MEIM: A Multi-source Software Knowledge Entity Extraction Integration Model

Entity recognition and extraction are the foundations of knowledge graph construction. Entity data in the field of software engineering come from different platforms and communities, and have different formats. This paper divides multi-source software knowledge entities into unstructured data, semi-structured data and code data. For these different types of data, Bi-directional Long ShortTerm Memory (Bi-LSTM) with Conditional Random Field (CRF), template matching, and abstract syntax tree are used and integrated into a multi-source software knowledge entity extraction integration model (MEIM) to extract software entities. The model can be updated continuously based on user’s feedbacks to improve the accuracy. To deal with the shortage of entity annotation datasets, keyword extraction methods based on Term Frequency–Inverse Document Frequency (TF-IDF), TextRank, and K-Means are applied to annotate tasks. The proposed MEIM model is applied to the Spring Boot framework, which demonstrates good adaptability. The extracted entities are used to construct a knowledge graph, which is applied to association retrieval and association visualization.