nlp

Ongoing, Coded backend, managed database and indexation for a Korean to Chinese sentence translation and POS-tagging dataset. Utilized TFIDF and Boolean indexing as methods to index and COP K-means to cluster sentences to topics based on PMI.

Fine-tuned multiple ERNIE and BERT models on given Chinese riddle multiple choice dataset using Huggingface. After hyperparameter optimization with TPE and data augmentation, improved original paper’s performance from 59.3% to 63.0%. Combined pretrained results to heuristic choice elimination methods and surpassed all other 20 participating teams accuracy-wise (72.3%).