attention-streams recommendations
DESCRIPTION
Real-time and Contextual Recommendations with Attention Modelling.TRANSCRIPT
Attention-Streams
Attention-Streams
Grégoire Burel, OAK Group, University Of Sheffield
ESWC 2010, Heraklion,30 May 2010
Attention-Streams
Introduction
• Attention-Streams– Attention-streams Recommendations:
• Contextual and real-time recommendations.• Passive recommendations.
– Modelling Attention streams :• Attention streams and existing recommendations.• Attention vs. Interests.• Modelling attention.• Monitoring Attention.• Attention based recommendations
– Demo:• Video
– Conclusions
Attention-Streams
Recommender Systems
• Contextualizing information and users using cross-domain attention modeling.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Information + Users + Interests
Attention-Streams
Recommender Systems
• Contextualizing information and users using cross-domain attention modeling.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Information + Users + Interests
Attention-Streams
Attention-Streams
(Real-time and Contextual Recommendations)
Attention-Streams
Attention-Streams RecommendationsContextual and Real-time Recommendations
Attention-Streams
Contextual and Real-time Recommendations
• Features:– Models users interests across networks and communities:
• Interests are not fragmented.
– Recommendations matches real-time user interests:• Information and user interests evolve rapidly independently of the users common interests.
– Real-time interests might be linked to FOAF profiles:• Real-time interests can be shared between different contexts and application.
– Contextual ‘bookmarks’:• Relevant recommendations might be bookmarked by the user.
•–
–
–
Attention-Streams
Contextual and Real-time Recommendations
• Features:– Models users interests across networks and communities:
• Interests are not fragmented.
– Recommendations matches real-time user interests:• Information and user interests evolve rapidly independently of the users common interests.
– Real-time interests might be linked to FOAF profiles:• Real-time interests can be shared between different contexts and application.
– Contextual ‘bookmarks’:• Relevant recommendations might be bookmarked by the user.
• Content Recommendations:– Local events using user location and current interests.– Information sources using contextual RSS subscriptions. – Real-time information streams given current interests.
Attention-Streams
Passive Recommendations
Local events + Information Streams + Contextual RSS
Mobile
DesktopCross-domain
Interests
Attention-Streams
Passive Recommendations
• Recommendations do not require any particular action to be accessed:– Users might ignore or access the recommendations without disturbing their current
workflow.
Attention-Streams
Modelling Attention-StreamsAttention-Streams and Existing Recommendations
Attention-Streams
Attention-Streams and Existing Recommendations
• Contextualizing information and users using cross-domain attention modeling.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Existing recommendations are fragmented, network specific, community dependent and long-term oriented (Resnick, 1997)
Attention-Streams
Attention-Streams and Existing Recommendations
Movies
Events
Music
Content
People
Products
Attention-Streams
Attention vs. Interests
• Modelling particular user Interests within a system or generic interests (Resnick, 1997).
– Explicit:• “Tell me what you like”
– Implicit:• “Let me guess what you like
given what you do”.
User Interests
• Modelling information access and usage across domains. – User Activity: (Dragunov, 2005)
• Work/Leisure.• News browsing, Finding a
Restaurant…
Attention Profiling
Contextual ‘Interests’Long-term Interests
(Middleton, 2004)
Attention-Streams
Attention vs. Interests
• Attention Management:– Attention models have been designed for dealing with interruption
overload (attention management):• Attention for information notification (Vertegaal, 2003) (Horvitz et al., 2003).
• Attention and Information Contextualisation:– Attention is currently applied to information presentation.
Attention:The cognitive process of selectively concentrating on one aspect of the environment while ignoring other things.
Attention-Streams
Attention vs. Interests
• Attention Management:– Attention models have been designed for dealing with interruption
overload (attention management):• Attention for information notification (Vertegaal, 2003) (Horvitz et al., 2003).
• Attention and Information Contextualisation:– Attention is currently applied to information presentation.
Attention:The cognitive process of selectively concentrating on one aspect of the environment while ignoring other things.
Attention-Streams
Attention-Streams
Attention vs. Interests
• Attention models can be used for recommending information:– Attention Interests / Interests Attention
• Cross-domain Recommendations:– Attention is community independent.
• Real-time recommendations:– Attention is real-time / Interests are not (e.g. Middleton, 2004).
• Ambient Recommendations:– Integration of the recommendations in the user workflow.– Passive application.
Recommender System:Type of information filtering system technique that attempts to recommend information items that are likely to be of interest to the user.
Attention-Streams
Modelling Attention using Attention-Streams
– Attention Tag:• AT = {agent, timestamp,
domain, tag, weight (…)}
– Attention:• AT = {agent, timestamp,
AT set (…)}
Attention-Stream:The continuous flow of interrelated information accessed through the Attention of an Agent.
Attention
Attention Tags
Attention-Streams
Attention Tag
• Attention is represented using lightweight semantics and weighted tags (APML Ontology).– Each web document has corresponding attention tags. – Attention-Tags might be linked to FOAF profiles.
curio: Document
curio: Agent
Attention-Streams
Attention
• At a specific instant, the attention of an Agent is characterized by a set of Attention Tags.– Attention exists across domains.
Model:• Attention-Tag Similarity:
— WordNet, PMI, NSS (NGD (Cilibrasi, 2004))...
• Attention-Range Affinity.• Attention-Range Calculation:
— Affinity-Gradient, EMA…
politics word wide web AJAX
computing
Attention-Streams
Monitoring AttentionMedia Extraction Service
Attention-Streams
Attention Based RecommendationsMedia Extraction Service
Attention-Streams
Demohttp://nebula.dcs.shef.ac.uk/sparks/astreams
Attention-Streams
Conclusions
• Attention-Streams Recommendations:– Contextual and Real-time information recommendations.– Real-time interests modelling and sharing.
• Interests derived from user attention.– Ambient recommendations.
Full Live Demo:Tomorrow, during the poster and demo session.
Attention-Streams
Conclusions
• Attention-Streams Recommendations:– Contextual and Real-time information recommendations.– Real-time interests modelling and sharing.
• Interests derived from user attention.– Ambient recommendations.
• Future work:– More recommendations ! (i.e: Social).– Integration with streaming ontologies and models (i.e: Sensor
Streams).– More Attention bookmarking.
Full Live Demo:Tomorrow, during the poster and demo session.