一种多阶段交互式线索驱动的设计模式识别方法
A multi-stage approach based on interactive clues driven for design pattern identification

摘要 针对传统设计模式自动检测不够精确及不易于扩展的问题，为提高设计模式实例恢复的精确性，提出一种多阶段交互式线索驱动的设计模式识别方法。在传统基于约束满足问题（CSP）的设计模式检测思想基础上引入了线索的思想，旨在经过调研对专家经验知识进行反馈，并将筛选后有价值的线索表示为CSP形式的信息，进而依据信息特征将线索分类，通过在设计模式检测过程中逐步增加线索，直至设计模式实例候选参与者集产生。实验结果表明，本文方法不仅分阶段筛选了设计模式检测实例的假阴性与假阳性结果，还解决了设计模式识别的重叠问题，通过与其他主流检测方法的F-score指标值对比，取得了较好的检测效果。
Abstract：Aimed at inaccuracy and difficulty to extend the traditional design pattern automatic detection method, and in order to improve the accuracy of the design pattern instance recovery, an interactive clue-driven approach for design pattern detection was presented. Concept of clue was introduced based on the constraint satisfaction problem (CSP) of traditional design pattern detection, and expert experience feedback mechanism was proposed by investigation. The significant clues were converted into the information based on CSP in the refining stage, the clues were classified by information characteristics, and the clues were added in design pattern detection process until the candidate sets of design pattern instance were produced. Experimental results show that the proposed method can gradually reduce the false negative results and the false positive results of design pattern detection instance, and further more, the novel method can solve design pattern overlap problem. Compared to the F-score index of other well-known algorithms, the proposed method shows better detection effect.