I think it would be cool if one of the features you add in the future would be something similar to the ‘reading journal’ on storygraph. Something that lets you record incrementally as you read + make extra comments. As an example, on tumbler I’ve been recording my comments about the japanese version of magic thief, example after I read every chapter. And I think that would be a fun thing to have on our profiles to actually show how far we are in the books and our current thoughts on it, etc.
2 Likes
Thanks for letting me know about storygraph - hadn’t heard of it before and always like finding a new book tracking site 
I agree! Updating progress, goal setting & journal entries are definitely in the roadmap and I really would love to add them too. They’re staples of book tracking sites. Bookmeter does a tweet interface for status updates that I kinda like, but i’ll have to think on it and chat with you all on what sort of interface we’d like best.
1 Like
How many users / active users do you have on the site so far?
There are about 1200 users at this point. It hasn’t grown a ton the past few months, but it’s relatively steady. There’s been a lot of time spent on my core infrastructure & interfaces (search page & functionality, personal library management, scaling up the books database) which I’m now feeling much more confident about.
The next few sprints, however, will be entirely focused on building out the community, which will not only include more outreach from me, but I expect there to be a bit more viral growth. Stuff like showcasing community activity, following, building out a great review interface (markdown, upvoting, sharing) & public boards for lists & reviews. The latter will involve me directly reaching out to certain users / influencers to bring their content on platform, along with more general announcements in other forums like reddit.
I’m very excited about it and think it’ll really add fuel to the fire.
6 Likes
Any plan to add a recommendations section too? Either customized for a given user by comparing their list (and ratings) with other users’ lists. Or just an “if you like book A you might like book B” kind of thing based on what other users who like the selected book have also liked.
3 Likes
I’d love to… but I think that’d be a v2 of community. My gut says that I’d need at least somewhere around 10,000 users (or around 10x my current size) or perhaps much larger before it’d be useful… i’m not really sure. I’m hopeful that the following feature could satisfy some of that need.
I know it’d be really nice though! Along with content tags
2 Likes
I’ll also add too - the big selling points for the website right now are things you can’t easily get anywhere else: a bunch of leveled Japanese books, language learner book reviews & lists, user profiles / language journeys. Right now I only do the leveling somewhat well - the other things I can really improve!
1 Like
PRODUCT UPDATE
Alright, I’ve finished the features proposed as ‘upcoming’ in the last announcement!
Summary:
Contracted out book additions, proactively adding >70 unique series a week
You can now expect the book database to continue to increase at a steady rate. We’ve gone through the top 500 manga on AniList and we are now working through the top 500 light novels, which will take us a little while. I’m aiming for > 100 unique series a week, but we’re only at 70 currently. We now have 733 manga series on the site!
Improved search - added romaji, kana and english to many titles, authors are clickable
Search is much better at handling non kanji searches. For most manga / light novels, there’s a good chance I now will return the right result for non kanji searches. I’ll be surfacing those alternative spellings on the book page soon, but they’re in search now. As for non manga / LNs, I use an automatic kana converter for the titles, but the kana / romaji performance is less reliable. Something like “konbini ningen” works, but strangely for 君の名は it converts to 君 to kun… so YMMV.
Author names are now clickable too!
Improved temporary ratings again
I’ve put a lot more work into temporary ratings and have now moved to a completely custom system. The previous system simply used an elo system with a large K value, but that had the tendency of weighting the last grade way too much. The new system is a bit complicated to explain, but it’s much more intelligent now. It first tries to make bounds around reasonable guesses and then utilizes a limited elo system (smaller k) starting from the middle of those bounds. Needless to say, I’ve gone back and recalculated some temporary ratings for some books and it produces much better results!
Preliminary investigation to Bradley-Terry model
While I was improving the temporary rating system, I was inspired to take a look at the bradley terry model (Bradley–Terry model - Wikipedia) utilizing a prebuilt python package (choix — choix 0.3.5 documentation). I was able to replicate my current rankings quite easily, which leads me to believe I could implement it if I wanted to. It would still take a lot of work to fully analyze the results, create mappings that make sense with the current leveling schemes and grading processes… but it’s promising.
The Bradley-Terry model can be thought of as a ‘static’ elo system, which would resolve the issue of recency with the current system. However, the larger problem of too large of influence from particular users would not be resolved, so the benefits may not be worth it until I figure that out.
Edit: I’ll also add that I’ve tried to look into partial pairwise rankings, but I must admit I’ve struggled to parse them and figure out something useful.
Improved Grading Interface (progress bar!), better comparison generation, 'Extra' gradings
The progress bar is backk!! You’ll notice that if you’re grading a book that is temporary, you may get some ~ because it can fluctuate how many gradings you have. You’ll also notice that I prioritize comparing books of the same type first. So if you’re grading a manga, I first surface other manga before going to other book types. I think that this will eliminate the need for custom comparisons… but let me know @seanblue!
You can now do ‘extra’ gradings on a book even if you’ve completed your six! You simply can go to that book and go to the grading section. If there are gradings available, you’ll see a ‘+ grade extra’. The gradings will not impact the difficulty ratings currently (I will use them in the future though, when I figure out a better algorithm), but they will show on the book pages. I expect that I will have a general ‘extra’ grader soon, so you don’t have to choose a particular book from a book page… but I think this works well for now!
Upcoming
I’m very excited for the next steps. As I’ve mentioned before, I’ll be focusing on community for the next few sprints. Originally I thought I would start with public boards for lists, but I’ve decided to start with following and following activity and then move onto reviews. For this sprint, I’ll be focusing on:
- following mechanism, showcase followers on profile, simple follower activity feed
- add follower privacy settings
This may be a bit shorter sprint, but I think my sprints have been going too long anyway.
7 Likes
I just did a grading an hour ago and saw the progress bar. That’s definitely a helpful addition. And I didn’t notice at the time, but thinking back they were all manga to manga comparisons, so that was definitely nice.
As for custom comparisons, I still want to compare All The ThingsTM. But I can survive until you work out the algorithm so that we can grade as much as we want without adverse effect. At a certain point, I want to compare books just for the fun of it.
But I do think there’s value in users being able to see many comparisons for any two books. Yes, the main draw of the site is the overall rating system (which doesn’t require many comparisons between the same books), but I think the more individual comparisons the better as well since it would make comparisons between two specific books more useful. For example, given books A and B with the same difficulty score, I think knowing that 10/10 people said A is similar to B versus 5/10 said A was easier than B but 5/10 said A was harder than B can be valuable. And the more people that compare books A and B, the more useful that information will be.
4 Likes
I agree! That’s the whole point of the ‘extra’ gradings currently. They don’t impact the difficulty ratings, but they do surface on the book pages, so they’re still useful. They are still limited to only showcasing comparisons within 5 levels, so not quite AllTheThings, but pretty close. Let me know if you run into any outliers there.
1 Like
I didn’t realize this was already possible. I’ll have to take a look tomorrow.
Yeah, I was exaggerating a bit anyway. I definitely don’t need to compare Aria with Re:Zero or anything crazy like that. I mostly want to be able to compare manga from the まんがタイムきらら magazines. Most of those should be within 5 levels, though I think one or two might be 6 apart. I get that this is +/- though, so increasing it to something like 10 would be quite the range. You could maybe justify +/- 7 if you ever want to expand it though, since that would allow a nice total of 15 levels for comparison, as opposed to the current 11.
1 Like
So I wanted to try this and I admit I’m not sure if I’m misremembering how it was before or if I’ve encountered a bug. Please disregard if I completely forgot/confused prior functionality.
What I think was the case before is that my rating of the first book the series was basically it, no more rating of the individual volumes. When I clicked the +grade extra however I was able to grade based on the same books twice. So…feature or bug? 
1 Like
Ah nice catch! Yes your impression of what should’ve happened was correct - this is a bug. Apparently if you request to grade a later volume on a series I don’t appropriately filter out the books you’ve already graded - easy to fix
Edit: @pocketcat that bug has now been fixed and those duplicate gradings removed. Thanks again for calling that out.
2 Likes
You weren’t kidding. 
(That’s うつろ by the way.)
Searching for “utsuro” also finds it, so that’s interesting. I don’t know if you’re going through to manually fix these mismatches as they come up, but if so please remove the “sora” match.
4 Likes
Yeah, the kanji conversion is pretty terrible… I mean if it’s not getting 君の名は, perhaps I could ditch it, but I found this package and was playing in that area of the code, so I threw it in. It does catch some things, especially kana → romaji, but yeah.
I’m not going to manually fix those for now. The good news is that you will get relatively good romaji results for manga / LNs because I can find the real kana names from myanimelist / anilist, but novels are harder as I don’t have that information readily available at scale (amazon api doesn’t provide it).
The reason ‘sora’ still returned 空ろ was that I still provide the autogenerated kana even if I have inputted the correct kana, which you’re right… I probably shouldn’t do. Japanese search is hard!
Right, you should probably not auto-generate if you already have it. And on that note, you should allow corrections from users so this can improve over time.
2 Likes
What’s the name of the lib you’re using right now? I’m not an expert but I poked around in text analysis / tokenization just for fun/personal projects. Wouldn’t be shocked if others on the forum could also contribute good resources
1 Like
Agreed that would be ideal - tbh though I’m not sure how critical it is outside of manga / light novels. Manga / LNs have a large english following so people might know the romaji / english title but not the kanji title. Think that’s less the case for novels? I guess I always seem to search with kanji if I know the title, but perhaps others are different.
I use PYKAKASI documentation — pykakasi 2.0.1 documentation (i run a python backend). If anyone knows a better one, please let me know
1 Like
I’ve used Kuromoji before and found it to be an excellent tokenizer. It is, unfortunately, in Java. I did some googling to see if there was a python port for it and this looks promising although it does require Elasticsearch - supports Kuromoji as a tokenizer, though!
FWIW I ran the title “空ろの箱と零のマリア” through it via Attilika’s website (here) and got this result:
1 Like
… well seeing as I use elasticsearch for search, that very well may be a great option! Really appreciate that pointer, would be very easy to implement. 
1 Like