In-domain data size

[turchmo]

Do u run any experience or results in running the algorithm using smaller size of the in-domain data (e.g. 15k)? How does the quality of the retrieved segments change?

[amirkamran]

The smallest in-domain size we experimented with is 50k sentences. However, it is not about size of the data, it is more about the quality of the data. If the 15k set is truly a representative of a particular domain then the algorithm should work.