A Recommender System based on the Immune Network

1
A Recommender System based on the Immune Network
Steve Cayzer, HP Labs Bristol Uwe Aickelin,
University of the West of England, Bristol

• The Recommendation Problem
• The AIS Approach
• Walk through algorithm
• Results and Discussion

2
The Recommendation Problem What movies would
you predict/recommend?

• Prediction
• What rating would I give this film?
• Prediction quality can be assessed by absolute
  error

• Recommendation
• Give me a top 10 list of films I might like
• Recommendation quality can be assessed by a
  ranking discordance metric

3
The Recommendation Problem
EachMovie database User profiles User Profile
set of tuples movie, rating Me My user
profile Neighbour User profile of someone else
Similarity metric Correlation score between
user profiles Neighbourhood Group of neighbours
similar to me Recommendations generated from
neighbourhood
4
The AIS Approach
EachMovie database User profiles User Profile
set of tuples movie, rating Me My user
profile Neighbour User profile of someone else
Similarity metric Correlation score between
user profiles Neighbourhood Group of neighbours
similar to me Recommendations generated from
neighbourhood
Antigen
Antibody
Stimulation
Suppression
Antibody Antigen Binding
Antibody Antibody Binding
Group of antibodies similar to antigen and
dissimilar to other antibodies
5
Algorithm walk through Encoding
Suppose we have 5 users and 4 movies
DATABASE u1(m1,v11),(m2,v12),(m3,v13) u2(m1,v
21),(m2,v22),(m3,v23),(m4,v24) u3(m1,v31),(m2,v
32),(m4,v34) u4(m1,v41),(m4,v44) u5(m1,v51),
(m2,v52),(m3,v53), (m4,v54)

• We do not have user votes for every film
• We want to predict the vote of user u4 on
  movie m3

6
Algorithm walk through (1)

• Start with empty AIS

DATABASE u1, u2, u3, u4, u5
Encode user for whom to make predictions as an
antigen Ag
DATABASE u1, u2, u3, u4, u5
Ag
7
Algorithm walk through (2)

• Add antibodies until AIS is full

Add next user as an antibody Ab
DATABASE u1, u2, u3, u4, u5
Ab1
8
Algorithm walk through (3)
After some more iterations the AIS has filled
up
Table of matching Scores between Antibodies and
Antigen MS14, MS24, MS34
Table of matching Scores between Antibodies MS12
CorrelCoef(Ab1, Ab2) MS13 CorrelCoef(Ab1,
Ab3) MS23 CorrelCoef(Ab2, Ab3)
9
Algorithm walk through (4)

• AIS is now at full size so begin iterations

Calculate a new CONCENTRATION for each Antibody,
considering interactions with the antigen
(stimulation) and other antibodies (suppression)
Ag
Notice that antibody 3 has been eliminated
10
Algorithm walk through (5)
If AIS not yet full and more reviewers available,
start the whole process again. Otherwise
GENERATE RECOMMENDATION from CONCENTRATION and
ANTIGEN MATCH.
Recommendation for user u4 on movie m3 will be
highly based on vote on m3 of user u2
11
Results 1 Stimulation and Suppression affect
neighbourhood size
12
Results 2 Simple AIS matches Simple Pearson
performance
13
Results 3 Small amounts of suppression improve
recommendation quality
14
Evaluation

• General purpose AIS tool
• Easily extensible
• Idiotypic effects for more varied population
• Potential for distribution
• Wider applicability
• BUT
• Not the killer app for CF
• Sensitivity to parameters
• Computationally expensive

15
Speculation

• Implications for online community formation
• Idiotypic effects alter nature of community
• How important is diversity?
• Are there other network effects that can be
  used? (hubs, routers etc)
• Distribution the snowball effect
• What about interacting communities?
• Application areas ad-hoc community formation,
  knowledge management, P2P routing

16
Discussion
BICAS site http//www.hpl.hp.com/research/bicas/
Steve Cayzer http//www-uk.hpl.hp.com/people/ste
cay/ Uwe Aickelin http//www.aickelin.com/