Summary

Method for collaborative-filtering tasks. We base the rating of item for user by checking ratings of users with similar tastes

License

CC-BY-SA 3.0

Tags

collaborative-filtering item-based neighborhood recommender systems

Feature Processing

No feature processing required, however standard outlier removal might be helpful.

Parameters

matrix of user/item ratings

Operating System

independent

Code

function X = itembased( y,M,sizeX,swap,sim,neib) % Neighbour based regression algorithm % calculates similarities between items and than rating for u as % r_i(u) = sum_j sim(i,j) * r_j(u) / sqrt(sim(i,i) * sim(j,j)

% Output % X - matrix to be estimated % Input % y - sampled entries % M - masking operator, applied to vectorized form of X % sizeX - size of matrix to be reconstructed % swap - switch to ITEM-BASED (default = 1) % sim - similarity measure (0 - PC, 1 - WPC, 2 - cos sim, default = 0) % neib - percentage of neibours taken into sum

% Copyright (c) ?ukasz Kidzi?ski 2011

Y = reshape(M(y,2),sizeX);

if nargin < 4
    swap = 1;
end
if nargin < 5
    sim = 0;
end
if nargin < 6
    neib = 0.8;
end
if swap
    Y = Y';
end;

if (sim < 2)
    corr = pearsonsim(Y,sim);
else
    corr = cosinesim(Y);
end;

corr(corr < 0) = 0;

rated = Y>0;
usedcorr = corr * rated';
X = corr * Y';
X = X  ./ usedcorr;
X(usedcorr==0) = mean(y);
X = X';

if (swap)
    X = X';
end;

end

Software Packages

matlab

Description

(No information yet)

URLs

http://students.mimuw.edu.pl/~lk234151/works/mgr-math.pdf

Publications

revision 1

by kidzik on 2011-09-13 19:58

revision 2

by kidzik on 2011-09-13 20:09

No one has posted any comments yet. Perhaps you would like to be the first?

Leave a comment

This item was downloaded 0 times and viewed 5804 times.