Combining Graph-Based Dependency Features with Convolutional Neural Network for Answer Triggering

Answer triggering is the task of selecting best-suited answer for a given question from set candidate answers if it exists. This paper presents hybrid deep learning model triggering, which combines several dependency graph-based alignment features, namely graph edit distance, similarity, and coverage, with dense vector embeddings Convolutional Neural Network (CNN). Our experiments on WikiQA dataset show that such combination can more accurately trigger compared to previous state-of-the-art models. Comparative study data shows $$5.86\%$$ absolute F-score improvement at level.