Dynamic QuillQuest using ML

Abstract

In a constantly evolving digital landscape, recommender systems have become crucial for filtering online content and meeting individual user preferences. These systems have challenges with sparse data and scalability despite their accuracy. This paper performs a comprehensive analysis of the latest developments in recommender systems for a range of applications. It examines datasets and performance indicators, analyzes algorithms, and establishes a taxonomy for system development. In order to facilitate future advancements in effective recommender system development, the paper highlights gaps and limitations in the field\'s current state of research.

Introduction

I. INTRODUCTION

 With today's sophisticated technology and widely available internet services, it is now easier to obtain large volumes of information rapidly. But the proliferation of computing has resulted in an overabundance of data, making it more difficult to find pertinent internet material due to this overload issue. In response to this difficulty, recommender systems have gained so much of attention. By providing individualized content, these systems function as information filters and seek to streamline and accelerate users' search operations. Recommender systems are being applied in large scale of industries, including e-learning, travel, movies, the web, and literature. In spite of their progress, recommender systems still require work to become more effective and useful in a wider variety of scenarios. In order to shrink the distance, this paper compares and evaluates previous research on recommender systems, examining their datasets, features, difficulties, applications, simulation platforms, algorithms, classification, and system performance. The objective is to steer future exploration in the scope of recommender systems across different kind of applications and offer insightful information for practitioners and academics.

The major types of recommender systems are: hybrid systems, collaborative systems, and content-based systems. These groups cover several methods via which consumers are recommended to. In Figure 1, which depicts the block diagram of recommender systems.

II. RECOMMENDER SYSTEM BASED ON CONTENT

Authors, publishers, directors, actors, and other fundamental characteristics of books and movies are used by content-based systems to classify them into unique profiles. When An object or an application that has been used by the customer and given highest rating possible, the system gathers other products with similar characteristics and creates a customized customer profile. The contents in this customer profile are then later used to generate recommendations, as seen in Figure 2.

Still there exist many other factors to keep in mind when using this strategy. For reliable recommendations, a full grasp of the item's features is needed, which is not necessary that it will always be available for every item. Additionally, the method can't go beyond a user's current preferences or areas of interest. Notwithstanding these difficulties, content-based systems provide a number of benefits. They show a rapid ability to adjust to shifting consumer preferences over time. The confidentiality and authentication of customer data are guaranteed by this algorithm, which works independently of other users' profiles because each customer profile is unique to that particular person.

Its capacity to pacify the "cold-start" issue—recommending an item even if we are not provided with the ratings given by the customer—is one of its standout features, especially for new items with adequate description content. Web page recommendation systems, tailored news, and publications all make extensive use of content-based filtering techniques. Although they are excellent in some areas, the approach's efficacy may differ on the basis of its accessibility and depth of item descriptions, also all possible difficulties in varying recommendations to accommodate changing user preferences

A. Collaborative Filtering-Based Recommender Systems

Recommender systems based on collaborative filtering make advantage of user similarity metrics to provide tailored recommendations. This entails locating a user neighbourhood where a target user's tastes coincide, and then suggesting products that the maximum customers in this group Favor. Collaborative filtering tackles issues like the cold-start problem even though it doesn't require knowledge of item features and handles privacy concerns.

The two categories of these techniques are memory-based and model-based. Memory-based systems employ the utility matrix to directly recommend things om the basis user's preferences and rating for their neighbourhood. A new item's rating is determined via collaborative filtering based on customers using the ratings of nearby customers those who have provided the same or same product or item a highest rating. By building an item neighbourhood with comparable user-rated items, item-based collaborative filtering, on the other side, it predicts the ratings given to a novel product by averaging the weighted ratings of the items inside this neighbourhood.

 Systems based on models utilize machine learning and mining data processes / techniques to build a model that forecasts user ratings for the products which have not been rated yet. In contrast to memory-based methods, they extract features to achieve the aim to construct a model rather of relying on the entire dataset. Model-based approaches are effective for group suggestions since they don't need to add a new user's profile to the utility matrix beforehand. Using methods for model learning and dimensionality reduction, they tackle problems like sparsity and scalability. The efficacy of the underlying learning method used to build the model determines the accuracy of model-based systems (Figures 3 and 4 show the item-based and user-based collaborative filtering techniques, respectively)

III. CHALLENGES FACED BY AN RECOMMENDER SYSTEM

This section briefly explains the different problems that exist in the recommender systems that are currently in use and provides several ways to solve these problems.

1. Cold start issue: This is an issue that arises when the system lacks sufficient data to make wise recommendations., especially for newly added goods or new users. This occurs during the initial setup of the system or whenever new entries are made to the database. The cold start hybrid approaches that combine cooperative and content-based screening techniques.
2. Latency issue: This issue is unique to collaborative filtering techniques and arises when new entries are consistently added to the database with new items. Because these reviewing new products is necessary. before being included, this problem makes it more difficult for the algorithm to propose them quickly. Content-based filtering has the potential to reduce latency, but it can also increase computation time and lead to overspecialization. Using clustering-based approaches and doing calculations in an offline environment are necessary to overcome this difficulty.
3. Scalability issue: Particularly, recommender systems that employ collaborative filtering approaches suffer greatly from scalability problems. As additional goods and people join the system, scalability problems are made worse by the growing volume of training data. There are two widely utilized techniques. to address scalability issues: dimensionality reduction and the application of clustering-based algorithms. By identifying people in smaller clusters rather than analyzing the entire database, these solutions seek to improve scalability through simplifying the system's processing in the big data era.

Conclusion

In summary, this paper provides a comprehensive and insightful survey of recommender system research conducted between 2011 and 2021. The data shows that recommender systems research has grown significantly, which emphasizes the necessity for additional study in this field. Interestingly, the large body of research on movie suggestions contrasts with the paucity of studies on recommender systems linked to health, tourism, and education, highlighting the need for broad datasets to support advances across many domains. The lack of a standard measurement to assess recommender system success is a crucial finding. The prerequisite for a single framework for performance evaluation is highlighted by the several different techniques that have been used in the articles that have been identified, with differing degrees of acceptance. The development of systems requires the use of programming languages., and the fact that Python and other languages account for 27% of contributions imply that because these languages have robust libraries, they are used extensively, that make development easier. Furthermore, the introduction of optimization and hybrid approaches shows a viable route to enhance recommender system performance. Additionally, the study demonstrates the growing popularity of deep learning and neural network-based methods., as they can be utilized to build very accurate recommender systems. The study\'s findings point to a dynamic area of research on recommender systems that can be improved, standardized, and investigated further by utilizing new technologies.

References

[1] Online Book Recommendation System using Custom Recommender Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2022 https://doi.org/10.1109/ICOEI53556.2022.9777131 [2] Improving recentness of the ICT book recommendation using an adaptive rules-based recommender system https://doi.org/10.1504/IJCAT.2022.130880 [3] Design and Implementation of the Web-Based Book Recommendation System https://doi.org/10.4028/www.scientific.net/AMM.427-429.2143 [4] Efficient Book Recommendation System Based on a MapReduce Model https://doi.org/10.4028/www.scientific.net/AMM.284-287.3405 [5] Castellano G, Fanelli AM, Torsello MA. NEWER: A system for neuro-fuzzy web recommendation. Appl Soft Comput. 2011;11:793–806. 2. [6] Crespo RG, Martínez OS, Lovelle JMC, García-Bustelo BCP, Gayo JEL, Pablos PO. Recommendation system based on user interaction data applied to intelligent electronic books. Computers Hum Behavior. 2011;27:1445–9. [7] Lin FC, Yu HW, Hsu CH, Weng TC. Recommendation system for localized products in vending machines. Expert Syst Appl. 2011;38:9129–38. [8] Wang SL, Wu CY. Application of context-aware and personalized recommendation to implement an adaptive ubiquitous learning system. Expert Syst Appl. 2011;38:10831

Copyright

Copyright © 2024 Ankit Sharma, Prabhakar Kumar Singh, Bahushruth M Pujar, Prof Manjula BS. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.