The popularity of mobile devices among the users has increased the dependency on the mobile servers. People get lots of information including business information, product information, and recommendation information from the mobile devices. One of the important applications of mobile servers is movie recommendation. A movie recommendation system has been an effective tool in recommending movies to the users which, in turn, helps the viewers to cope with multiple movie options available and help them in finding the appropriate movies conveniently. However, recommendation is a complicated task as it involves various tastes of users, different genres of movies, etc. Hence, many techniques have been used to enhance the performance of the recommendation system [32].
We have a massive platform that can be used for giving individual thoughts and reviews. As there is so much data flowing over the internet, it is significant to derive new ways to collect and produce the information. Recommender system is an important component of many businesses, especially in the ecommerce domain. It usually exploits the preference history of the users to provide them with the suitable recommendations, whereas a traditional recommender system can provide only one rating value to an item [5, 24–26].
3.2 Work Related Multi-Criteria Recommender System
Multi-Criteria Recommender System (MCRS) is widely used in almost every sector. It has developed with time. Nowadays, we have many advance recommender systems. Recommender system models can be made by various methods like clustering technique, machine learning technique, deep learning techniques, neural networks, and big data sentiment analysis. There are many open source projects that are developing in the domain of MCRS. So, let us take a look on here.
Wasid and Ali came up with a MCRS based on the clustering approach. The primary objective of their method was to enhance recommendation performance by identifying more similar neighbors within the cluster of a specific user. To implement this method, they had done two major things. First, they extracted the users’ preferences for the given items based on multi-criteria ratings. Second, on the basis of the preferences of the user, the cluster centers were defined [2].
Zheng proposed a utility-based multi-criteria recommender system that depends on the utility function. He built the utility function by applying the multiple-criteria ratings to measure the similarity between the vector of user evaluations and the vector of user expectations. To calculate the utility score, they had incorporated three similarity measures. In addition, three optimization learning-to-rank methods were used to learn the user expectations [3].
Tallapally et al. adopted a deep learning–based ANN architecture technique known as stacked autoencoders to ease the recommendations problems. The functionality of the traditional stacked autoencoders was enhanced to include the multiple-criteria ratings by adding an extra layer that acted like an input layer to the autoencoders. The multiple-criteria ratings input were connected to the intermediate layer. This intermediate layer comprised of the items or the criteria. This intermediate layer was further linked to N consecutive encoding layers [4].
Musto, Gemmis, Semeraro, and Lops used MCRS using aspect-based sentiment analysis. They utilized a structure for sentiment analysis and opinion mining. It automatically extracts sentiment scores and relevant aspects from users’ reviews. They estimated the efficiency of the proposed method with other state-of-the-art baselines and compared the result [5].
García-Cumbreras et al. method utilizes the pessimistic and optimistic behaviors among users for recommender systems. The objective was to categorize the clients into distinct classes of two, namely, pessimist class and optimist class based on their cognition or behavior. The classes are defined on the report of the mean polarity of clients’ rating and reviews. Then, the derived client’s class is added as a latest attribute for the collaborative filtering (CF) algorithm [6].
Zhang et al. proposed an algorithm that considers virtual ratings or overall rating from the users’ reviews by analyzing the sentiments of the user’s opinions by using the emoticons that were also included in the reviews to mitigate the sparsity problem which still lies in the recommender systems [7].
Bauman et al. presented a recommendation system that suggested the items that comprised of the most significant aspects to improve the user’s overall experience. These aspects were identified using the Sentiment Utility approach [8].
Akhtar et al. presented a technique for analyzing the hotel reviews and extracted some valuable information or knowledge from them to assist the service providers as well as the customers to help them identify the loopholes and strengths in the service sector to improve their business performance [9].
Yang et al. presented a technique consisting of three main components namely aspect weight, opinion mining, and overall rating inference. The opinion mining component was responsible for extracting only the key aspects and opinions from the users’ reviews based on which it computed a rating for each extracted aspect [10].
Dong et al. presented a method for CF that merges feature similarity and feature sentiments for recommending items, that having higher priority that are similar and better than the items in the users query [11].
Wang et al. proposed an approach on solving a problem when a user is particularly new to an environment. This problem is known as cold start problem. We will discuss about the cold start problem later in this paper. Most recommender systems collect the preferences of the users on some attributes of the items [12].
Musat et al. explained a method called topic profile CF (TPCF) that solved the problems occurring due to the data sparsity problems and non-personalized ranking methods that led to difficulty in finding sufficient reliable data for making recommendations [13].
Jamroonsilp and Prompoon presented an approach for ranking the items based on user’s reviews. They had considered five pre-defined aspects for the software items. The ranking of the software was computed by comparing the sentences analyzing the different clients’ ratings for every software aspect. This was performed in three phases include gathering user reviews, analyzing the gathered reviews and doing the subsequent software ranking [14].
Zhang et al. proposed a method that utilized the aspect-level sentiment of the users’ reviews with the support of helpfulness reviews [15].
Zheng, Shekhar, Jose, and Rai proposed a multi-criteria decision-making approach in the discipline of educational learning. At first, they integrated the context-awareness and the multi-criteria decision-making in the recommender systems considering the educational data. Their experimental results were quite satisfactory, and it was realized that they were able to produce additional correct suggestions based on two different strategies of recommendations [17].
These are some of the works done by various scientists around the globe. There are thousands of projects which has been conducted or ongoing in the field of MCRS to make the system fully efficient. Nowadays, the leading companies are making using of the recommender systems. One of the best examples is the company Amazon that uses the recommender systems to give proper recommendation to their customers. Netflix is another company that uses multi-criteria recommender system to give a list of movies and web series suggestions to the user on the basis of the user’s details and user’s previous choices. So, day by day, new techniques are being applied in recommender systems to improve the accuracy.
3.3 Working Principle
The most basic question comes in mind that what is recommender system. A recommender system is a software or model that analyze a client’s preference, and based on that, it generates a list of items for that client. A multi-criteria recommender system can