I, along with other MALL lab members, participated in the AI Science Challenge hosted at Kaggle, which aimed at answering 8th grade multiple choice science questions. We achieved an overall accuracy of 51.2%, and stood 10th on the leaderboard.

We used an ensemble method consisting of the following approaches

• M1: Solr and Elasticsearch-based information retrieval
• M2: textual entailment model
• M3: a statistical association-based model (PMI)
• M4: a neural network-based representation learning approach over question and answer pairs
• M5: inference over a knowledge graph constructed out of existing knowledge graphs and by running OpenIE over large unstructured text data
• M6: retrieval over the constructed KG
• M7: matrix factorization for question type prediction

I was involved in M1, M3 an M7 modules of the entire effort.