What question did this study set out to answer?

This research aims to develop an effective information retrieval system that integrates text and image retrieval using deep learning techniques.

December 3, 2020

Collaborative Text and Image based Information Retrieval Model using BiLSTM and Residual Networks

Key Points

This research aims to develop an effective information retrieval system that integrates text and image retrieval using deep learning techniques.
Developed a collaborative model using bi-directional long short term memory for text retrieval and CNN-ResNet50 for image retrieval.
Employed differential evolution for optimizing the Bi-LSTM technique.
Analytical assessment of precision, recall, and F-score was conducted.
DE-BiLSTM achieved a precision of 99.08%, recall of 99.24%, and F-score of 99.21%.
CNN-ResNet50 model yielded an average precision of 0.96 and average recall of 0.91.

Abstract

Information retrieval (IR) becomes a crucial process in different applications like digital libraries, medicine, entertainment, and so on. Presently, the deep learning (DL) techniques are found to be applicable in several computer vision areas like object tracking, image classification, and natural language processing. So, the DL models are mainly applied in this work for the retrieval of text and images effectively. This paper develops a collaborative text and image based information retrieval system using DL models. The presented model involves two distinct processes namely text retrieval and image retrieval. Primarily, the convolutional neural network (CNN) based residual network 50 (ResNet) model called CNN-ResNet50 with Manhattan distance based similarity measurement technique is applied for the effectual retrieval of images. On the other side, bi-directional long short term memory (Bi-LSTM) technique optimized by differential evolution (DE) is applied for text retrieval. The proposed system highlights that the accuracy can be gained by increasing the depth( intermediate layers ) in DL model and providing an optimal result when using differential evolution algorithm with Bi-LSTM. A detailed implementation analysis was carried out to highlight the proficient results of the presented model. The outcomes indicated that the DE-BiLSTM technique has resulted in effectual outcome with the precision of 99.08%, recall of 99.24%, and F-score of 99.21%. Similarly, the CNN-ResNet50 model has also proficient retrieved the images with the average precision of 0.96 and an average recall of 0.91.

Demander à l'IA

Bookmark