A Review on Consumer Behavior towards Online Shopping using Machine Learning

Study of consumer behavior in online shopping, as a rule, manages identification of consumers and their purchasing behavior. The purpose of such studies is to verify who purchases where, what, when, and how. The analysis of such consumer behavior is useful to get the buyer's prerequisites and requirements for their future aims towards the product. Through this review, E-commerce organizations can follow the utilization and sentiments appended to their items and adopt suitable promoting strategies to give a customized shopping experience to their buyers, consequently expanding their hierarchical benefit. This paper purpose to utilize information-driven promoting models, for example, information perception, natural language processing, and AI models that help in getting the demographics of an association. Additionally, make recommender frameworks through cooperative filtering, sentiment analysis, and neural networks.


Introduction
The use of E-commerce has immeasurably expanded in this modern innovative era. People like to shop online as opposed to shopping in business sectors. In online business, the information created by consumers as reviews and surveys of a specific item can be utilized for authenticity and exposure of the item. Customers settle on whether to purchase a given item by checking out these evaluations and surveys. Such happiness can be positive or negative reviews made by customers/consumers who have recently utilized the item. An exact investigation of this consumer-produced content can be useful to the online business association to acquire bits of knowledge and get their consumer's expectations and necessities ( T.Yoshida,M.Hasegawa,T.Gotoh,H.Iguchi,K.Sugiokaand K. Ikeda,, 2020).
Sentimental analysis is regularly used to assemble a social conduct diagram on human internet-based conduct to observe the connection among exchanging and volume costs of stocks and products.
Sentimental analysis was a help to perform on the information removed from SentiWordNet utilizing a hybrid selection model to show what market patterns mean for item notoriety and rate (Valecha, 2018

Literature Review
The literature review has been committed to showing the exact visual portrayal of buyer conduct to such an extent that it covers every one of the areas of premium expected for an internet business organization to make upgrades in their items and advertising methodologies. Past studies have focused on techniques on how friendly research can be a significant element in deciding the purchasing conduct of any client.  The exploration led for this design was a survey and the outcomes were examined measurably. Trust was asserted as such a human quality that influences their purchasing propensities (Martin, 2020

Datasets
To analyze character characteristics from informal communities, this research utilized the myPersonality dataset as a contextual investigation. This research developed the review with 250 clients and 9917 announcements from the my Personality test. The dataset of Facebook clients was named by the Big 5 model. As indicated by the dissemination of character types in Table,

Outcomes
The dataset comprises Amazon item audits sorted by different classifications like apparel items, home and kitchen items, mobiles, and substantially more. For our review, this research has zeroed in on a particular sort i.e Home and Kitchen items to deal with. The Count Vectorizer is used to tokenize a collection of text documents and build a vocabulary of all the words in that document. This vocabulary helps in encoding new documents. The focus is on the occurrence of words in that document to find out a pattern. This research has used the following classification models for analyzing the sentiment score of each associated product review. In all of these models, 80% of data is used for training and 20% is used for testing purposes. This classification model is used to find K nearest neighbors of a particular data point, according to a certain distance measure, and decide which category it belongs to. Random forests or random decision forests are an ensemble learning method that takes multiple learning algorithms or multiple instances of an algorithm and puts them together to give the best possible result. The SVM classification model is generally used to classify between different categories using some decision boundary. A decision boundary separates the points into two classes.

Conclusion
The goal of this research is to aid e-commerce organizations in analyzing the sale of products and also gain insights into customer intentions while purchasing any particular product. These studies help the organizations in knowing their customers and incorporating targeted marketing techniques to increase their customer base and profits (Martin, Vincent). To understand consumer interactions with e-commerce sites, this research employed various data visualization techniques and did a thorough sentiment analysis on product reviews. Sentiment analysis helped us evaluate consumers' sentiments related to various products which in turn helped us analyze the product's performance in the market (Surendro, 2019). This research used five classification models for sentiment analysis of reviews with the help of natural language processing and obtained the best results in the Naive Bayes classifier. This research also trained an LSTM network which went one step further and found out the amount of positive and negative sentiment in the review, which helps in understanding neutral reviews as well.

Future work
This research employed three recommender systems using nearest neighbors, a priori, and Boltzmann machines. The objective of a recommender system is not to get the best accuracy, but to recommend similar items to a consumer based on their recent purchase activity. This personalized approach to a consumer's needs increases their trustworthiness towards the website and influences them to buy more products. Nearest neighbors helped us filter out similar products and a priori algorithms helped us gather item sets that are frequently bought together. The Boltzmann machine helped in predicting the ratings of all the products that the consumer hasn't bought yet, and gives the most promising result among all.
Using all these results n this research concluded that any customer decision is influenced by multiple aspects when he/she is using any e-commerce platform for shopping. Such factors can be identified by ecommerce organizations and they can take necessary steps to facilitate better service to the customer and add more lifetime value to their business.