ISSN 2587-814X (print),
ISSN 2587-8158 (online)

Russian version: ISSN 1998-0663 (print),
ISSN 2587-8166 (online)

Alexander D’yakonov1
  • 1 Lomonosov Moscow State University; Dorodnitsyn Computing Center, Russian Academy of Sciences , 1, build. 52, Lomonosov Moscow State University, Leninskie Gory, GSP-1, Moscow, 119991, Russian Federation

Algorithms for a recommender system: LENCOR technology

2012. No. 1(19). P. 32–39 [issue contents]

Alexander D’yakonov – Associate Professor, Department of Mathematical Methods of Forecasting, Faculty of Computational Mathematics and Cybernetics, Lomonosov Moscow State University.
Address: 1, build. 52, Lomonosov Moscow State University, Leninskie Gory, GSP-1, Moscow, 119991, Russian Federation.
E-mail: djakonov@mail.ru

The article describes algorithms that took first places in International ECML/PKDD Discovery Challenge 2011 and are focused on development of recommendation systems for VideoLectures.Net website, a multimedia repository of video lectures (VideoLectures.Net Recommender System Challenge). The study describes all alternative problem-solving methods that can be divided into two groups: collaborative filtering and content-based (information filtering) methods.

The first group uses user behavior statistics (e.g. they recommend goods and services that were of interest of similar users); the second group uses description of goods and services (e.g. they recommend goods and services of the same category, price range, related products, etc.). It is possible to use simultaneously methods of both groups (hybrid prediction) and algorithms based on a priori knowledge of user needs (knowledge-based).

LENKOR methods analyzed in this study are focused on tasks with complex object setting (using different characteristic features and/or non-feature description), relatively small samples (insufficient to use statistic methods) and non-traditional functionality of algorithms. It was proposed to introduce objects similarity functions (each function evaluates similarity using its own information type), to create a general similarity formula (usually, in the form of linear combination of functions introduced) and a way of answer receipt. Then it is required to set up an algorithm and amend a general formula adding all nonlinearities.

Described algorithms are simple and universal and concede paralleling. The solution comes in the form of evaluation vector. To recommend a number of lectures, it is enough to select some largest vector elements. At the same time it is possible to obtain popularity rating for each lecture. LENKOR technology is based on ideas of algebraic approach: correct evaluation vector space is selected, and then algebraic expression is set.

Proposed methods can be used in tasks of other types, e.g. Cold Start solution algorithm can be easily adapted for loan scoring and project evaluation tasks.

Citation: D’yakonov A. G. (2012) Algoritmy dlia rekomendatel'noi sistemy: tekhnologiia LENCOR [Algorithms for a recommender system: LENCOR technology] Biznes-informatika, 1(19), pp. 32-39 (in Russian)
Rambler's Top100 rss