Artificial Intelligence and Data Science in Recommendation System: Current Trends, Technologies and Applications

This study provides an overview of recommendation systems and machine learning and their types. It briefly outlines the types of machine learning, such as supervised, unsupervised, semi-supervised learning and reinforcement. It explores how to implement recommendation systems using three types of filtering techniques: collaborative filtering, content-based filtering, and hybrid filtering. The machine learning techniques explained are clustering, co-clustering, and matrix factorization methods, such as Single value decomposition (SVD) and Non-negative matrix factorization (NMF). It also discusses K-nearest neighbors (KNN), K-means clustering, Naive Bayes and Random Forest algorithms. The evaluation of these algorithms is performed on the basis of three metric parameters: F1 measurement, Root mean squared error (RMSE) and Mean absolute error (MAE). For the experimentation, this study uses the BookCrossing dataset and compares analysis based on metric parameters. Finally, it also graphically depicts the metric parameters and shows the best and the worst techniques to incorporate into the recommendation system. This study will assist researchers in understanding the summary of machine learning in recommendation systems. 

 The revolution of Industry 4.0 will leave an impact on the domain of everyone's lives directly or indirectly. Several new complex applications will be developed in the days to come that are complicated to predict in the current scenario. With the help of machine learning approaches and intelligent IoT devices, people will be relieved from extra overheads of redundant work currently being performed. Industry 4.0 has become a significant catalyst for innovation and development in various industrial sectors like production processes and quality improvement with greater flexibility. This chapter applied different machine learning algorithms for spam detection and classifying emails into legitimate and spam. Seven classification models: Decision Trees, Random Forest, Artificial Neural Network, Gradient Boosting Machines, AdaBoost, Naive Bayes, and Support Vector Machines are applied. Three benchmark spam datasets are extracted from standard repositories to conduct the experiments. The chapter also presents a quantitative performance analysis. The results from rigorous experiments reveal that ensemble methods, Gradient Boosting and AdaBoost, outperformed other methods with an overall accuracy of 98.70% and 98.18%, respectively. The ensembled models are effective on a large-sized dataset embedded with more extensive features. The performance of non-ensemble methods, ANN and Naïve Bayes, was instrumental on large datasets as a viable alternative, with an overall accuracy of 98.38% and 97.63% on test data.

The ever-increasing information on the internet and the rapid development of online movies, songs, and stores have enhanced the demands of customers to obtain the required information within the least time. A Recommendation System (RS) is designed to help customers with personalized information and interests to avoid overloading issues in entertainment and social media. Though traditional methods have made noteworthy developments, RS encounters challenges such as data limitations and cold starts. The present study aims to review the developments in the field of deep learning-based RS, thereby providing the required information for researchers. In addition, several applicable domains of employing deep learning-based RS have been analyzed. The review has been organized into RS type, deep learning approaches, deep learning-based recommendation systems in various applications, and evaluation metrics.

Traditionally, single-domain recommender systems (SDRS) can suggest suitable products for users to alleviate information overload. Nonetheless, cross-domain recommender systems (CDRS) have enhanced SDRS by accomplishing specific objectives, such as improving precision and diversity and solving cold-start and sparsity issues. Rather than considering each domain separately, CDRS uses information gathered from a particular domain (e.g., music) to enhance recommendations for another domain (e.g., films). Context-aware Recommender System (CARS) focuses on optimizing the quality of suggestions, which are more appropriate for users depending on their context. Integrating these techniques is helpful for many cases where knowledge from several sources can be used to enhance recommendations and where relevant contextual information is considered. This work describes the main challenges and solutions of the state-of-the-art in Cross-Domain Context-Aware Recommender Systems (CD-CARS), taking into account the abundance of data on different domains and the systematic adoption of contextual data. CD-CARS have shown efficient methods to tackle realistic recommendation scenarios, preserving the benefits of CDRS (regarding cold-start and sparsity issues) and CARS (assuming accuracy). Therefore, CD-CARS may direct future research to recommender systems that use contextual information from multiple domains in a systematic way. 

In the era of information overloading, enormous scholarly data poses a challenge in identifying potential authors for productive outcomes. Researchers collaborate with fellow profiles to improve the eminence of Research and their academic profiles. This chapter proposes a content-based recommender system to generate author recommendations for collaborations that extracts the relevant keywords from the titles of research papers using MapReduce. To specify author specialization, the proposed technique comprises the feature extraction from the entire document using Latent Dirichlet Allocation (LDA), followed by an influence model which generates recommendations for the target authors. A ranking algorithm, such as TOPSIS is implemented to get Top-N recommendations based on multiple criteria. In this chapter, we investigated how the MapReduce framework is helpful in obtaining improved computational time for large-scale scholarly data and scalability. Experimental results on DBLP articles prove the relevance of ranking methods as an efficient and scalable platform for computing content-based recommendations

Recently, most retail-based and e-commerce companies have been using recommender systems aggressively. It retains a customer's interest by giving exclusive offers on personalized preferences. The primary purpose of a recommender system is to get at an increase in sales by providing an enriched experience to the customer. With the emergence of many video streaming services like Netflix, Hotstar, and amazon prime video, the dependency on movie recommendation systems has increased. It facilitates the users in faster search and easier access for shows matching their tastes and helps them choose what they are looking for without getting lost in the flood of available options. The user most often gets surprised by seeing an offer that they possibly would never have searched. The system is based on information retrieved and processed user preferences, ratings, likings, disliking, etc., to use this understanding to recommend the products. In this chapter, we have discussed the various popular algorithm used for the movie recommendation, along with an insight into the extensive use of models based on machine learning especially deep learning. The performance of different movie recommendation systems with a comparative analysis is also given to encourage further research in this area. 

The viewpoint, together with a feeling examination, helps in distinguishing the perspective and characterizing the survey as good, nonpartisan, or negative. The angle-based conclusion investigation incorporates pre-processing of surveys, extremity figuring, and distinguishing the perspective and execution assessment. Primary assets of constant assessment, Twitter, messages, and other interpersonal organizations have captivated the significant interests of the exploration business and network. Wistful analysis (supposition mining) of brief, casual writings via web-based media sums up feelings as good, unbiased, or negative expressions of the sentiment holder. We will utilize a viewpoint-based estimation examination by using profound learning calculations. This methodology receives language preparation procedures, rules, and dictionaries to address a few opinion examination difficulties and produce summed results. By utilizing deep learning, we can yield more exactness than Artificial Intelligence (AI) calculations.

 In the Internet age, we perceive the use of recommender systems all around us. The exponential growth of information from intelligent devices on the internet creates confusion for customers to pick a preferred product. Suggestions are a noble way to guide shoppers to discover fascinating products to impress customers. These recommender systems influence our browsing or watching or listening, searching patterns, and guess what customers might like in the future based on our patterns. For instance, a customer searching for baby products recommend diapers. Two significant categories of recommender systems exist, which are either collaborative or content filtering. The core of the recommender system resides in filtering similar users (or products). We address the introduction, existing works focusing on collaborative and content recommender filters, and their merits and demerits. Later, we classify types therein and thoroughly discuss similarity metrics used to filter neighborhood and evaluation measures used in the recommender system. 

Point-of-Interest (POI) recommendation helps to find new places for users to visit, along with the popularity of locations. Recommendation of POI is the most important in location-based social networks (LBSNs). This paper discusses different parameters that significantly impact the POI recommendation process and make the prediction much more accurate. A comprehensive review of a few research works and the methodologies employed for POI recommendation have been presented. POI recommendation techniques have been classified based on many, such as the interest of the tourist in particular POI, popularity of the POI, weather conditions, etc. A summary of related research work is presented for each category, along with their respective drawbacks. Finally, the possible directions toward future work in this area are included, along with the conclusion.

Mobile Tourism Recommendation System recommends to a tourist the best attractions in a particular place according to his preferences, profile and interest. First, a Recommender system offers a list of the city places likely to interest the user. This list estimates the user demographic classification, likes in former trips, and preferences for the current visit. Second, a planning module schedules the list of recommended places according to their characteristics and user limitations. The planning system decides how and when to perform the recommended activities. For implementing these recommender methods, we have applied different machine learning algorithms, which are the K-nearest neighbors (K-NN) for both Clean Boot (CB) and Consolidation Function (CF) and the decision tree for all Data Framing (DF). Thus, executing a recommendation system for tourists helps them with user-friendly planning. Blind people can also use this. This application provides complete voice assistance for easy navigation via a simple button click. Vibratory and voice feedback is provided for accurate crash alerts for visually challenged people. The application extracts its smartness by incorporating Android and Internet of Things (IoT) support. Since blindsupported applications and devices are more expensive and many blinds can not afford them, we aim to put forth a novel, low cost and reliable approach to help the blind explore the possibilities and power of smartphone technology in navigation. We additionally expect to find the static variables that should be tended to, food, tidiness, and opening times, and valuable to suggest a tourist place depending on the travel history of the client. In this investigation, we propose a cross-planning table methodology depending on the area’s prevalence, appraisals, idle points, and conclusion. A targeted work for proposal streamlining is defined as dependent on these mappings. Our outcomes show that the consolidated highlights of Latent Dirichlet Allocation (LDA), Support vector machines (SVM), appraisals, and cross mappings are helpful for upgraded execution. The fundamental motivation of this study was to help businesses related to tourism. 

In the recent era, recommendation systems have marked their footsteps and have changed the way of the travel industry. The recommendation system deals with massive amounts of data to identify users’ interests, making the location search easier. Many methods have been used so far for making predictions much more desirable regarding users’ interests by collecting Information from a large set of other users. The main objective of this paper is to show various methods and techniques used for generating recommendations. These recommendation processes are classified into different forms, such as traditional methods and tensor-based methods. A brief review of these methods was described with the help of some challenges faced by the recommendation system. Apart from that, the advantages and disadvantages are discussed, along with the highlights of future directions.

In this article, the insights of a garment retail store have been studied with respect to the attributes of the dresses and sales information. Mention that each dress in fashion retail has several attributes or features. These features play a critical role in the selection of consumers or customers. This study tries to establish the relationship among these features by which the importance of the attributes is evaluated concerning sales. Furthermore, this paper tries to automate the process of the recommendation of the dresses by using these attributes. It is merely a binary classification but useful for retail sales. Moreover, the demand for sales is estimated over a period. All these objectives are achieved through using one or more data science techniques. The case study shows that the algorithms of data science are helpful in the decision-making of fashion retail.

Human resource data analytics are more important now than ever before. An increasing number of businesses are delving ever deeper into the data they collect about their employees, their success, and their well-being. Recruitment analytics can assist in making smarter, data-driven selection, recruiting, and sourcing decisions. This technology will scan hundreds of resumes at once to provide the best possible fit for a particular job opening. An organisation can submit automated emails with an interview appointment that automatically sinks with the work calendar using modern recruiting software. The organisation may use automated disqualification of unqualified applicants to automatically screen application forms and exclude candidates who aren't qualified. Effective recruitment is a mix of science and art. It necessitates the implementation of repeatable processes that produce consistent results.

Early and accurate crop yield estimates at a local and national level are essential to oversee industry and trade planning and to mitigate the price hypotheses. The major challenge for farmers in the agricultural field is selecting an appropriate crop for planting. Crop selection is dependent on several factors like climate, soil nature, market, etc. Majorly, crop yield production depends on weather conditions and soil types. Yield anticipation is essential for farmers nowadays, which significantly adds to the appropriate yield selection for sowing. There needs to be a framework to recommend what type of crops to produce for farmers. It is essential and challenging to make the right farming decisions at a future steady cost and yield balance. This article proposes an Artificial Neural Network (ANN) model for rice crop yield prediction by utilizing weather parameters like rainfall, temperature, sunshine hours, and evapotranspiration. Generally, Default-ANN has only one hidden layer. But in this work, a Personalized Artificial Neural Network (PANN) approach has been designed by varying the number of hidden layers, the number of neurons, and the learning rate. P-ANN model accuracy is computed using R-Square (R2) and Percentage Forecast Error (PFE). Outcomes demonstrate that the P-ANN model performs precisely with a greater R2 and smaller PFE values than existing methods. For this research, the seasonal (Kharif & Rabi) weather dataset and rice yield data of Guntur district, Andhra Pradesh, India, from 1997-2014 have been used. Better paddy yield was forecasted by utilizing the P-ANN approach.