Skip to content

Latest commit

 

History

History
86 lines (57 loc) · 3.17 KB

File metadata and controls

86 lines (57 loc) · 3.17 KB

Opinion Mining based Fake Review Detection using Deep Learning Technique

Table of Contents

Overview

About

Recently, the field of text mining gained more attention due to its enormous opportunities and challenging problems in the exponential growth of unstructured textual data. People frequently express their thoughts in complicated ways, making automated labeling of textual data challenging. The usage of mislabeled datasets reduced the efficiency of automated labeling tasks. In this paper, we proposed an Opinion Mining-based Fake Review Detection using Deep Learning technique to express word semantic sentiment. SentiWordNet and WordNet lemmas are used to retrieve word synsets and sentiment scores. The Amazon Shoes Review Dataset is used for implementation. The proposed model achieved 92 % accuracy for the Amazon Shoes Review Dataset.

Screenshot

My Process

Algorithms Used

The Model is built using a deep learning technique known as Attention mechanism which will give importance to certain words which is important for decision making process. Using this technique we can classify a review whether it is positive or negative. Further we can analyze whether a negative review is a fake review or not.

Built with

Libraries used to built the model:

  1. pandas.
  2. numpy.
  3. tensorflow.
  4. seaborn.
  5. sklearn.

What I learned

Through this project I got a better understanding how deploy a deep learning model on amazon review data set. How to do sentiment analysis and also got familiar with different libraries such as seaborn, sklearn, numpy etc.

Continued development

The main motto of this project is to help a ecommerce site to understand which project is getting positive reviews and will bring more customers. In this paper, we proposed an opinion mining based fake review detection using deep learning technique approach for detecting the fake customer reviews in an online website. The implementation result shows that, the proposed model achieved 92 % accuracy. In future, the presence of anomalies in the validation dataset can be further refined to improve the accuracy

Useful resources

Authors

Sayan Poddar

Koustav Pal

Madhumita Choudhury

SK Saif Ahmed

Soumyajit Halder