An intelligent personalized web user information retrieval using partial least squares and artificial neural networks

Abstract

The Internet has more than Bronto bytes of available information. However, not all the information matches the appropriate meaning of the search. In information retrieval (IR), the documents are organized based on the terms and inverse document frequencies. The major issues in the information retrieval systems are vocabulary mismatch, lexical similarity, and performance threshold. This paper analyses the query optimization problem for personalized web searching and it proposes an intelligent method for developing an intellectual, personalized searching scheme by machine-learning algorithms. Here, the optimized features are selected by Deep Belief Network (DBN). For accelerating the search process, inverse filtering (IF) is used because text matching is time-consuming. The similarity between the query and the document is estimated using a Genetic Algorithm (GA)-cosine similarity. Furthermore, a user interest prediction scheme by PLS-ANN (partial least squares-artificial neural network) hybrid model is executed and interconnected with a web personalized search engine. To enhance the personalized search query, the feedback module is significant in this search engine. The PLS-ANN is first time developed for web user IR, which has been not yet to be implemented for the feedback module. The proposed approach is implemented in the Python platform and the performances are evaluated using precision, recall, f-measure, and accuracy. Then, the proposed scheme performance is compared with previous machine learning methods and several classical methods. The implementation results proved that the PLS-ANN achieved better performance than the existing algorithms.