Task 1 Given a question and a set of paragraphs, predict if the question can be answered with the given paragraphs. If yes, return the paragraph that answers the question. Each question and paragraph is associated with a specific theme. This could be “Sports”, “English” or “Mathematics” etc. A question of a theme can be answered by one of the paragraphs in that theme Task 2 For the given questions, also predict the exact answer from the predicted paragraph. Predict the start_index and the answer_text field for the given question. Note: Both the tasks will be marked individually. However, to perform better in Task 2, your model needs to perform better in Task 1
For the entire PS check out the link:
PS LinkOur team successfully developed a highly accurate Question-Answering Pipeline, achieving an accuracy rate of over 90%. This advanced pipeline efficiently handles inquiries within a specific theme by retrieving the most relevant context paragraph. It then proceeds to search for the answer within that paragraph. To meet the space and time constraints of the competition, we utilized the powerful combination of Delade and Cross Encoder Models for paragraph retrieval, along with the Electra Small Model for question answering. These models were chosen based on their effectiveness and ability to deliver optimal performance within the given limitations.
Achievement:
Absolute Silver Rank 4
People Involved:
Abhijit Panda
Aryan Rastogi
Atharva Mohite
Krish Agrawal
Parth Bhore
Prajakta Darade
Pranjal Gadge
Rupal Shah
Tanisha Sahu
Yatharth Gupta
Domain Worked On:
Natural Language Processing
