Stream Leaders: Professor Barry Smyth and Professor Alan Smeaton
This Research Stream provides organisation and management to content information, this information having been gathered from various physical and online sensor sources. It achieves this by exploring the intersection and interaction between users and the sensor web.
Within CLARITY, raw sensor data is captured from whatever applications and sources are appropriate or available and we then enrich this data using contextual information. In the case of current work the raw sensor data includes a wide variety of sources including wearable cameras, Bluetooth signatures from mobile phones, EEG readings, environmental sensor readings, information from video sources, data from physiological sensors, location information, energy consumption information, data about weather conditions, Twitter streams and so on. This contextual enrichment is an automated process and depends on the domain and the data sources. The approaches we take include machine learning techniques, sentiment analysis, extracting semantics from video, automatic classification, adding trust and reliability, adding semantics to lifelogs and enriching sensor readings with location information.
Research Stream 4 also explores the intersection and interaction between users and the sensor web, combining ideas from user modelling, recommender systems, and information retrieval with a primary goal to develop context-aware, personalized sensor web technologies that are capable of adapting to the changing needs and contexts of individuals, small groups of users, and even larger communities of users. Based on this interaction, three areas are examined:
- Profiling, Preferences, Activities, and Context – focusing on the understanding the needs and preferences of users based on the availability of diverse sources of sensor data, including online and physical-world sensor sources.
- Recommendation & Collaboration – focusing on the development of recommendation techniques to support proactive information discovery techniques that are capable of automatically adapting sensor web data for the needs of individual users.
- Communities, Trust, and Privacy – focusing on the relationship between groups and communities of users (as opposed to individual users) from an information discovery standpoint.
Sentiment Analysis on Social Networks
Sentiment analysis refers to the automatic analysis of text with a view to determining its ‘sentiment’ or view, towards a particular subject. For example this can include analysis of a newspaper article to determine whether is is positive or negative to a person, a company or a product. Within CLARITY we have developed techniques to determine sentiment on real time events using text from social networks and from blogs. We apply this to real time events as diverse as football matches on TV, reality TV shows, financial markets and political parties during elections.
The world of the web has become more social and real-time. Facebook and Twitter are examples of a new generation of social, real-time web services which provide a fertile ground for recommender systems research. CLARITY researchers are evaluation a range of different profiling and recommendation strategies based on a large data set of Twitter users and their tweets to demonstrate the potential for effective and efficient followee recommendation.