Classifying Source Code Using an Expertise Model

1 . A method for evaluating source code, comprising:

extracting, using a processor, features from source code written in a programming language; and

classifying, using the processor, the source code by comparing the extracted features to an expertise model, the expertise model modeling a usage frequency of programming features of the programming language according to a plurality of skill levels.

2 . The method of claim 1 , wherein the expertise model further models other metrics relating to usage of the programming features.

3 . The method of claim 1 , further comprising assigning a skill level evaluation to an author of the source code based on the classification of the source code into one of the plurality of skill levels.

4 . The method of claim , further comprising assigning a risk level to the source code based on the classification of the source code into one of the plurality of skill levels.

5 . The method of claim 4 , wherein the plurality of skill levels comprise at least a first skill level and a second skill level, the second skill level being associated with a lower risk level than the first skill level, the second skill level being represented in the model at least by a usage frequency histogram indicating a more frequent usage of a wider set of programming features than the first skill level.

6 . The method of claim 1 , wherein the programming features comprise at least one of lexical features of the programming language, syntactic features of the programming language, and semantic features of the programming language.

7 . The method of claim 6 , wherein the syntactic features of the programming language comprise statements, expressions, and structural elements.

8 . The method of claim 6 , wherein the semantic features of the programming language comprise relationships between the lexical features and syntactic features.

9 . The method of claim 6 , wherein extracting features from the source code comprises:

extracting lexical features of h source code based on a lexicon of the programming language;

extracting syntactic features of the source code using a parser; and

extracting semantic features of the source code based on the extracted syntactic features using a static program analysis tool.

10 . The method of claim 1 , further comprising:

performing the extracting and classifying steps on multiple modules of source code in a code base to estimate a level of expertise of authors of the modules; and

assigning a risk level to each of the modules based on the the estimated level of expertise of the author(s) of the module.

11 . A system, comprising:

a model generator to generate an expertise estimation model based on labeled examples of source code written by multiple developers each associated with one of a plurality of skill levels, the expertise estimation model modeling a usage frequency of programming features; and

a classifier to classify source code into one of the plurality of skill levels using the expertise estimation model.

12 . The system of claim 11 , further comprising:

an extractor to extract features from a module of source code. wherein the classifier is configured to classify the module of source code by comparing the extracted features to the expertise estimation model.

13 . The system of claim 12 , further comprising:

a risk estimator to estimate a risk level of the module of source code based at least on the classified skill level and an additional software risk metric.

14 . The system of claim 10 , wherein the programming features comprise lexical features, syntactic features, and semantic features.

15 . The system of claim 13 , further comprising:

a parser to extract syntactic features from the source code; and

a static program analysis tool to extract semantic features from the source code.

16 . A non-transitory computer readable storage medium storing instructions that, when executed by a processor, cause a computer to:

extract lexical features, syntactic features, and semantic features from source code; and

classify the source code by comparing the extracted lexical, syntactic, and semantic features to an expertise model, the expertise model modeling a usage frequency of the lexical, syntactic, and semantic features according to a plurality of skill levels.

17 . The storage r medium of claim 1 , further storing instructions that cause a computer to:

assign a risk estimate to the source code based on the classification.