Footnotes

  1. http://movielens.umn.edu/
  2. Datasets used in this thesis are available for download from http://rslab.movsom.com/data/ .
  3. Our recommendation technique "framework" which is used for the evaluations in this thesis is available for download from http://rslab.movsom.com/tools/rslab/ , it includes all of the recommendation techniques described in this thesis.
  4. Our public experimental recommender system MOVSOM can be found at http://www.movsom.com/ .
  5. http://trec.nist.gov/
  6. http://www.amazon.com/
  7. http://www.imdb.com/
  8. rec.arts.movies
  9. http://www.imdb.com/a2z/
  10. http://www.imdb.com/moka/
  11. http://www.allmovie.com/
  12. http://www.a9.com/
  13. Figures are taken from the website http://vv.carleton.ca/~neil/neural/neuron-a.php
  14. In optimization theory, the object of a computational problem is to find the best possible solutions among all feasible solutions which has the minimum (or maximum) value of an objective function. Objective functions can be thought of as cost functions which determines how good a solution is, for instance, the total cost of edges in the travelling salesman problem [sant04dictonary]
  15. The historical notes here are largely (but not exclusively) based on the following two sources: chapter 1 of the book by Haykin [haykin94neuralnetworks] and the introduction part in [arbib03handbook] . If the authors finds i neccessary, specific references will be made.
  16. Pattern recognition is one of the most common applications of a neural network. Pattern recognition is a combination of feature extraction and classification. Feature extraction results in feature vecors that are used as input vectors that are to be classified. Essentially, one may say that any application for neural networks can been transformed to a pattern recognition task even if the actual application wasn't pattern recognition, since the input to the network can be called patterns in any case
  17. A closed form (expression) is any formula that can be evaluated in a finite number of standard operations.
  18. Maths packages like MATLAB have inbuilt functions for doing such calculations, see the section entitled SOM Toolbox for MATLAB
  19. Can be downloaded at http://www.cis.hut.fi/projects/somtoolbox/
  20. MATLAB is a widely used programming environment for technical computing. The matrix is the main data structure used in MATLAB, so it is a very efficient tool for performing matrix calculations. MATLAB is developed by MathWorks Inc. and can be found at http://www.mathworks.com/
  21. WEBSOM can be found at http://websom.hut.fi/websom/
  22. Digital Equipment Corporation's Systems Research Center (SRC) originally developed EachMovie in order to gain experience with a collaborative filtering algorithm. EachMovie had been operated as a free public service for 18 months between the years 1995 and 1997. In September 1997 the EachMovie service was terminated. The dataset was then stripped of identifying data and made available so that collaborative filtering researchers could use the data to test their algorithms. The EachMovie Dataset remained available until October 2004 when it was finally retired.
  23. http://www.movsom.com/
  24. http://www.netzspannung.org/
  25. http://www.gnovies.com/