Content categorization using enhanced convolutional neural network for library book availability prediction

Abstract

Deep learning (DL) techniques analyze textual data, but existing research has not explored heuristic approaches for extracting crucial text vectors. Essential preprocessing techniques for cleaning global and local text are also required. This study proposes a DL-based framework to predict book availability based on book-title. The proposed methodology has four phases: data preprocessing, text preprocessing, feature selection, and classification. Missing values, duplicates, and feature extraction are handled during data preprocessing to obtain clean data. Then, lemmatization and bidirectional encoder representations from transformers (BERT) vectorization are applied to derive feature vectors from each word in text preprocessing. Four feature selection algorithms-firefly algorithm (FA), artificial immune system (AIS), genetic algorithm (GA) and ant colony optimization (ACO) were used to select the crucial feature vectors. Comparatively, ACO obtained 90.87% accuracy, 90.87% precision, 100% recall, 91.94% F-score, and was selected as the optimal feature selection algorithm. The extracted features from ACO are trained using convolutional neural network (CNN) to predict book availability. Experimental results show that CNN achieved best performance than decision tree (DT), random forest (RF), k-nearest neighbor (KNN), support vector machine (SVM) and other state-of-art methods, i.e., 99.07% accuracy, 99.83% precision, 99.24% recall, and 99.54% F-score for analyzing textual data.