Sentiment Analysis on Product Reviews: Amazon

Project Title

Sentiment Analysis on Product Reviews: Amazon

Project Area of Specialization Artificial IntelligenceProject Summary

Our project is based on machine learning. We will analyze the sentiments that user gave in comments of any amazon product reviews. This is basically natural language processing. We will get labelled data which has two features. First one is review comment the next is the label (positive or negative). We will preprocess data and apply feature extraction and convert words in vector representation. Then we will train our model and evaluate it and select the best parameters for it. Then we will integrate this model in website platform in which user enter the URL of any product. Our web scrapping model will fetch the reviews of that product and then predict the sentiments of users by applying our trained model and then our platform will show the result to user. In this Project, we will make sense of the sentiment expressed in product review section of Amazon. We will help the seller through our model that how customers are feeling about their product, in this way seller can improve the area of product which needs to be improvised or think of ways to make it better for customer use. On customer point of view, Customer will know what are the general sentiments of the product are.

Project Objectives

• The objective of the project is to create a model that trains on previous reviews of the product as data from specific seller and implements these values in the set of information exploration rules to decide how the product is doing in customer’s hand and what are the areas that can be improved using the reviews by customer. From customer point of view it will simply give him/her the idea that what are the general remark/sentiment of the users of the product
• Also objective is to make the seller investment based on this model for example the product areas that are good and is appreciated by the customer can be kept and the negative reviews or the things people don’t like about seller’s product can be improved/changed/removed/modified or any other improvement method can be used.
• Given an opinion document (with amazon product reviews) d, discover all opinion quintuples (e, a, s, h, t) in d. where e is the target entity, a is the target aspect of entity e on which the opinion has been given, s is the sentiment of the opinion on aspect a of entity e, h is the opinion holder, and t is the opinion posting time; s can be positive, negative, or neutral, or a rating (e.g., 1–5 stars).
• The challenge will be more useful for amazon seller as they will know what are the short-comings of the product are and another very prime objective lies within sentiment analysis of amazon product is that for new sellers they can surf to the best seller reviews can sell what people are most liking in their product and what are the things they are disliking or the thing they want to improve so they do not have to go through the survey process either online or on ground they simple have to run our model on the product review and later on can introduce their very new product in the market with ease. This anticipated and analyzed records can be used to recognize the general repute of corporations that how they are caring for their customers. We can certainly determine the styles of the usage of information mining algorithms.

Project Implementation Method

Some key deliverables before Phase – I are:

• Project proposal
• Detailed proposal
• Preliminary Project Studies Report

Phase I (Machine Learning Model)

1. 1. Download Data set Labelled

First we have to download a supervised data available on the internet. So that our learning be supervised. There are numerous sources from where data can be gathered but we came across a dataset of 1.6 Gb that was not specifically directed towards one category of the products rather it was kind of a universal type of data with entries in train data more than 10 lac and in the test data it was around or maybe near 05 lac. The link to the data is given as https://figshare.com/articles/dataset/Amazon_Review_Polarity/13232501/1

1. 1. Cleaning and feature extraction of the dataset

Cleaning the data is not a simple step it requires time, effort and energy. Also some of the feature extraction techniques is also in this step in our case is the vocabulary of words. Now that we have downloaded the data now our next step would be to clean the data in the most effective way.

1. 1. Transfer Learning For (Word Embedding)

We can use any pre-trained models but we’ll use Google’s word2vec model because as per internet research we came to know that the google model is very well trained than any other model also it can be one of the best source of transfer learning this model is also robust and perform wells on any of the natural language processing dataset. This will require high computation and processing power to load this much data into the model.

1. 1. Model Selection RNN:

The next step is the model selection. So we have decided to use the “Recurrent Neural Network”.

1. 1. Train/Test split

Then the model will be divided into two part train and test 80% will be used in training and 20% be used in testing.

1. 1. Save model

Model will be saved on the local machine.

Phase II (Web Scrapping Model)

2.1 Web Scrapping

We have scrape the reviews data from Amazon.com. One will write a program that queries internet servers, requests and retrieves information, parses it to extract info and stores it in a csv file.  

• Request 
• Beautiful soap 
• Selenium

2.2 Preprocessing

After the data is being scraped the next is step is to clean the data. If any missing values etc are present then remove it. As Scrapped data may be unstructured or not good for the model so clean and check the data before feeding into the model. Save the data in csv.

Phase III (Website building and ML - Model Integration)

3.1 Let the input of saved model be the scrapped csv file

3.2 Dashboard creation using flask or Django and website building

Tools and techniques

Python, Numpy, Pandas, Jupyter lap, Tensorflow, beautiful soup, flask, web hosting , a workstation with high computation and processing power and enough space in memory to save and run the model

Benefits of the Project

• The findings of these study help sellers on Amazon. By examining what are the common topics that people write good review or bad review, sellers can make an information driven investment on product and service improvements.
• The challenge will be more useful for amazon seller as they will know what are the short-comings of the product are and another very prime objective lies within sentiment analysis of amazon product is that for new sellers they can surf to the best seller reviews can sell what people are most liking in their product and what are the things they are disliking or the thing they want to improve so they do not have to go through the survey process either online or on ground they simple have to run our model on the product review and later on can introduce their very new product in the market with ease. This anticipated and analyzed records can be used to recognize the general repute of corporations that how they are caring for their customers. We can certainly determine the styles of the usage of information mining algorithms.
• Amazon itself can use the findings to identify anomalies in reviews. This study itself cannot be used for detecting anomalies but it can provide supplemental information to study that try to find anomalies in review.
• Live insights with sentiment analysis, one can see the mood of each customer in a real-time.
• Tracking overall customer satisfaction that is it tells you how effective your service is.
• Upselling opportunities one can find the happy customers and happy customers will tend to recommend the product to others so it will boost your sale as well.

Technical Details of Final Deliverable

• By using real-time scrapped reviews data from Amazon product review section, we analyse sentiments of reviews to decide the outcome in the form of beautifully designed dashboard showing percentage of positive, negative, or neutral reviews. We also observed the potential impact of reviews on selling trends. The RNN model is methodical in making prediction on the time-series data. Also, we will help the seller to improve the areas of the product which are not liked by the general audience that is buyer in our case.  

• Our model will advise both the buyer and the seller of the product. For instance, if a product has many positive reviews then our machine learning model will advise the buyer to buy this product and will not recommend to buy the product if product has many negative reviews.  

• While, on the other hand from seller point of view it will advise him to upgrade certain area of his service or product to be improved so that he/she can improve his/her sale and meet customer expectations.   

• Our Web app will use the “Recurrent Neural Network” at backend as a machine learning model. Out of many models, RNN are supposed to be the one that can be used in providing us better prediction results. The unique aspect of NLP data is that there is a temporal aspect to it. Each word in a sentence depends greatly on its context. In order to account for this dependency, we use a recurrent neural network.  

• Purpose of the project is to optimize performance of product in the global market and driving traffic to ones’ product in Amazon Listing on online portal, by introducing a state-of-the-art sentiment analysis model on product reviews section.  

• Main driving force behind our end software product will be the Web Scraping Model which will scrape and extract the reviews of our client’s product dynamically whenever a new review is added to the product reviews section. We will use the most powerful web scraping libraries of python like selenium, beautifulSoup4 and Scrapy. By scraping the product reviews from amazon and labelling the reviews as (positive, negative or neutral) we will be all set to train our main machine learning Sentiment Analysis Model on the basis of the data gathered in the scraping step. On the basis of the training data given to the model, our model will predict the overall product’s reviews sentiments. The full and final deliverable of our product will be the full fledge graphical user interface designed and integrated with our trained machine learning model which will be available for use by the professional Amazon product sellers who will insert the URL of their Amazon product and our model will scrape the reviews and then the model will predict about the product sentiment and deliver a detailed report to the seller to assist them to sell their product better on Amazon. This is the ready to use system and the most vital deliverable in this project.  

Final Deliverable of the Project Software SystemCore Industry ITOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development GoalsRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
512 GB SSD	Equipment	1	15000	15000
intel Core i-7 8th gen Processor	Equipment	1	45000	45000
16 GB Ram DDR4 3200 MHz	Equipment	1	10000	10000
Web Hosting	Miscellaneous	1	5000	5000
Co Lab paid cloud functions	Miscellaneous	1	5000	5000