Θεωρητική μελέτη των συνειρμικών τεχνικών ανάκτησης προκειμένου να υπάρξει βελτίωση στο πρόβλημα ανεπάρκειας του συνεργάσιμου φιλτραρίσματος
Theoretical study of associative retrieval techniques in order to improve the sparsity problem of collaborative filtering
The aim of this master thesis is the study and application of associative retrieval techniques that result is to have an improvement in sparsity problem. The recommendation systems are widely applied many installed applications. So as to offer to potential customers, products, services and information they need. Collaborative filtering, which is perhaps the most successful recommendation approach classifying them based on past experience and therefore no feedback from consumers who shared the same interests.A major drawback limiting the use of the co filtering is sparsity problem refers to a situation in which the transactions or data feedbacks are sparse and insufficient to have identification of similarities in consumer interests. n this thesis we propose considering the sparsity having problems posting using the application framework associative retrieval relatively disseminated in order to investigate the transitional links between consumers based on their previous transactions and then place the appropriate feedback. Such transitional compounds are a valuable source of information is a valuable source of information that will help us to conclude consumer interests to be consideration of having problems posting sparsity. To make the evaluation of effectiveness of our approach, we will be carrying out an experimental study he used a data set from an on-line bookstore. Research theoretically done in three widespread activation algorithms including a capacitor leakage algorithm, a connected partial symbolic algorithm, and a Net parallel hop plantations search algorithm. Also made a report to the nearest neighbor algorithm and the algorithm of collaborative filtering to focus on the graphics The plantation relaxation search algorithm are compared with different filtering approaches where not review the transitional Union. A simple graph search approach,two variants of the use of comparison with the approach based a comparison variation element based approach. The experimental results help us to show that a comparison of disseminated activations with approaches based surpassed, to a very large extent, other filtering methods are measured based on the recommendation accuracy, reflection, the measure F and dense result. In addition we observe the effect of the trigger longitudinally, which means that after making integration transitional compounds were included element that was used in the past and not dilute the possibility of dilution. These data are used, so as to draw conclusions relating to user preferences, and ultimately lead to performance degradation recommendation.