WikiQA,复现论文《APPLYING DEEP LEARNING TO ANSWER SELECTION: A STUDY AND AN OPEN TASK》
-
Updated
Jul 25, 2019 - Python
WikiQA,复现论文《APPLYING DEEP LEARNING TO ANSWER SELECTION: A STUDY AND AN OPEN TASK》
LSTM model for WikiQA dataset using Keras
A simple Question Answering system built using the WikiQA dataset. It takes a user question, retrieves candidate sentences from Wikipedia, and selects the most relevant answer or returns "No answer found."
Answer selection (reranking) on WikiQA: TF-IDF + Naive Bayes, Siamese BiLSTM with contrastive loss, and a fine-tuned DistilBERT cross-encoder (0.775 MAP / 0.793 MRR). Includes class-imbalance handling and error analysis.
To associate your repository with the wikiqa topic, visit your repo's landing page and select "manage topics."