COMBINING INFORMATION RETRIEVAL MODULES AND ...

indicated how traditional information retrieval methods can be augmented with the features
distilled from the domain knowledge of JavaScript and software analysis to improve
classification performance. Their program comprehension included that static and dynamic
features alone do not perform well, but their combination greatly reduces individual mistakes.
The combined feature set also does not outperform the simple lexical approach, but serves to
augment its performance.
Kanellopoulos et al. 2007 presented a methodology and an associated model for
extracting information from object oriented code by applying clustering and association rules
mining. K-means clustering produces system overviews and deductions, which support further
employment of an improved version of Multiple Minimum Support (MMS) Apriori that
identifies hidden relationships between classes, methods and member data. By such a novel
algorithmic framework which combines two different kinds of data mining algorithms, one for
clustering and one for mining association rules from code, developers could have a more
comprehensive view of the system under maintenance, at various levels of abstraction.
Hill et al. 2007 proposed a technique that retrieves neighborhood information for a
software component. Their tool Dora was compared with a structural technique Suade, and two
base line techniques: Boolean-AND (AND) and Boolean-OR (OR), and Dora performed best.
Their integrated lexical-based and structural-based approach was significantly more effective in
helping programmers explore programs.
Maskeri et al. 2008 investigated latent Dirichlet allocation (LDA) in the context of
comprehending large software systems and proposed a human assisted approach based on LDA
for extracting domain topics from source code. Their results indicate that their tool was able to
satisfactorily extract some of the domain topics but not all, and certain human input is needed in
11