The increase in cyberattacks has made the cybersecurity problem worse. To navigate the ever-changing and intricate cyber landscape, cyber threat intelligence is essential. Since most cyber threat intelligence is unstructured, security analysts have a hard time keeping up with the massive amounts of data. Entity extraction, co-reference resolution, relation extraction, and knowledge graph building are the four cornerstones of the new approach to threat intelligence information extraction that is proposed in this research. A number of models are employed in this process. In order to extract the word dependence relationships for the entity extraction task, a multihead self-attention strategy is used. To enhance mention representation, the co-reference resolution method incorporates both mention embedding and contextual information. A convolutional neural network can retrieve features from multiple dimensions at once. The relation extraction task enhances the embedding representation by incorporating entity type, distance between entity pairs, part of speech, mention breadth, and relational distance. Lastly, a knowledge graph is created to formally define things and their interactions. Our model outperforms the baseline model in entity extraction (F1 score of 8.87), coreference resolution (F1 score of 9.82), and relation extraction (F1 score of 10.56). Our technology is demonstrated by Neo4j’s knowledge graph.