Українська   Русский
DonNTU   Masters' portal

Abstract

Introduction

Modern internet space provides the user with a huge range of information, which is becoming more difficult to navigate, so the use of conventional means to search and organize can not fully meet the user's needs: you can not view all the materials to choose for themselves suitable. In this regard, began to appear more and more so-called systems of recommendations that focus on the provision of information, most fully satisfying user interests and best suits his request.

1. Subject urgency

Recommendation system – programs that try to predict which objects (books, movies, music, web sites) may please the user with certain information about his profile. Such programs are usually used for commercial purposes (primarily at online stores or specialized sites "of interest" for the purpose of supply of goods). On the other hand, it is urgent of the search process on the Internet. Many Internet users believe objectively that modern impact opportunities search engine will not allow them to find the necessary documents or data. For such a user views a worldwide network of the following conditions occur:

– The explosive growth of data available in society in general (increasing the number of books, movies, news, advertisements, etc.);

– Increase of online data

– Actual volume of information, the human, is considerably better than what it can actually pass through itself, to discover the necessary and sufficient as well as his favorite.

Actual use of the system recommendations for online retailers. This allows the user to recommend a popular, high–quality goods that could interest him, or in the absence of any stock on hand to advise him of the requested analog products.

2. Research goals and objectives

Objective is to develop and study algorithmic support intelligent system to develop recommendations based on collaborative filtering techniques To achieve this goal it is necessary to solve the following problems:

Main tasks of the research:

  1. Review existing guidelines and calculation methods to prove the choice of methods based on collaborative filtering;
  2. To analyze the measure of proximity ( similarity ) users;
  3. Improved algorithm for calculating the recommendations.

3. Alleged scientific novelty

A modified algorithm for calculating recommendations based on collaborative filtering methods: the matrix will not count as four- dimensional. It besides objects and users will have 2 parameters: time and geolocation. Time will show what period of time users spend on the viewing of an object, and will be responsible for geolocation advisable to order goods (if no stock on hand in your city whether to order it from the other end of the country). The proposed algorithm can be implemented in any online store, for improving its work and provide more accurate information of its visitors.

4. Calculation methods recommendations

Calculation methods recommendations. As discussed below algorithms systems recommendations, the following definitions.

Object – a song, a movie, a product the user (in the case of the recommendation of friends) Ie that users consume system recommendations. This is what they should be encouraged.

User – a person registered in the system, he can buy, listen, watch, evaluate objects and use the service recommendations.

Recommendation – an object or multiple objects that the system gives the user recommendations.

Recommendation system allows a person to identify their tastes and returns results interesting for him, based on estimates from other users and their assumptions.

Unlike search engines, for a response from the system does not require a clear job request. Instead, the user is prompted to evaluate some of the objects from the collection, and on the basis of its assessments and compare them with previous estimates users speculate about the tastes of the user and returns the closest results to them, creating a personalized issue for him.

As a set of objects can be estimated, for example, be: directory of links to web sites, news, electronic goods store, a collection of books in the library, etc.

The scope of these systems also includes situations where the user is not looking for information on a specific keyword, and, for example, wants to get a list of contemporary articles on topics similar to those he looked up to this.

5. Сollaborative filtering

To meet the challenges set out in the paper, we propose to use collaborative filtering method, so consider this method in more detail.

Сollaborative filtering

Figure 2 – Сollaborative filtering
(animation: 17 frames, 5 cycles of repeating, 647 kilobytes)

Collaborative filtering (Collaborative filtering) – a method of recommendations for which only analyzed response of users to objects. People leave objects in the evaluation system . Moreover, the evaluation can be both explicit (eg assessment on a five-point scale) and implicit (eg, views of one roller). The ultimate goal of this method is how to assess a more accurate prediction , which would put the current user of the system they previously unappreciated objects. The more assessments collected, the more accurate the recommendations obtained. By is obtained, users help each other to filter objects. Therefore, this method is also called collaborative filtering.

Conclusion

In this paper we summarize the main algorithms used in systems recommendations. Different methods use a variety of data about users and about the objects. Each approach has its advantages and disadvantages. For example, the method of collaborative filtering recommends objects, having no idea. what they represent Issue recommendations of new facilities decide the methods of analysis of content. But for their good work required text data about the objects. If information about users, objects, and estimates enough for these algorithms applied methods utilizing knowledge base. While interactively identified user requirements.

The more data available, the exact system bases its recommendations can be developed.

This master's work is not completed yet. Final completion: December 2014. The full text of the work and materials on the topic can be obtained from the author or his head after this date.

References

  1. Jannach D., Zanker M., Felfernig A., Friedrich G. Recommender Systems / An Introduction. New York: Cambridge University Press 32 Avenue of the Americas,  — 2011. — 352 p.
  2. Гомзин А., Коршунов А. Системы рекомендаций: обзор современных подходов // Препринт. Москва: Труды Института системного программирования РАН. 2012. 20 c.
  3. Melville P., Sindhwani V. Recommender systems. Encyclopedia of Machine Learning. — 2010.
  4. Su X., MagyhT. Khoshgoftaar Survey of Collaborative Filtering Techniques / Advances in Artificial Intelligence. — 2009.
  5. Ansari A., Essegaier S., Kohli R., “Internet Recommendations Systems,” / J. Marketing Research, — Aug. 2000. — pp. 363–375.
  6. Resnick P., Iacovou N., Sushak M., Bergstrom P., Riedl J. "GroupLens: An Open Architecture for Collaborative Filtering of Netnews," / Proceedings of the 1994 Computer Supported Cooperative Work Conference, ACM, — 1994.
  7. Konstan J., Miller B., Maltz D., Herlocker J., Gordon L., RiedlJ. "GroupLens: Collaborative Filtering for Usenet News," / to appear in Communications of the ACM special issue on collaborative filtering, — March 1997.
  8. Miller B., Riedl J., Konstan J. "Experiences with GroupLens: Making Usenet Useful Again," / Proceedings of the Usenix 1997 Winter Technical Conference, Anaheim, CA, — January 1997.
  9. Арбузова О.В. Разработка и исследование алгоритмов для повышения эффективности интеллектуального анализа web-контента // реферат выпускной работы магистра Факультет вычислительной техники и информатики ДонНТУ. 2013.