chandnii7/Transfer-Loss-NLP

Program to analyze transfer loss across domains using TF-IDF vectors with Chi squared into logistic regression model.

Transfer Loss Across Domain

Program to analyze transfer loss across domains using example of books and electronics. Creating TF-IDF vectors and selecting k best with Chi Squared. Logistic regression model is used for training.

Dataset:

1. Source Domain: Books

   • number of positive reviews = 1000
   • number of negative reviews = 1000
   • source domain training set vector: (2000, 4500)

2. Target Domain: Electronics

   • number of positive reviews = 1000
   • number of negative reviews = 1000
   • target domain training set vector: (1600, 4500)
   • target domain test set vector: (400, 4500)

Result:

1. Direct Transfer:
   • Training a logistic regression classifier on the Electronics training dataset.
   • Evaluating it on the Electronics test dataset.

2. Cross-domain Transfer:
   • Training a logistic regression classifier on the Books training dataset.
   • Evaluating it on the Electronics test dataset.

3. Transfer Loss Across Domains:
   • LOSS = direct_transfer_accuracy - cross_domain_transfer_accuracy = 0.39