Download presentation
Presentation is loading. Please wait.
Published byKaya Bainum Modified over 10 years ago
1
1 Whats Up: P2P news recommender Anne-Marie Kermarrec Joint work with Antoine Boutet, Davide Frey (INRIA) and Rachid Guerraoui (EPFL) Gossple workshop 2010
2
2 The social Web Web content is generated by you, me, your friends and millions of others The Web has turned social
3
Content comes from everywhere Gossple workshop 20103
4
Is it equally relevant? Gossple workshop 20104
5
Is it equally relevant? Gossple workshop 20105
6
Is it equally relevant? Gossple workshop 20106
7
Whats wrong with news feed? Amazon recommends me a fryer Some of my Facebook write in Italian LeMonde.fr wants to inform me on the Champions ligue Gossple workshop 20107
8
Why is it so difficult? Even a space restricted to users explicit subscriptions is too large a database Dynamic Recommendations not always user-centric Explicit links not always that relevant Classical pub/sub do not filter enough Granularity of a user seems too coarse Gossple workshop 20108
9
Cascading over explicit links Gossple workshop 20109
10
Fine grain tuning calls for decentralisation Gossple workshop 201010
11
Whats up Decentralised information dissemination channel Simple interface: I like it or I dont Exploit implicit social links Gossple workshop 201011
12
Gossple workshop 201012
13
Whats up in a nutshell Gossple workshop 201013
14
Whats up challenges Gossple workshop 201014 Who are my social acquaintances How to discover them? How to disseminate news ? Similarity metric Through gossip Biased epidemic protocol
15
Whats up: Gossple net Gossple workshop 201015
16
Whats up challenges Gossple workshop 201016 Who are my social acquaintances How to discover them? How to disseminate news ? Similarity metric
17
An implicit social network Gossple workshop 201017
18
Which nodes should be considered as social acquaintances? Model U(sers) × I(tems) (news) Profile(u) = vector of liked news Minimal information Similarity metrics Overlap Cosine similarity Multi-interest similarity Gossple workshop 201018
19
Item cosine similarity 19Gossple workshop 2010 Normalized overlap Profile(u)= Vector of news Items{u}
20
Individual rating might be too restrictive Rate the set of users instead of individuals 20 Items of interest for nodes in set(n) Items of interest for nodes in set(n) Distribution Normalized not to take into account non shared interests Normalized not to take into account non shared interests Gossple workshop 2010
21
Whats up challenges Gossple workshop 201021 Who are my social acquaintances How to discover them? How to disseminate news ? Through gossip
22
22 The Gossple network Gossple workshop 2010 Copyright: E. Rivière Gossip similarity protocol. Gossip-based peer sampling service
23
Gossple social network Gossple workshop 201023 @IP: port102.14.18.1:2110 Bloom Filter100100000110 Update time30 @IP:port 132.154.8.5:2020 Profile234527 387690 672986 Update time5 Friends Uniform sample c entries k entries
24
Gossple workshop 201024 Building the social network Two gossip protocols Similarity-based Peer Sampling Random Peer Sampling When p encounters q Evaluate potential new view, based on set similarity metric Use of Bloom filters to limit the communication overhead RPS SPS RPS SPS
25
Whats up in a nutshell Gossple workshop 201025
26
Whats up challenges Gossple workshop 201026 Who are my social acquaintances How to discover them? How to disseminate news ? Biased epidemic protocol
27
Dissemination Gossple workshop 201027 Heterogeneous Homogeneous HeterogeneousHomogeneous Involvement (fanout) Expectations Epidemic Dissemination F=log(N) Heterogeneous Gossip F log(N) on average
28
BEEP: orientation and amplification Orientation: to whom? Gossple workshop 201028 Forward to friends Forward to random Amplification: to how many? Increase fanout Decrease fanout
29
Beep: I like it Gossple workshop 201029 I like it!
30
Beep: I dont Gossple workshop 201030 I dislike it!
31
Tuning BEEP Orientation The news carries the list of visited users A profile: sum of interests of users who liked it Amplification F log(N) friends Amplification depends on the similarity between the news and the user F 1 or 2 random Gossple workshop 201031
32
Evaluation User Metrics Spam Recall Precision System metric Number of messages Redundancy (useless messages) Traces Synthetic clustered traces Real dataset: 700 Digg users/2000 news/1 week Gossple workshop 201032
33
Preliminary results Gossple workshop 201033 AlgorithmPrecisionRecallSpam Perfect110 Gossip fanout=log(n)=7 0.280.940.74 Cascading through explicit friends from Digg 0.390.71 WhatsUp fanout=11/1 ; ttl=12 0. 520.6 WhatsUp without no social users
34
To take away Automatic light news recommender Analysis through mean field theory Experimental evaluation Next: diversity of sources, trust, privacy Gossple workshop 201034
35
Thank you 35 www.gossple.fr Gossple workshop 2010
36
36
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.