A Data-Based Approach for Computer Domain Knowledge Representation

Representation learning is a method to compute the corresponding vectorized representations of entities or relationships. It is one of the most basic and essential natural language processing tasks. Current computer domain knowledge modeling techniques have two flaws: (1) the neglect of fine-grained knowledge hierarchies, and (2) the lack of a unified reference standard for modeling domain information. The fine-grained knowledge hierarchy includes knowledge domains, units, and topics. We use the Computer Science Guidelines as a standard to annotate an unstructured and unlabeled corpus in the computer domain with knowledge annotation and topic mapping. We organise the corpus into a computer domain knowledge system with a three-level hierarchy. We propose a knowledge representation method that incorporates contextual semantic information and topic information. The method can be applied to discover connections between knowledge of entities of different granularity. We compare it with several existing textual representation methods. Experimental results on extracting knowledge representations in computer domains show that combining contextual semantic information and topic information methods are more effective than single ones.