COMET is a generative model designed for automatic knowledge graph construction, particularly for commonsense knowledge bases like ATOMIC and ConceptNet. The model leverages pre-trained language models to generate rich and diverse commonsense knowledge, enabling the creation of new knowledge tuples. COMET is trained on existing knowledge tuples to learn how to generate new ones, achieving high precision in generating correct knowledge. The model outperforms existing methods, with up to 77.5% precision for ATOMIC and 91.7% for ConceptNet, approaching human performance. COMET uses a transformer architecture, with multi-headed attention and feed-forward networks, to process and generate knowledge. It is trained on seed knowledge tuples, allowing it to learn and generate new knowledge. The model's performance is evaluated using automatic metrics and human evaluations, showing high quality and novelty in generated knowledge. COMET's effectiveness is demonstrated across various domains, with results indicating its ability to generate accurate and diverse knowledge. The model's approach offers a promising alternative to extractive methods for commonsense knowledge base construction.COMET is a generative model designed for automatic knowledge graph construction, particularly for commonsense knowledge bases like ATOMIC and ConceptNet. The model leverages pre-trained language models to generate rich and diverse commonsense knowledge, enabling the creation of new knowledge tuples. COMET is trained on existing knowledge tuples to learn how to generate new ones, achieving high precision in generating correct knowledge. The model outperforms existing methods, with up to 77.5% precision for ATOMIC and 91.7% for ConceptNet, approaching human performance. COMET uses a transformer architecture, with multi-headed attention and feed-forward networks, to process and generate knowledge. It is trained on seed knowledge tuples, allowing it to learn and generate new knowledge. The model's performance is evaluated using automatic metrics and human evaluations, showing high quality and novelty in generated knowledge. COMET's effectiveness is demonstrated across various domains, with results indicating its ability to generate accurate and diverse knowledge. The model's approach offers a promising alternative to extractive methods for commonsense knowledge base construction.