About Problem Statement:

• Genre: NLP
• Problem Type: Contextual Semantic Similarity, Auto generate Text-based answers

Submission Format:

• You need to generate upto 3 distractors for each Question-Answer combination
• Each distractor is a string
• The 3 distractors/strings need to be separated with a comma (,)
• Each value in Results.csv's distractor column will contain the distractors as follows: distractor_for_QnA_1 = "distractor1","distractor2","distractor3"

About the Evaluation Parameter:

• All distractor values for 1 question-answer will be converted into a vector form
• 1 vector gets generated for submitted distractors and 1 vector is generated for truth value
• cosine_similarity between these 2 vectors is evaluated
• Similarly, cosine_similarity gets evaluated for all the question-answer combinations
• Score of your submitted prediction file = mean ( cosine_similarity between distractor vectors for each entry in test.csv)

Final Model

Used GPT-2 Small and finetuned it on the test set so as to generate the output. More information is available on this Link

This final model helped me to land in top 0.5% in the hackthonon Hackerearth thus helping me getting shortlisted.