Transfer Learning Using Bi-Lstm With Attention Mechanism On Stack Exchange Data

Stack Exchange is a collection of diversity of topics from different fields which needs the most dominant methods for tags prediction. In our work, we have opted to improve the accuracy by using transfer learning. The Transfer learning reports the problem of how to use amply of categorized data in the source to crack related but diverse problems in a target domain, even though the problems to be trained and tested have distinct features. We trained about 87K instances of data using Bi-LSTM with attention mechanism. We used about 81K of unknown data from physics background for predicting the tags through the knowledge gained by our model. Previous implementation works could hardly achieve an accuracy of 66.89%. Our work has greatly improved the result by achieving an accuracy of 86.54%.