia material can be worked on, in the context of music, similar artists or songs are of highest interest. Sources to be considered include word patterns on music pages, user-generated tags, ratings, tweets, search engine page counts, peer-to-peer networks, or playlists. Mining these sources can not only give information on the artistic context of music and help estimating similarity, but also to defer measures and indicators such as popularity, hotness, novelty, diversity, and genre prototypicality. In this context, topics related to user modeling and personalized as well as user-aware retrieval and recommendation can be considered, too.

Examples for projects:
  • Multimedia recommender systems
  • User modeling and user-aware retrieval and recommendation
  • Analyzing and preventing hubness in recommender systems
  • Implementation and comparison of different similarity measures
  • Automatic playlist generation
  • Taste models for music (or multimedia) consumption behavior (diversity, mainstreaminess, openness to new content, etc.)
References:
Contact: Peter Knees, Markus Schedl



last edited by ms on Sep 24, 2015