Effective Big Data Retrieval Using Deep Learning Modified Neural Networks

Abstract

In big data, the information retrieval (IR) embraces the discovery of documents from a compilation of dataset which are related to the user query. Usually, the data retrieval systems are used to scan the data. The existent approaches that employ customary IR are wearisome for big document collections. Recently, IR approaches were developed, albeit these are faster comparing to the traditional method but the quality of the document retrieved is less. To overcome such difficulties, here, effectual big data retrieval utilizing Deep Learning Modified Neural Networks (DLMNN) is proposed. Initially, the general pre-processing along with feature extraction steps are taken place. In preprocessing stage, unwanted data are removed and also used for changing the unstructured data in to structured data then in FE is done using frequency and entropy calculation of the given input data. Secondly, find the closed recurrent item dataset, after that find the weight of provided data using entropy measure and frequent item measure. In the 3rd step, the documents are clustered utilizing the k-means algorithm and then classified using DLMNN. The K-Means algorithm is utilized to partition the collection of documents into several clusters then the DLMNN is used for classifying the documents into positive and negative classes. The proposed DLNN weight parameters are optimized utilizing the cuckoo search (CS) optimization algorithm. The last stage on the training process section is generating a training data-base. In the other part, the retrieval process is performed, in this section it pre-processes the user query and discovers the frequency item set then gets retrieval data. Finally, check the similarity assessment, if the information is found then it is visualized, otherwise the document is returned to the initial position. Experimental results contrasted with the previous MRT and IRI-RAS techniques concerning precision, recall, F-measure along with computation time. The proposed document IR is better when comparing with existent methods.