Code Fragment Summarization through Expertise Modeling
Annie Ying , McGill, SOCS, PhD student

Area: Software Evolution

Abstract:

When a programmer uses general search engines to find code examples, the information accompanying a returned link does not always contain adequate cues for determining whether the link is worth-while to pursue. To mitigate this issue, in this talk, I will introduce the problem of code fragment summarization for extracting succinct cues on Web pages containing code fragments. We developed a machine learning approach using expertise-related, syntactic, and query-related features, guided by an oracle we collected: 73 summaries and 1000 human judgements of whether a code line is in a summary. Our results show that syntactic and query-related features can form summaries that approximate summaries in the oracle, with a precision of 0.55. The novel use of light-weight expertise-related features shows promise, increasing the precision to 0.58. A qualitative analysis shows that even the summaries with lower precision can be useful.

This is joint work with Prof. Martin Robillard.