Contextual Recommendation of Social Updates, a tag-based framework

1,521 views

Published on

How to cope with information overload?

In this presentation (and the corresponding paper), we propose a framework to improve the relevance of awareness information about people and subjects, by adapting recommendation techniques to real-time web data, in order to reduce information overload. The novelty of our approach relies on the use of contextual information about people's current activities to rank social updates which they are following on Social Networking Services and other collaborative software. The two hypothesis that we are supporting in this paper are: (i) a social update shared by person X is relevant to another person Y if the current context of Y is similar to X's context at time of sharing; and (ii) in a web-browsing session, a reliable current context of a user can be processed using metadata of web documents accessed by the user. We discuss the validity of these hypothesis by analyzing their results on experimental data.

Presented by Adrien Joly, on the 28/08/2010, at the Active Media Technology (AMT) conference, Toronto, Ontario, Canada.

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,521
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
30
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • w t,d is the number of people p who annotated a given resource d using the term t .
  • Counts the occurrences of each term that is semantically identified in the document’s content.
  • Contextual Recommendation of Social Updates, a tag-based framework

    1. 1. Contextual Recommendation of Social Updates a tag-based framework Adrien JOLY PhD Candidate, supervisor: Prof. Pierre MARET Alcatel-Lucent Bell Labs France + INSA-Lyon, LIRIS, UMR5205 [email_address] / [email_address]
    2. 2. Agenda of this presentation <ul><li>Motivation — Awareness and information overload </li></ul><ul><li>Approach — Context-based filtering </li></ul><ul><li>Framework — Contextual tag clouds </li></ul><ul><li>Evaluation — Perceived relevance </li></ul><ul><li>Conclusion & future work </li></ul>
    3. 3. Motivation Approach Framework Evaluation Conclusion Introduction to Awareness <ul><li>Awareness is the state or ability to perceive, to feel, or to be conscious of events, objects or sensory patterns […] without necessarily implying understanding. </li></ul><ul><li>[wikipedia.org] </li></ul><ul><li>Social Awareness </li></ul><ul><li>current/recent people activities, moods, availability, status… </li></ul><ul><li>[Dourish, Ericksson, Gutwin…] </li></ul><ul><li>Context Awareness </li></ul><ul><li>location, surrounding environment… [Dey’2000] </li></ul>
    4. 4. Motivation Approach Framework Evaluation Conclusion Web « 2.0 » social / communication tools <ul><li>Social Networking Platforms increase Social Awareness </li></ul>
    5. 5. Motivation Approach Framework Evaluation Conclusion Web « 2.0 » social / communication tools <ul><li>Social Networking Platforms increase Social Awareness </li></ul>… through Social Updates
    6. 6. Motivation Approach Framework Evaluation Conclusion Web « 2.0 » social / communication tools <ul><li>Social Networking Platforms increase Social Awareness </li></ul><ul><li>But it can steal a lot of attention  productivity loss </li></ul>
    7. 7. Motivation Approach Framework Evaluation Conclusion Our proposal Filter “ Aware” user Activities / Status Updates / Contacts Needed <ul><li>Social updates </li></ul>and productive
    8. 8. Agenda of this presentation <ul><li>Motivation </li></ul><ul><li>Approach </li></ul><ul><li>Framework </li></ul><ul><li>Evaluation </li></ul><ul><li>Conclusion </li></ul>
    9. 9. Motivation Approach Framework Evaluation Conclusion Filtering possibilities <ul><li>Motivated goal: </li></ul><ul><ul><li>Filter social network updates to enable awareness without information overload </li></ul></ul><ul><li>What criteria should we adopt to find the most relevant updates ? </li></ul><ul><ul><li>Popularity ? (most spread updates) </li></ul></ul><ul><ul><li>Response rate ? (most commented updates) </li></ul></ul><ul><ul><li>Content-based filtering ? (according to preferences) [Budzik’2000, Bauer’2001] </li></ul></ul><ul><ul><li>Collaborative filtering ? (according to similar ratings) [Agosto’2005, Bielenberg’2005] </li></ul></ul><ul><ul><li>Similarity of context </li></ul></ul>
    10. 10. Motivation Approach Framework Evaluation Conclusion Similarity of context, our hypothesis <ul><ul><li>C A is the context of a user U A sharing a piece of information I A . </li></ul></ul><ul><ul><li>C X is the context of a user U X that is a potential recipient of this information. </li></ul></ul>Hypothesis: I A is relevant to U X if C A is similar to C X A A = Travel in Asia U A = Alice I A = « Check out my amazing picture ! » A B = Working Java U B = Bob I B = « What database should I use ? » A C = Browsing map U C = Christine I C = « Looking for holiday locations… »
    11. 11. Motivation Approach Framework Evaluation Conclusion Similarity of context, our hypothesis <ul><ul><li>C A is the context of a user U A sharing a piece of information I A . </li></ul></ul><ul><ul><li>C X is the context of a user U X that is a potential recipient of this information. </li></ul></ul>Hypothesis: I A is relevant to U X if C A is similar to C X C A = Travel, Asia C C = Travel C B = Java Dev. A A = Travel in Asia U A = Alice A B = Working Java U B = Bob I B = « What database should I use ? » A C = Browsing map U C = Christine I C = « Looking for holiday locations… » Similar context: travel No relevant match for this context I A = « Check out my amazing picture ! »
    12. 12. Motivation Approach Framework Evaluation Conclusion What is context ? <ul><li>Context [Dey, 2001] : </li></ul><ul><li>«  any information that can be used to characterize the situation of an entity  » </li></ul><ul><ul><li>From physical sensors: </li></ul></ul><ul><ul><li>From computer-based actions: </li></ul></ul>Location Surrounding people Other sensors Communication history Web browsing history Document history
    13. 13. Motivation Approach Framework Evaluation Conclusion From sensors to applications Context sensors Applications Interpretation Acquisition db <ul><li>Usual representation scheme for context information: </li></ul><ul><li>Ontology-based </li></ul><ul><li>/ semantic </li></ul><ul><ul><li>Requires ont. modeling </li></ul></ul><ul><ul><li>Lack of semantic data </li></ul></ul><ul><ul><li>Complex to manipulate </li></ul></ul><ul><ul><li>Scaling issues </li></ul></ul>Context Management Framework
    14. 14. Motivation Approach Framework Evaluation Conclusion From sensors to applications Context Management Framework Context sensors Social Applications Interpretation Acquisition db <ul><li>Proposed representation scheme for context information: </li></ul><ul><li>Contextual tag clouds </li></ul><ul><ul><li>Easy to browse </li></ul></ul><ul><ul><li>Easy to edit </li></ul></ul><ul><ul><li>Simple & interoperable </li></ul></ul><ul><ul><li>Crowds-friendly </li></ul></ul>Updates Paris Notre-Dame Café Cloudy Crowded Sitting with:Pierre
    15. 15. Agenda of this presentation <ul><li>Motivation </li></ul><ul><li>Approach </li></ul><ul><li>Framework </li></ul><ul><li>Evaluation </li></ul><ul><li>Conclusion </li></ul>
    16. 16. Motivation Approach Framework Evaluation Conclusion Context Aggregation and Filtering process Social updates Aggregator Sniffers Notifier Filter User Actions and tags Contextual clouds Notifications Context Interfaces Abstraction and weighting Services
    17. 17. Motivation Approach Framework Evaluation Conclusion Context Aggregation and Filtering process –- in the enterprise Social updates Aggregator Sniffers Notifier Filter User Actions and tags Contextual clouds Notifications Context Interfaces Abstraction and weighting Services
    18. 18. Motivation Approach Framework Evaluation Conclusion How to synthesize the contextual tag cloud from web browsing ? The user opens a web page…
    19. 19. Motivation Approach Framework Evaluation Conclusion How to synthesize the contextual tag cloud from web browsing ? Low level and static author description Automatic content analysis Mining semantic concepts from content People-entered tags (wisdom of crowds) 1) URL is sent to the Context Aggregator 2) Content is analyzed by enhancers (including web services)
    20. 20. Motivation Approach Framework Evaluation Conclusion Contextual Tag Clouds, vector space model and algebra Sample tag cloud R : (normalized) <ul><li>Aggregation of a set V of normalized Tag Clouds </li></ul><ul><li> normalized sum: </li></ul><ul><li>Relevance of Tag Cloud R with S </li></ul><ul><li> cosine similarity: </li></ul>0.1 0.1 0.3 0.5 « Discount » « Flight » « Asia » « Travel »
    21. 21. Motivation Approach Framework Evaluation Conclusion Contextual Tag Clouds, extraction and enhancement functions <ul><li>1. Extracting weighted terms from: Resource Metadata </li></ul>Title Keywords Description Parameters = 50 = 10 = 1
    22. 22. Motivation Approach Framework Evaluation Conclusion Contextual Tag Clouds, extraction and enhancement functions <ul><li>2+3. Extracting weighted terms from: </li></ul>2. Search Query ambient, awareness 3. Resource Location video, all, alcatel-Lucent
    23. 23. Motivation Approach Framework Evaluation Conclusion Contextual Tag Clouds, extraction and enhancement functions <ul><li>Extracting weighted terms from: Social Annotations </li></ul>w poster = 11, w work = 11, w gtd = 10, w done = 10, w inspiration = 7, …
    24. 24. Motivation Approach Framework Evaluation Conclusion Contextual Tag Clouds, extraction and enhancement functions <ul><li>Extracting weighted terms from: Semantic Analysis of content </li></ul>MIT, Tim Berners-Lee, …
    25. 25. Agenda of this presentation <ul><li>Motivation </li></ul><ul><li>Approach </li></ul><ul><li>Framework </li></ul><ul><li>Evaluation </li></ul><ul><li>Conclusion </li></ul>
    26. 26. Motivation Approach Framework Evaluation Conclusion Requirements and plan <ul><li>Hypothesis: </li></ul><ul><ul><li>Recommended social updates are relevant when users’ contexts are similar </li></ul></ul><ul><li>To evaluate: </li></ul><ul><ul><li>Tag cloud similarity for relevance ranking </li></ul></ul><ul><ul><li>Relevance of social updates to the context of their posting </li></ul></ul><ul><li>Experimentation plan: </li></ul>(1 week)  1 tag cloud every 10 minutes  2 personalized surveys per user
    27. 27. Motivation Approach Framework Evaluation Conclusion From browsing activity to social matching Temporal indexing period = 10 mn. Common tags: JAVA, DEV Common tags: TRAVEL  Recommend u5’s social update to u1  Recommend u3’s social update to u7
    28. 28. Motivation Approach Framework Evaluation Conclusion Survey #1 <ul><li>… and 3 social updates with various relevance scores, for each context </li></ul>upd1 upd2 1 2 3 4 1 2 3 4 <ul><li>Survey #1 : For each user, 5 personal contextual clouds are proposed… </li></ul>
    29. 29. Motivation Approach Framework Evaluation Conclusion Survey #1 results 1/2 <ul><li> rarity of good matches </li></ul><ul><li>(few participants  few common tags) </li></ul>
    30. 30. Motivation Approach Framework Evaluation Conclusion Survey #1 results 2/2 <ul><li> Accuracy = 72% (based on MAE between relevance scores and ratings) </li></ul>Accuracy
    31. 31. Motivation Approach Framework Evaluation Conclusion Survey #2 <ul><li>Survey #2 : For each user’s social update, Evaluation of relevance between social updates and context of posting </li></ul>rating <ul><li>Results </li></ul><ul><li>Average relevance rating: 50.3% (over 59 social updates), including: - 71% for social bookmark notifications - 38% for tweets ( ≈ 41% of “me now” statuses on twitter [Naaman’2010]) </li></ul>1 2 3 4
    32. 32. Agenda of this presentation <ul><li>Motivation </li></ul><ul><li>Approach </li></ul><ul><li>Framework </li></ul><ul><li>Evaluation </li></ul><ul><li>Conclusion </li></ul>
    33. 33. Motivation Approach Framework Evaluation Conclusion Contribution <ul><ul><li>Goal : </li></ul></ul><ul><ul><li>Increase awareness, reduce information overload </li></ul></ul><ul><ul><li>Proposition : </li></ul></ul><ul><ul><li>Use contextual information to rank relevance of social updates </li></ul></ul><ul><ul><li>Approach : </li></ul></ul><ul><ul><li>Tag-based context representation, instead of ontology-based </li></ul></ul><ul><ul><li>Findings (using web browsing activity as context) : </li></ul></ul><ul><ul><ul><li>Encouraging results: 72% accuracy </li></ul></ul></ul><ul><ul><ul><li>Half social updates are relevant to web browsing context, depending on nature </li></ul></ul></ul>
    34. 34. Motivation Approach Framework Evaluation Conclusion Future work <ul><ul><li>Improve quality of contextual tag clouds </li></ul></ul><ul><ul><ul><li>Semantic analysis, clustering, and filtering of tags </li></ul></ul></ul><ul><ul><ul><li>Dynamic weights (based on time) </li></ul></ul></ul><ul><ul><li>Deeper study of social updates </li></ul></ul><ul><ul><ul><li>Relevance factors between specific social update and contextual properties </li></ul></ul></ul><ul><ul><li>Gather context from other sources </li></ul></ul><ul><ul><ul><li>Additional types of documents (e.g. emails, PDF/word documents…) </li></ul></ul></ul><ul><ul><ul><li>Physical context information </li></ul></ul></ul><ul><ul><li>Develop a contextual tag cloud manipulation interface (HSI) </li></ul></ul><ul><ul><ul><li>Graphical extension, multidimensional/hierarchical tag cloud ? </li></ul></ul></ul><ul><ul><ul><li>How to edit tags and their weights ? </li></ul></ul></ul>
    35. 35. www.alcatel-lucent.com Thank you for your attention! Your questions are welcome 

    ×