5 ms·

Show HN: Idina, a customizable news aggregator powered by machine learning

39 points by idina_news 11y ago

rw2 11y ago
Curious what the machine learning is optimizing for and how this is different from getting the most popular article from twitter
- laotzu 11y ago
  Can't speak for this particular application but just off the top of my head you could use NLP to aggregate news based on more complex queries. Instead of saying "grab the most popular article" you could say something like "grab the most popular articles that talk about both Hillary Clinton in a positive tone and Bernie Sanders in a negative tone and is about the economy."
  - _b 11y ago
    You should be able to build that topic on idina. If you start with a search for "clinton sanders economy," then communicate the "clinton good, sanders bad" by rating examples of that high, and everything else low. It is hard for algorithms to understand words like "data-driven political journalism that is slightly left leaning," and, but easier to understand "stuff like this couple of Nate Silver article." So for the short term, I think we'll continue to need positive and negative examples to really get topics good for people, but someday I'd love to figure it all out from NLP parsing a general description.
- _b 11y ago
  For customized categories, the machine learning is optimizing for ordering the articles as closely as possible to how the user rated them. There are other options, like optimizing to clicks, but that can end up with more tabloidy articles than most users like. Our hope is by optimizing to explicit ratings, users will be able to find content that is more deeply interesting.
bfrog 11y ago
Its a cool idea, and definitely one that has crossed my mind a few times. I would guess you could build up common groupings of user preferences to try and provide group content filtering/rating? If not, would be fun to try
- _b 11y ago
  Agreed, moving towards collaborative filtering is a good idea. We have only a few users right now, so customized categories is mostly content based filtering and figuring out how to recombine the ranking signals for them. That often work fairly well to improve topics (sometimes drastically), but mixing in collaborative filtering seems like an exciting opportunity for improving. Also hi, I'm Ben, and I worked on this.
webwanderings 11y ago
This is pretty good. I am not a fan of any news website which does not provide RSS, but yours is where it is not needed by design. Now, if only I can increase the font size.
- _b 11y ago
  Thanks! What do you want regarding font? I'll make it happen for you. Would you like an option to set the font-size for when you are logged in? Or is there someway we could be doing our CSS differently so that zooming-in on our site meets your needs?
  - webwanderings 11y ago
    I would really prefer being able to increase the font size. With age, I do not enjoy reading anything with small fonts anymore (on desktop). I have also yet to test your site on iPhone and my Android tablet. BTW, I am a non-programmer and a type of guy who follow my own curated news (hence a proponent of RSS). Your system looks closer to my ideals.
    - DanSmooth 11y ago
      Have you tried Ctrl and +, aka the browsers capability to increase text-size if needed? Works nicely with this site (at least in Chrome/Win).
eecks 11y ago
I am making a similar site right now
- shabinesh 11y ago
  Any blog posts, refs on how you are building it? I am interested, I am also making one for my personal use and hobby.
  - eecks 11y ago
    Hadn't thought of that. How far along are you?
devit 11y ago
This seems potentially amazing and capable of replacing Google News and Reddit, although it might a while to achieve that. How is it ranking stories? It might be worth considering trying to optimize placement based on number of clicks and the time passing until the user goes back to using Idina (this is slightly privacy-intrusive, but unfortunately browsers allow this if JavaScript is enabled by listening for mouse move events etc., so might as well take advantage of it). The website could use thumbnail images for the articles, and more colors and more contrast on the pages (have titles and clickable things stand out more). It seems that most of the top articles should have a Reddit thread, but only very few have links to Reddit; also could link to HN. The popular/followed/customized UI seems quite confusing. It might make sense to autofollow all defaults, and make "followed" the default and the current popular be "explore other topics". Also, the default should probably be customized, with the non-customized version being offered in another tab. Might want a "since last visit" time range (which obviously needs to be intelligent and ignore things like refreshes, maybe make it "since last visit on day XX time YY" so the user knows right away if it makes sense). Should consider removing the "star rating" from customization, which is very hard/impossible to assign (is this worth 2 or 3 stars? how do I decide that?), and instead let users customize by ordering articles (up/down arrows + drag&drop), which is natural and easy. Maybe consider copying Google News' click-to-expand mechanism.
- idina_news 11y ago
  These are great ideas. I'll need to ponder some of them a bit, although a few immediate thoughts are below. If you have any thoughts in the future, my email is chas@idina.com (and _b is ben@idina.com). Like _b mentioned below, stories are ranked by ML fitting a bunch of signals and the article text to a user's ratings. Usage metrics is something we want to figure out how to use well. Currently we track them, but don't use them. Confession regarding images/colors/contrasts/etc: neither of us are UI people. It is something we struggle with, so doubly appreciate the feedback here. Will try to improve this soon. Regarding some missing links to reddit, are there good threads on reddit we should be linking to for them, or has reddit just not picked them up? Our algorithms try to find every related reddit thread to a story, although they can make mistakes. Regarding autofollowing all defaults -- it's tricky. We used to do this, and got feedback that people didn't link having to unfollow a bunch of topics when they created an account. Exactly how to package the experience is an open question, one that is complicated by the desire to have a good logged-out experience that transitions as smoothly to the logged-in experience as possible. Point being, you hit on a good issue here that we don't yet have a perfect answer to. 'Since last visit' time range sounds neat. Will look into adding. Re: star ratings -- manually ordering articles (this should rank above that) or having up/down arrows for each article are also natural ways to provide ranking feedback. A big concern of ours here is the ability for a user to have a clean mental model of their own feedback over a long period of time. For example, feedback in the form of up/down arrows comes necessarily in the context of a specific set of articles shown at a specific time. If someone wants to review their feedback in aggregate, it might be difficult to represent it all in one place in a way that's easy to understand. On the other hand, your criticism of it is also correct. Adding additional ways to make training data, or changing how we do this, is something we are really interested in getting right.
DanSmooth 11y ago
Some suggestions because I would like to use the site more often but right now there is too much confusion going on (although I registered an account, which was probably the least confusing part :-): - to get the ball rolling as soon as a new user has signed up, lead him to one or two steps where he would pick some topics of interest of an alphabetic sorted list and then to the "customized" tab, where you show him three stories of a selected topic and explain what wonders the stars can do (problem with the star imho: in many other applications it's often used to bookmark something, which could be happening here; replace it with a simple "rate" and in the dropdown have just three options: "Read [more], [same], or [less/fewer] like this article.") - this "my" business gets confusing really fast; perhaps rename it to "topic (customized)" so you also get the connection to the "customized" (perhaps "personalized" is a better fit? It's also used already somewhere on the site) tag; also how do I switch between "my" and the normal view? Edit menu should have an option for this - more confusion: why can I only rate the first article (in the normal view not the expanded one) and what defines what is listed at the top? Do you also save sources, so I get more suggestions from the same source? If so, a better view to see and rate articles might be in order. I might even would consider adding an option to rate articles as well as stories. One could show that you are not at all interested in a HP prequel but that you like articles from perezhilton.com nevertheless. - in the popular tab: I can follow stuff easily but have to "edit" then "unfollow" - this might actual be a bug? - perhaps there is a technical reason for it but you seem to be afraid of too much text. Instead of "X articles" write out what the user can expect behind this link, like: "View 99 additional articles about this story" ("additional" seems to be correct or else your counting is off) or "View all 102 articles about this story" or even "View and rate 99 other articles about this story"; perhaps make it italic for same basic distinction - you already "open" new tabs when I select a topic-name but it is named "search" (probably due to your underlying js-structure, where everything is a search) and - you guessed it - it confused me. Why not rename it to the correct name (e.g.: my:entertainment; or if you did go for some of my suggestions: "entertainment (personalized)". Additional tabs for clicking instead of scrolling to get to a topic might also be nice to have - that the search tab is removed if I select another tab is intrusive and unexpected - unsorted feature suggestions: hide rated articles/stories as a setting for users; sort by time; a huge topics page (popular ones at the top, less popular ones at the bottom; both sorted by ABC) with an easy to use option (a simple + at the right end of the topic link would suffice) to select topics; later on suggest topics based on selected topics (yay, more data crunching); implement a "stream"-view where instead of sorted by topics it's sorted by time just using topics you follow - bug? when I remove a rating I get a 3 next to the star but there is no actual rating - replace stars with hearts - have you not heard the news? I'm kidding with this one Sorry for probably confusing you too with my unsorted babble.
gerner 11y ago
I've been in the Idina closed beta for a couple of weeks. The #1 thing I like about this is the power of "rate and refresh". The low latency feedback loop is really impressive. Maybe I don't use enough news sites, but I haven't seen anything like that before. The breadth of content is pretty good. I was surprised I was able to build a good gaming topic that's not just AAA titles or some niche category. I'm a little worried about rating my way into an echo chamber. Having used Idina for a few weeks, that seems like a risk. For example, 4 of top 6 stories in my gaming topic are from the same site. I really like those stories and that site, but I wonder if I'm missing stuff from sites I've never seen before.
est 11y ago
My personal experience is that customizable is a great thing, however, it might turn out to be an echo chamber if over customized. Too many familiar contents, need more surprises. also need a "hide" button. The most frequently used button in reddit was mark a thread as seen by hide it, don't wanna see it again.
- idina_news 11y ago
  The risk you're talking about re: echo chambers makes sense -- so far, we're trying to mitigate the issue by making it possible to 'follow' both the customized and uncustomized versions of the topic. Any other ideas are always welcome. 'Hide' sounds like a really neat feature!

39 points by idina_news 11y ago

rw2 11y ago

Curious what the machine learning is optimizing for and how this is different from getting the most popular article from twitter

laotzu 11y ago
Can't speak for this particular application but just off the top of my head you could use NLP to aggregate news based on more complex queries. Instead of saying "grab the most popular article" you could say something like "grab the most popular articles that talk about both Hillary Clinton in a positive tone and Bernie Sanders in a negative tone and is about the economy."
- _b 11y ago
  You should be able to build that topic on idina. If you start with a search for "clinton sanders economy," then communicate the "clinton good, sanders bad" by rating examples of that high, and everything else low. It is hard for algorithms to understand words like "data-driven political journalism that is slightly left leaning," and, but easier to understand "stuff like this couple of Nate Silver article." So for the short term, I think we'll continue to need positive and negative examples to really get topics good for people, but someday I'd love to figure it all out from NLP parsing a general description.
_b 11y ago
For customized categories, the machine learning is optimizing for ordering the articles as closely as possible to how the user rated them. There are other options, like optimizing to clicks, but that can end up with more tabloidy articles than most users like. Our hope is by optimizing to explicit ratings, users will be able to find content that is more deeply interesting.

bfrog 11y ago

Its a cool idea, and definitely one that has crossed my mind a few times. I would guess you could build up common groupings of user preferences to try and provide group content filtering/rating? If not, would be fun to try

_b 11y ago
Agreed, moving towards collaborative filtering is a good idea. We have only a few users right now, so customized categories is mostly content based filtering and figuring out how to recombine the ranking signals for them. That often work fairly well to improve topics (sometimes drastically), but mixing in collaborative filtering seems like an exciting opportunity for improving. Also hi, I'm Ben, and I worked on this.

webwanderings 11y ago

This is pretty good. I am not a fan of any news website which does not provide RSS, but yours is where it is not needed by design. Now, if only I can increase the font size.

_b 11y ago
Thanks! What do you want regarding font? I'll make it happen for you. Would you like an option to set the font-size for when you are logged in? Or is there someway we could be doing our CSS differently so that zooming-in on our site meets your needs?
- webwanderings 11y ago
  I would really prefer being able to increase the font size. With age, I do not enjoy reading anything with small fonts anymore (on desktop). I have also yet to test your site on iPhone and my Android tablet. BTW, I am a non-programmer and a type of guy who follow my own curated news (hence a proponent of RSS). Your system looks closer to my ideals.
  - DanSmooth 11y ago
    Have you tried Ctrl and +, aka the browsers capability to increase text-size if needed? Works nicely with this site (at least in Chrome/Win).

eecks 11y ago

I am making a similar site right now

shabinesh 11y ago
Any blog posts, refs on how you are building it? I am interested, I am also making one for my personal use and hobby.
- eecks 11y ago
  Hadn't thought of that. How far along are you?

devit 11y ago

This seems potentially amazing and capable of replacing Google News and Reddit, although it might a while to achieve that. How is it ranking stories? It might be worth considering trying to optimize placement based on number of clicks and the time passing until the user goes back to using Idina (this is slightly privacy-intrusive, but unfortunately browsers allow this if JavaScript is enabled by listening for mouse move events etc., so might as well take advantage of it). The website could use thumbnail images for the articles, and more colors and more contrast on the pages (have titles and clickable things stand out more). It seems that most of the top articles should have a Reddit thread, but only very few have links to Reddit; also could link to HN. The popular/followed/customized UI seems quite confusing. It might make sense to autofollow all defaults, and make "followed" the default and the current popular be "explore other topics". Also, the default should probably be customized, with the non-customized version being offered in another tab. Might want a "since last visit" time range (which obviously needs to be intelligent and ignore things like refreshes, maybe make it "since last visit on day XX time YY" so the user knows right away if it makes sense). Should consider removing the "star rating" from customization, which is very hard/impossible to assign (is this worth 2 or 3 stars? how do I decide that?), and instead let users customize by ordering articles (up/down arrows + drag&drop), which is natural and easy. Maybe consider copying Google News' click-to-expand mechanism.

idina_news 11y ago
These are great ideas. I'll need to ponder some of them a bit, although a few immediate thoughts are below. If you have any thoughts in the future, my email is chas@idina.com (and _b is ben@idina.com). Like _b mentioned below, stories are ranked by ML fitting a bunch of signals and the article text to a user's ratings. Usage metrics is something we want to figure out how to use well. Currently we track them, but don't use them. Confession regarding images/colors/contrasts/etc: neither of us are UI people. It is something we struggle with, so doubly appreciate the feedback here. Will try to improve this soon. Regarding some missing links to reddit, are there good threads on reddit we should be linking to for them, or has reddit just not picked them up? Our algorithms try to find every related reddit thread to a story, although they can make mistakes. Regarding autofollowing all defaults -- it's tricky. We used to do this, and got feedback that people didn't link having to unfollow a bunch of topics when they created an account. Exactly how to package the experience is an open question, one that is complicated by the desire to have a good logged-out experience that transitions as smoothly to the logged-in experience as possible. Point being, you hit on a good issue here that we don't yet have a perfect answer to. 'Since last visit' time range sounds neat. Will look into adding. Re: star ratings -- manually ordering articles (this should rank above that) or having up/down arrows for each article are also natural ways to provide ranking feedback. A big concern of ours here is the ability for a user to have a clean mental model of their own feedback over a long period of time. For example, feedback in the form of up/down arrows comes necessarily in the context of a specific set of articles shown at a specific time. If someone wants to review their feedback in aggregate, it might be difficult to represent it all in one place in a way that's easy to understand. On the other hand, your criticism of it is also correct. Adding additional ways to make training data, or changing how we do this, is something we are really interested in getting right.

DanSmooth 11y ago

Some suggestions because I would like to use the site more often but right now there is too much confusion going on (although I registered an account, which was probably the least confusing part :-): - to get the ball rolling as soon as a new user has signed up, lead him to one or two steps where he would pick some topics of interest of an alphabetic sorted list and then to the "customized" tab, where you show him three stories of a selected topic and explain what wonders the stars can do (problem with the star imho: in many other applications it's often used to bookmark something, which could be happening here; replace it with a simple "rate" and in the dropdown have just three options: "Read [more], [same], or [less/fewer] like this article.") - this "my" business gets confusing really fast; perhaps rename it to "topic (customized)" so you also get the connection to the "customized" (perhaps "personalized" is a better fit? It's also used already somewhere on the site) tag; also how do I switch between "my" and the normal view? Edit menu should have an option for this - more confusion: why can I only rate the first article (in the normal view not the expanded one) and what defines what is listed at the top? Do you also save sources, so I get more suggestions from the same source? If so, a better view to see and rate articles might be in order. I might even would consider adding an option to rate articles as well as stories. One could show that you are not at all interested in a HP prequel but that you like articles from perezhilton.com nevertheless. - in the popular tab: I can follow stuff easily but have to "edit" then "unfollow" - this might actual be a bug? - perhaps there is a technical reason for it but you seem to be afraid of too much text. Instead of "X articles" write out what the user can expect behind this link, like: "View 99 additional articles about this story" ("additional" seems to be correct or else your counting is off) or "View all 102 articles about this story" or even "View and rate 99 other articles about this story"; perhaps make it italic for same basic distinction - you already "open" new tabs when I select a topic-name but it is named "search" (probably due to your underlying js-structure, where everything is a search) and - you guessed it - it confused me. Why not rename it to the correct name (e.g.: my:entertainment; or if you did go for some of my suggestions: "entertainment (personalized)". Additional tabs for clicking instead of scrolling to get to a topic might also be nice to have - that the search tab is removed if I select another tab is intrusive and unexpected - unsorted feature suggestions: hide rated articles/stories as a setting for users; sort by time; a huge topics page (popular ones at the top, less popular ones at the bottom; both sorted by ABC) with an easy to use option (a simple + at the right end of the topic link would suffice) to select topics; later on suggest topics based on selected topics (yay, more data crunching); implement a "stream"-view where instead of sorted by topics it's sorted by time just using topics you follow - bug? when I remove a rating I get a 3 next to the star but there is no actual rating - replace stars with hearts - have you not heard the news? I'm kidding with this one Sorry for probably confusing you too with my unsorted babble.

gerner 11y ago

I've been in the Idina closed beta for a couple of weeks. The #1 thing I like about this is the power of "rate and refresh". The low latency feedback loop is really impressive. Maybe I don't use enough news sites, but I haven't seen anything like that before. The breadth of content is pretty good. I was surprised I was able to build a good gaming topic that's not just AAA titles or some niche category. I'm a little worried about rating my way into an echo chamber. Having used Idina for a few weeks, that seems like a risk. For example, 4 of top 6 stories in my gaming topic are from the same site. I really like those stories and that site, but I wonder if I'm missing stuff from sites I've never seen before.

est 11y ago

My personal experience is that customizable is a great thing, however, it might turn out to be an echo chamber if over customized. Too many familiar contents, need more surprises. also need a "hide" button. The most frequently used button in reddit was mark a thread as seen by hide it, don't wanna see it again.

idina_news 11y ago
The risk you're talking about re: echo chambers makes sense -- so far, we're trying to mitigate the issue by making it possible to 'follow' both the customized and uncustomized versions of the topic. Any other ideas are always welcome. 'Hide' sounds like a really neat feature!