Automatic construction method of software bug knowledge graph

An automatic construction method of a software bug knowledge graph, comprising:

• (1) scraping bug information in a related bug library, scraping a BugID, bug description information, and four main properties of platform, product, component and status in a bug report, and performing natural language processing, wherein the natural language processing mainly comprises steps of word segmentation and part-of-speech tagging;
• (2) identifying a noun phrase (NP), a verb phrase (VP), and a verb phrase with an open clausal complement (VVP) through regular expressions for phrase identification and according to part-of-speech tags of words in a sentence subjected to natural language processing;
• (3) performing dependency syntax parsing on the bug description information to find out a dependency relationship between specific “dependent” and “dominant” words, and extracting a grammatical structure in a bug description sentence;
• (4) constructing a relationship triple of the bug description information according to the dependency relationship between words and based on the NP and the VP extracted in step (2), wherein the constructed relationship triple is (NP1, VP, NP2), the bug description sentence usually comprises more than one relationship triple;
• (5) adding the BugID in the bug information scraped in step (1) to supplement the relationship triple of the bug description information extracted in step (4), so as to generate a relationship quadruple of a bug, wherein the generated relationship quadruple is (BugID, NP1, VP, NP2);
• (6) using property X of the bug collected in step (1) to form a property triple of the bug, wherein the property triple is (BugID, property, X), the property triple is used for further description of the bug information and is prepared for subsequent construction of a software bug knowledge graph;
• (7) extracting a domain feature for bug classification according to the property triple, and using the domain feature to promote a learning and training process of a classifier in a form of (BugID, NP1, VP, NP2, features);
• (8) performing the learning and training process with a semi-supervised support vector machine (SVM) classifier by using the extracted domain feature, and performing domain classification for the bug;
• and,
• (9) after labeling by the classifier, storing data of the labeled bug of each type in a visual database and generating the software bug knowledge graph.