A Controlled Experiment on the Process Used by Developers During Internet-Scale Code Search - Finding Source Code on the Web for Remix and Reuse - page 63

When we looked at all of the queries, without regard to the independent variables,
we found two statistically significant query sequences at p < 0 . 05: specialization →

generalization, and new null.

We found that when developers specialized their query, their next likely step
was to generalize their query. We refer to this transition as SG, i.e. a specialization
followed by a generalization. We also found that developers often used a single
query to carry out search without reformulating the query. We refer to this transition
as N-null transition, i.e. a new query followed by no reformulation. In other words,
it was a session that consisted of a single query.
Further, we analyzed the sequence of query revisions by size of search target,
intention of search, and the search engine. The significant results are presented in
the Tables 4.4 - 4.6 .

→

Size of search target Significant query transitions
Block SG, RS, NR, N-null
Subsystem N-null

Table 4.4: Effect of target size on revision sequences

Intention of search Significant query transitions
As-is reuse N-null
Reference examples SG, N-null

Table 4.5: Effect of intention on revision sequences

Search engine Significant query transitions
Koders SG, NR, N-null
Google code search SG, N-null
SourceForge SG, N-null
Google RR, NS, N-null
Krugle N-null

Table 4.6: Effect of search engine on revision sequences