Master of Science (MS)
convolutional neural networks, deep learning, dropout, e-commerce product categorization, keras, transfer learning
The global e-commerce market is snowballing at a rate of 23% per year. In 2017, retail e-commerce users were 1.66 billion and sales worldwide amounted to 2.3 trillion US dollars, and e-retail revenues are projected to grow to 4.88 trillion USD in 2021. With the immense popularity that e-commerce has gained over past few years comes the responsibility to deliver relevant results to provide rich user experience. In order to do this, it is essential that the products on the ecommerce website be organized correctly into their respective categories. Misclassification of products leads to irrelevant results for users which not just reflects badly on the website, it could also lead to lost customers. With ecommerce sites nowadays providing their portal as a platform for third party merchants to sell their products as well, maintaining a consistency in product categorization becomes difficult. Therefore, automating this process could be of great utilization. This task of automation done on the basis of text could lead to discrepancies since the website itself, its various merchants, and users, all could use different terminologies for a product and its category. Thus, using images becomes a plausible solution for this problem. Dealing with images can best be done using deep learning in the form of convolutional neural networks. This is a computationally expensive task, and in order to keep the accuracy of a traditional convolutional neural network while reducing the hours it takes for the model to train, this project aims at using a technique called transfer learning. Transfer learning refers to sharing the knowledge gained from one task for another where new model does not need to be trained from scratch in order to reduce the time it takes for training. This project aims at using various product images belonging to five categories from an ecommerce platform and developing an algorithm that can accurately classify products in their respective categories while taking as less time as possible. The goal is to first test the performance of transfer learning against traditional convolutional networks. Then the next step is to apply transfer learning to the downloaded dataset and assess its performance on the accuracy and time taken to classify test data that the model has never seen before.
Khanuja, Rashmeet Kaur, "Optimizing E-Commerce Product Classification Using Transfer Learning" (2019). Master's Projects. 679.