Doing a Necro post on this as I have recently began to mine from mass media.
First some thoughts
These projects we've been doing over the last 18 months: UBJG, KO2001, AATP, Kanzen master, etc. are not sentence mining. These are transferring structured resources from the page to the computer.
Structured resources usually offer diminishing returns when you go through them item by item. What I mean is the first 100 sentences will likely be beneficial than the next 100. Learning the most useful 1000 kanji takes as much time as the next 1000 kanji. However, the benefit from the former is much greater. The same can be said of vocabulary and grammar points.
What happens is you can get into a cycle where you're wanting to cover all your bases, but you get to a point where you're investing more and more time for less gains. For example, it took me about 200 hours to finish 2000 vocabulary words from iKnow. Most of those words will be seen a lot, although I doubt some I'll ever see. However, to invest 200 more hours for the next 2000 words will not net me immediate benefits. Nothing new here for those that went from RTK1 to RTK3.
Anyway, last summer I started doing RTK3. Now, what I did with RTK3 is I stopped at 2500 kanji. However, while doing iKnow, if a new kanji from RTK3 comes up, I activate in my RTK deck. I was treating RTK3 as a resource that I didn't systematically go through, but instead called upon it when the kanji presented itself in other studies. Wasn't too hard, just write down the kanji and how it's pronounced then use Anki's search feature (or I used JWPce).
I am now doing something similar with iKnow's Core 6000 list. Instead of systematically going through each new word, I suspended them all in my iKnow deck. I am then sentence picking from a TV show (subs2srs use here). I go and get J-J definitions for words I know, and J-E definitions for new words. For the new words, I copy to a text file till I build up 30 new words. By that time, I pull up iKnow in Anki, and look for the words. Then it's a matter of unsuspending the sentence with that word.
Basically, I'm using RTK 3 and Core 6000 as a resource for new words and kanji that have actually come up when mining.
This is a personal choice. I could just as easily get a sample sentence from yahoo.jp or kenkyuusha. However, I like Core 6000 as the resource here due to having audio, English translation, and the likely hood of Core 10,000 coming out in time. My other thinking is: Core 6k (and later Core 10k) are common words. There's merit to memorizing words in that list. I'm just not going to do it systematically from now on.
The benefit: Actual dialogue. This is the one thing you're not getting from structured sources. That this is dialogue from a show you're interested in helps immensely. When I pick a sentence, sometimes there's not one single new word. However, that sentence is new and used is a way I would not have done spontaneously. It has a social situation attached to it (guy to girl, boss to worker, worker to president, landlord to tenant, at a bar, at lunch, at a meeting, etc.). It breaks grammar rules the right way.
Plus, I noticed something very quickly. When you "mine" a show, you slow it down to a snail's pace (sentence by sentence) and analyze it. So while I watched a show in Japanese, much of it I gloss over as I'm still getting the gist of it. Essentially, I'm "translating" the show, just without the English. With mining, I have to look at it and understand it and appreciate it, else I throw it out.
.
