Amazon Product Recs via SVD Algorithm
Amazon Product Recs via SVD Algorithm
Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Amazon Product Recommendation System using
SVD Algorithm
Manne. Lahari1
Maddu. Devi Prasanna2
Eluri. Naveena Kumari3
P.Srinu Vasarao4
Swarnandhra College of Engineering and Technology
Abstract:- Amazon is the world's largest retailer by I. INTRODUCTION
revenue and business. Nearly a third of Amazon's sales
come from referrals, accounting for $470 billion of the Product recommendation is a solution that provides
company's ecommerce revenue in 2021. The program is instant product information to customers. It is a powerful
called recommenders and uses machine learning to select data filter based on algorithms, artificial intelligence,
specific features from larger data sets. When there is a machine learning and other data analysis methods. It
lot of filtering, the suggested ideas are often based on involves collecting, storing, analyzing and filtering customer
what the user has interacted with, purchased, viewed, data to deliver personalized and relevant products to
etc. Matches other similar products. The customers. Appropriate products that meet your customers'
recommendation engine recommends a product based on needs, tastes and preferences. Quality data is needed to show
this understanding of the user. This is how Amazon's detailed individual plans. But most importantly, you need
product recommendation engine works. The the right tools to understand your customer profile and
recommendation engine filters products based on business needs.
product functionality and user characteristics (for
example, find other users who are similar to you and Recommendations are the most popular among
have purchased the product you are looking at or will ecommerce companies like Amazon. These
buy). Amazon's recommendations use different filters to recommendations are valid only if they are based on the
recommend products. Amazon uses various artificial user's behavior and preferences. The collaborative filter
intelligence algorithms to power all aspects of the system is at the heart of Amazon's recommendations and is
platform. To enable smart product search on the used to make recommendations based on user behavior and
Internet, the company also uses a proprietary technology preferences. During this process, Amazon analyzes a lot of
called A9. Amazon recently updated its A9 algorithm, information, such as: A customer's purchasing history,
now called the A10 algorithm. The update changes many viewing habits, and ratings. The goal is to identify users'
aspects of the product's functionality, shifting the focus patterns and trends. The KNN algorithm is a supervised
of the product to the buyer's behaviour. machine learning algorithm used to solve classification and
retrieval problems. The system includes user ID, product
Machine learning algorithms in recommendations ID, rating, etc. using factors obtained an RMSE value of
generally fall into two groups: contextual methods and 0.9941 for this model.
collaborative filtering. Affiliate marketing is the most
common way to make online recommendations. It is In short, recommending products to users based on
"collaborative" because it predicts a customer's their behavior is not an easy task and there is a lot of user
preferences based on other customers. A better way data. This review shows that Amazon's product
would be to recommend the product based on the recommendations are created using a supervised learning
relationship between the products and not on the process. The RMSE value of this system is 0.9941. Finding
customer's consistency. Through user interaction, similar users and their purchasing behavior based on reviews
Amazon.com visitors are matched with other customers will help us find the top k products recommended by a
with similar purchasing history and personalized particular user. This will increase the company's revenue
recommendations are provided. Come and see. There and give satisfaction to the customers as it will recommend
are many ways to create a unified consensus model. popular products based on customer's preferences.
Machine learning algorithms such as SVD and Top-k are
used to find the most popular products. II. LITERATURE SURVEY
Keywords:- Recommendation System, Filtering Techniques, Many studies have been conducted in favor of the
Artificial Intelligent Algorithms, A9 Algorithm, A10 product. Previous works have used collaborative filtering in
Algorithm, Content-Based Filtering, Collaborative various applications based on user-user and project
Filtering, SVD, Top-K. interaction [1]. Recommendations using the collaboration
filter include videos, articles, and product recommendations.
IJISRT24MAR010 www.ijisrt.com 70
Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. METHODOLOGY
In [2] the author talks about a system that will help users
find movies of their choice. The idea behind using This research was designed to build models using
interactive techniques in this research paper is to find videos Google collaboration and Jupyter Notebook. The modeling
that match users' interests among the millions of videos steps are shown below. The first step is to get the dataset
available on online platforms today. from the Kaggle website; Afterwards, the data is processed
first, including data cleaning and quality work; Then, the
In [3] the authors describe their proposed consensus as A data retrieval process and only users who have commented
Priori. It is a universal product convention that encourages 50 or more. Store data is divided into two: training data and
people (especially customers) to actively share product test data. Use SVD to train the model. The table below
knowledge as they use it. shows some information about Amazon seller profiles and
their analytics. Data from Kaggle.com.
Fig 1: Amazon Sales Data
Data can be cleaned, visualized, and reduced in size using machine learning and is best suited for single-value parsing
algorithms. The value of the value parsing algorithm is used to reduce the size of the data. Using SVD, we obtained the RMSE
value of 0.9696. After parameter transformation, the RMSE value was obtained as 0.858.
IJISRT24MAR010 www.ijisrt.com 71
Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Fig 2: Graph Plotted Between Users and Ratings
IV. SVD
In SVD, rating is predicted as-
Fig 3: Model used to estimate the score
If user u is known, then the bias bu and the factors pu are assumed to be zero. The same applies for item i with bi and qi.
To estimate all the unknown, we minimize the following regularized squared error:
Fig 4: Formula used for estimating unknowns
Below we will fix our hyperparameters –
n_epochs : Number of iterations SGD algorithm
lr_all : learning rate of all parameters
reg_all : normalization time of all parameters
IJISRT24MAR010 www.ijisrt.com 72
Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
USING SVD:
Fig 5: Training the data using SVD
Fig 6: Accuracy of the model
After Correcting the Parameters:
Fig 7: Accuracy of the parameters improved model
V. CONCLUSION
The data we use in this analysis is provided by Kaggle
and is based on the world where Amazon actually sells data.
In this analysis, we performed data cleaning, data
visualization and data reduction. We conclude that the
recommendation of Amazon products can provide 0.858
RMSE with filtering and SVD (machine learning algorithm).
REFERENCES
[1]. G. Linden, B. Smith, and J. York, “Amazon.com
recommendations: Item-to-item collaborative
filtering,” IEEE Internet Computer, vol. 7, no. 1, pp.
76–80, 2003, doi: 10.1109/MIC.2003.1167344.
[2]. Furtado, F. and Singh, A. (2020). Movie kev pom zoo
siv tshuab kev kawm. International Journal of
Industrial Engineering Research, 9(1), 84-98.
[3]. Von Reischach, F., Guinard, D., Michahelles, F. and
Fleisch, E. (2009, March). The mobile app
recommends products related to the tagged product.
2009 IEEE International Conference on Ubiquitous
Computing and Communications (pp. 1-6). IEEE no.
IJISRT24MAR010 www.ijisrt.com 73
