Recommendation system for frequent item sets using multi-objective chaotic optimization with convolutional BiLSTM model

Abstract

A recommendation system offers a creative way to handle the limitations of e-commerce services by using item and user details. It is used to ascertain the user’s preferences in order to suggest products they would likely purchase and identify frequently used items from the data. The recommender model is designed with several common collaborative filtering techniques, but it has some complications. To overcome this drawbacks, a novel technique is proposed to find the frequent item in the given dataset. This research paper used two types of data, namely the product image data and user rating matrix data. Initially, image characteristics are retrieved using a residual dense network (RDN) to extract relevant features from the images. Then, the extracted features are fed into Multi-Objective Chaotic Horse Herd Optimization (MO-CHHO) to find common item sets from many items. Here, support, confidence, lift, and conviction are considered multi-objective functions. The text data is classified using Convolutional BiLSTM (CBiL) model based on significant sentiment features like All-caps, hashtags, emoticons, negation, elongated units, bag-of-units, punctuation, and numerical values to identify whether the item is common or not. Finally, the fusion process is performed using correlation to find the final frequent item sets from the image and data sets. The evaluation results show that the proposed method achieved 98% accuracy, precision of 99%, 98.3% of sensitivity, 99.5% of specificity, and 98.7% F-1 score using the amazon product review dataset.