Loading...
Publication
Abordagem multiagente em sistemas de recomendaĆ§Ć£o Web
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.56 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O crescimento exponencial da informaĆ§Ć£o disponĆvel na Web torna difĆcil para os utilizadores a tarefa de obter a informaĆ§Ć£o que pretendem e quando dela necessitam. Para ultrapassar o problema, os sĆtios Web tĆŖm vindo a incorporar sistemas de recomendaĆ§Ć£o que, baseados no histĆ³rico de acessos, tĆŖm como objetivo maximizar a satisfaĆ§Ć£o dos utilizadores, disponibilizando-lhes recomendaƧƵes de alta qualidade.
A complexidade do problema e a natureza distribuĆda da Web justificam abordagens baseadas na tecnologia dos agentes inteligentes autĆ³nomos e sistemas multiagente, permitindo combinar mĆŗltiplos algoritmos de recomendaĆ§Ć£o, aumentando assim as hipĆ³teses das recomendaƧƵes sugeridas serem efetivamente do interesse do utilizador. Ć este o tipo de abordagem explorada pelo sistema de recomendaĆ§Ć£o multiagente AMAAFWA (A Multi-Agent Approach for Web Adaptation) (Morais, 2013). Os testes realizados em modo offline mostraram que essa abordagem multiagente, baseada em agentes implementando diferentes algoritmos, apresenta um desempenho superior ao dos algoritmos considerados individualmente.
O objetivo desta dissertaĆ§Ć£o Ć© adaptar e testar o sistema AMAAFWA em tempo real, com o objetivo de validar os resultados obtidos em modo offline, pelo que se procedeu Ć sua adaptaĆ§Ć£o para funcionamento online, integrando-o num sĆtio Web. O sistema AMAAFWA baseia-se numa classificaĆ§Ć£o implĆcita dos itens e os algoritmos de recomendaĆ§Ć£o sĆ£o baseados em memĆ³ria e incrementais. Foi tambĆ©m criada e testada uma versĆ£o do sistema que considera uma classificaĆ§Ć£o explĆcita dos itens por parte dos utilizadores, com o propĆ³sito de comparar o desempenho de ambos os tipos de classificaĆ§Ć£o.
Demonstra-se na presente dissertaĆ§Ć£o que o sistema de recomendaĆ§Ć£o multiagente AMAAFWA, em funcionamento online, apresenta um desempenho superior ao dos algoritmos considerados individualmente, sendo ainda capaz de melhorar a satisfaĆ§Ć£o do utilizador e contribuir para o aumento do sucesso do sĆtio Web em que se insere. Relativamente Ć comparaĆ§Ć£o dos tipos de classificaĆ§Ć£o implĆcita e explĆcita dos itens, os resultados mostram desempenhos similares.
The exponential growth of information available on the Web makes it difficult for users to get the information they want and when they need it. To overcome the problem, the Web sites are using recommender systems in order to provide high-quality recommendations to the users and, in that way, improve user satisfaction. The complexity of the problem and the distributed nature of Web justify the use of the autonomous intelligent agents and multi-agent systems technology approaches, which allows the combination of multiple recommendation algorithms in order to increase the chances of the suggested recommendations to be actually of interest to the users. The multi-agent recommender system AMAAFWA (A Multi-Agent Approach for Web Adaptation) (Morais, 2013) explores this approach. The results of the tests performed offline showed that this multi-agent approach, based on agents implementing different algorithms, has a higher performance when compared to individual algorithms. The goal of this dissertation is to adapt and test the AMAAFWA system in real-time operation, in order to validate the results obtained in offline mode. So, we adapted the system for online operation and integrate it on a website. The AMAAFWA system is based on implicit classification of items and the recommendation algorithms are memory and item-based and incremental. It was also built and tested a version of the system that considers explicit classification of items by users, with the aim of comparing the performance of both types of classification. It is shown in this dissertation that the multi-agent recommender system AMAAFWA, in online and real-time operation, has a higher performance when compared to individual algorithms, being able to improve user satisfaction and contribute to the increasing success of the website. Concerning the comparison between implicit and explicit classification, the results show a similar performance for both.
The exponential growth of information available on the Web makes it difficult for users to get the information they want and when they need it. To overcome the problem, the Web sites are using recommender systems in order to provide high-quality recommendations to the users and, in that way, improve user satisfaction. The complexity of the problem and the distributed nature of Web justify the use of the autonomous intelligent agents and multi-agent systems technology approaches, which allows the combination of multiple recommendation algorithms in order to increase the chances of the suggested recommendations to be actually of interest to the users. The multi-agent recommender system AMAAFWA (A Multi-Agent Approach for Web Adaptation) (Morais, 2013) explores this approach. The results of the tests performed offline showed that this multi-agent approach, based on agents implementing different algorithms, has a higher performance when compared to individual algorithms. The goal of this dissertation is to adapt and test the AMAAFWA system in real-time operation, in order to validate the results obtained in offline mode. So, we adapted the system for online operation and integrate it on a website. The AMAAFWA system is based on implicit classification of items and the recommendation algorithms are memory and item-based and incremental. It was also built and tested a version of the system that considers explicit classification of items by users, with the aim of comparing the performance of both types of classification. It is shown in this dissertation that the multi-agent recommender system AMAAFWA, in online and real-time operation, has a higher performance when compared to individual algorithms, being able to improve user satisfaction and contribute to the increasing success of the website. Concerning the comparison between implicit and explicit classification, the results show a similar performance for both.
Description
DissertaĆ§Ć£o de Mestrado em Tecnologias e Sistemas InformĆ”ticos Web apresentada Ć Universidade Aberta
Keywords
InformĆ”tica PĆ”ginas Web Sistemas de recomendaĆ§Ć£o Internet Web recommender systems Multi-agent systems Association rules Collaborative filtering JADE
Citation
Neto, Joaquim - Abordagem multiagente em sistemas de recomendaĆ§Ć£o Web [Em linha]. [S.l.] : [s.n.], 2015. 101 p.