Vocabulary dictionary

Kanji dictionary

Grammar dictionary

Sentence lookup

test
 

Forums - Dictionary results

Top > renshuu.org > Feature Requests/Improvements



avatar
Nicolas2
Level: 31
If I search "general", and are at the top (#2 and #4, but doesn't appear in the first page.
For comparison, they are the top 3 of Aedict.
Your #6 is , because the 10th definition or so is "golden general" (whatever that may or may not be).
So, yes, you have quite a conundrum, because common words have priority. But words whose definition matches better should also have priority, and here there are 2 aspects:
1-if the matching definition of one word is exactly the request, then it should trump the words which contain the request but diluted in a longer expression (if request="general" then ("general") > スト ("general strike").
2- if the request is in the first or second definition of the word, then it should trump the words that have the request in the 6th, 8th or 10th definition
And by looking at the first page of results, I see . Well, this 3rd rule (that could be applied before?) would seem reasonable:
3- if 2 words match the request, but the first one is contained in the second one, the small one should always have priority on the compound.

And there's one (or two) more thing: I thought it could get better if I filter nouns only, but no, is still out. And the "has sentences" doesn't work. Also, the filters take half of the space. I tapped a bit everywhere outside the filter zone before trying to tap the cog button again. Maybe if the cog button was colour-reversed (or another obvious colour change) when pressed down, it would be more intuitive.

kqdhtyonjktj8gtk47uhbue7m.jpg
1
5 years ago
Report Content
avatar
じゅうり
Level: 1685
I think it's the new "common" and "news" tags that are skewing the results. They are making the dictionary function more like a search engine.

If you search "window",  is the second result and す is the fourth. Anything related to window is halfway down the page.
Searching for "night",  い is the first result. is fourth.  ナイト is about halfway down the page.

Could we make the popularity tags an optional feature? (off by default). That way those of us who want a traditional dictionary can have one, and those who want a search engine get what they want.

The dictionary needs to be usable by everyone from the greenest newbie who only knows hiragana to native level users.   Right now, except for unique words, it is not functional.

0
5 years ago
Report Content
avatar
マイコー
Level: 256
I agree that the English side of things needs to be improved, but the tags have nothing to do with it - those are 100% visual. You can verify this by doing the same search outside of beta (same results).

I'll add some more settings to the dictionary settings tab soon, @じゅうり! One to turn off images would be good, too (but turned on by default).

In fact, there is *no* ordering done on the terms outside of the search database software used by the site (and for the last few years), and that hasn't been tweaked in a *long* time. So, improvements definitely needed, but not a new problem.

Will update again when I have more news!
1
5 years ago
Report Content
avatar
Nicolas2
Level: 31
I think everyone could be happy with one algorithm, if you can attribute a score to each word, based on how the request X appears in its definitions. I hope this is useful…
Something like the following would put the right words on top for "window", "general" and "night":
Definition 1
: If the word's definitions are D1, D2, D3... (definitions being the words separated by commas or semicolons)
Definition 2
: eachh definition is L1, L2, L3… words long (1-letter words, 2-letter words and "the" count for 0.3 or something small like that)
Rule 1
: each definition Dk in which the request X appears gives the word a score that is (1-k/10)*(1/Lk) (If k≥10, then the first factor stays at 0.1 instead)
Rule 2
: If X appears in several of the definitions, then the scores of the definitions are added to get the total score of the word
Example 1
:if the first definition of the word is exactly X, the word has a score of 1 from this definition.
Example 2
: If the request is "general", and the 12th definition of is "golden general", then has a score of 0.05 for this request (0.1*0.5)

Once you have these scores, you have words with high scores (≥0.4, maybe ?), words with average scores (≥0.13, maybe?), and words with low scores (and has 0, no???).
Rule 3
: Within each group you can sort by popularity, then finally display high score words, followed by average score words, finishing with low score words (that could be hidden a "show less pertinent words?)
These rather simple rules (not necessarily simple to implement!) would give excellent results in the examples here.

Rule 4?: If the problem with the compound being higher than its parts persist, there could be an additional rule to push up the part just above the compound?
0
5 years ago
Report Content
avatar
マイコー
Level: 256
Thanks for taking the time to think all of that out!

That being said, mixing an algorithm into a complex system and maintaining good search speeds is another thing. I took a lot of inspiration from what you wrote, and taking advantage of what was already in the system, I think I was able to put out results that, if not ideal, are a lot closer there then they were before. It takes placement within the definition into consideration (both the # of the def and the location within the def). It also takes the common-ness into account, which will nudge the previous rankings up somewhat, but not overwrite them completely.

As of finishing this post, it is still generating the data in the search tables for all the terms, but it should look a lot better, sooner than later.

I will add the options for more dictionary customization later on!
0
5 years ago
Report Content
avatar
Nicolas2
Level: 31
It really is much better, I'm glad my little attempt helped you. And I'm quite impressed how quickly you managed this improvement. Just some fine tuning ahead now.
If it's possible (CPU intensive?), Implementing the weight of the request within each definition (1/Lk) would push (Lk=1) above all Lk=2 words: "general manager", "general opinion", and other general stuff (btw in the calculation of Lk, the word "to" should probably count for 0, as when you search for a verb you omit "to"). But hey, is now much, much higher than before.
I wondered how you handle words requests that could be either English or Japanese. Like the other dictionaries, you don't. But at least you don't convert requests like fun into fun*, so it's not so bad (good luck finding しみ among Aedict's answers for "fun"). There are actually not that many of such words, which you could find once and for all: a, do, go, man, men, fun, ban, tan, hire, mama, mane...shidekobushi (is that the longest?). There could be an exceptional button? "These Japanese words' meaning correspond to the English X. See Japanese -> English instead"
But more interestingly is that while thinking about requests that could be either Ja or En, I noticed that your dictionary does not search de-inflected forms
 (There's at least one open source module on github that does that. Not that you could use it directly, of course...). Or maybe that's CPU intensive and limited to premium?
0
5 years ago
Report Content
avatar
じゅうり
Level: 1685
Much better. Saru is gone from the "window" search. The only problem I can see is that rare and unusual terms will not appear in searches any more. Maybe you can work the algorithm to put them at the end of the list of related terms?

The pictures don't really bother me, but the tags do. They just force more scrolling....and scrolling....to find what you want.

I usually search in English simply because it returns more words. Searching in kana or kanji usually returns one term. I'm usually looking for related words or expressions to add to lessons.

Off topic, any way these can be combined into a single entry? To me, this is the kind of orthographic variation that SHOULD be combined.

0hhy6530omx3hidnwg0jkdkvu.png
0
5 years ago
Report Content
avatar
マイコー
Level: 256
ジュウリ、currently, I've only adjusted the ordering, not any filtering. Let me know if you can find an example of a term not appearing at all on any of the pages of the results.

as to the katakana you gave, I agree. However, it'll need to be done manually instead of using the dictionary data I used in the recent update, so it'll take some time before I am able to write a script to start hanging that. Needs to be done, though.
0
5 years ago
Report Content
avatar
じゅうり
Level: 1685
Not a problem and no rush. Just offering an idea to reduce clutter in the dictionary. Some terms have dozens of variations.

I was only suggesting putting rare and obsolete terms at the end of searches.
1
5 years ago
Report Content
avatar
マイコー
Level: 256
Thanks!

So, I turned the tags off by default, and threw an option in there under Dictionary in settings :).
0
5 years ago
Report Content
Getting the posts




Top > renshuu.org > Feature Requests/Improvements


Loading the list
Lv.

Sorry, there was an error on renshuu! If it's OK, please describe what you were doing. This will help us fix the issue.

Characters to show:





Use your mouse or finger to write characters in the box.
■ Katakana ■ Hiragana