[dataset] Improve typos dataset quality

[zurk]

After we have the final dataset collected we should improve our filtering tool: https://github.com/src-d/style-analyzer/blob/master/lookout/style/typos/research/eval_dataset/prepare_dataset.py

There are still many "typos" where -ed or -s ending was changed. One of the ways is to use stemmer to base and changed tokens (identifier parts) to see if there any real difference in it. But we should be sure that real typos remain unfiltered.

@irinakhismatullina if you have undestanding how many typos are bad in our dataset please tell.

[irinakhismatullina]

I haven't checked the exact numbers, I only know that great part of wrong corrections comes from such cases.

Here're the first several wrong corrections:

pos	wrong	sugg 0	correct
4	RightSingleQoutationMark	RightSingleRotationMark	RightSingleQuotationMark
5	specifiyHeaders	specificHeaders	specifyHeaders
6	getFeatures	getFeatures	getFeature
15	SupportTypes	SupportTypes	SupportedTypes
16	ReportedDisabledException	ReportedDisabledException	ReporterDisabledException
19	LiferyPageTopPhaseListenerCompat	LifePageTopPhaseListenerCompat	LiferayPageTopPhaseListenerCompat
20	fuelguage	fuelguage	fuelgauge
21	Issue150RegressionAsync	Issue150RegressionAsync	Issue1507RegressionAsync
24	deliver_method	deliver_method	delivery_method
27	insertingAEntityWillSetItsIdentifier	insertingAEntityWillSetItsIdentifier	insertingAnEntityWillSetItsIdentifier
30	RunAlbumGrouWorkflows	RunAlbumGroupWorkflow	RunAlbumGroupWorkflows
32	colors_initialized	colors_initialized	colorsinitialized
33	getBlacklistedAlgorithmsURIs	getBlacklistedAlgorithmsURIs	getBlacklistedAlgorithmURIs
35	AcknowledgementCode	AcknowledgementCode	AcknowledgmentCode
36	oldToNewCfgNameMappingImpl	oldToNewCfgNameMappingImpl	OldToNewCfgNameMappingImpl

So what problems I see here:

1. Plurality/tense/part-of-speach changes - even people would have trouble correcting that.
2. Changes in non-letter symbols (lines 21 and 32 for example).
3. Case changes (last line).

And that's only the first 15 lines, so maybe there're some more further, I suggest we look into it when the time comes to work on the issue.

[zurk]

@irinakhismatullina I think this one is done in https://github.com/src-d/style-analyzer/pull/763, right?

[irinakhismatullina]

Most part of it, yes. I've removed as much bad examples as possible, so that not to remove any real typos. In the future we may need more filtering, but for now imo it's done.