2021/04/04 by Endri Kacupaj, Kacupaj, Endri, Joan Plepi +9 · 1 citation
Computer Science · #Advanced Graph Neural Networks #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2104.01569
openalex publication_date 2021/04/04 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
This paper addresses the task of (complex) conversational question answering\nover a knowledge graph. For this task, we propose LASAGNE (muLti-task semAntic\nparSing with trAnsformer and Graph atteNtion nEtworks). It is the first\napproach, which employs a transformer architecture extended with Graph\nAttention Networks for multi-task neural semantic parsing. LASAGNE uses a\ntransformer model for generating the base logical forms, while the Graph\nAttention model is used to exploit correlations between (entity) types and\npredicates to produce node representations. LASAGNE also includes a novel\nentity recognition module which detects, links, and ranks all relevant entities\nin the question context. We evaluate LASAGNE on a standard dataset for complex\nsequential question answering, on which it outperforms existing baseline\naverages on all question types. Specifically, we show that LASAGNE improves the\nF1-score on eight out of ten question types; in some cases, the increase in\nF1-score is more than 20% compared to the state of the art.\n