I have 6.000.000 RSS articles in a DB. What can I do? | Hacker News Reader

I have 6.000.000 RSS articles in a DB. What can I do?

30 pointsrobertpohl37 comments

I created a website a couple of years ago (thatstoday.com) where members can add RSS feeds and read them in our "news reader". The years went by and we've indexed all feed articles in our db (and in Lucene).

Now I'm not sure what to do with all the data since the db it's getting large and difficult to handle.

Should I delete it or create a new service? What do you guys suggest?

Thanks, Rob

View on HN

> the db is getting large and difficult to handle.

Did your store the raw content in the database? If so, you might consider writing files instead. These blobs, like pictures, are better stored as files. In your database, you should keep relational data: url, time, adder, etc, properly indexed (probably by adder, time and maybe keywords). Then a 6M rows table is quite a small thing for any RDBMS (if your SELECTs are filtering on indexed columns).

Good idea! When i'm importing the raw data I add it to lucene to make similar-articles-matching, but after that it could be dumped to text.

You've got the basis for a "you-like-this-so-perhaps-you-will-want-to-read-this" recommendation engine. Perform some - mentioned in another reply - n-gram analysis on the corpus; do some basic cosine-similarity analysis with the feeds people subscribe to and see what pops up. Try other techniques (from e.g. ICWSM[1]) (last time I did something like this is April 2007); iterate; analyse results; publish.

At least you'll have fun (YMMV)...

[1] http://www.icwsm.org/[2007-2011]

I already use "find similar articles" techniques using Lucene. But I'm thinking it might be more interesting in another more data-centered website?

Does your reader record which items in a feed were read? I'm sure lots of feed providers would be interested in metrics on which of their articles were frequently read (and possibly some clustering of interests based on what else readers subscribe to)

Yes it does! The problem is the low activity on the site. If the traffic grew the meta data would explode ;)

You should dump a copy of the database somewhere so we can all take a look at it and perhaps analyse it.

Do want. Please. :-)

Would also love this!

that is what I thought

I'd love to do that, but how can I finance storage and bandwidth? Paypal donations?

Hmm.. I know a few peeps at MS/Azure.. maybe they can sponsor..I'll get back on this one.

Torrent that. Exactly the kind of thing that BitTorrent is for.

There's no e-mail address in your profile. Please contact me at the e-mail address in mine.

I have an encrypted version there now ;)

Sell it or lease it. I'm sure there's some value in the aggregation of feeds (though I can't imagine what besides user habits).

For God's sake don't delete it! First of all dump it to files instead of DB. (Or use some NoSQL Document Storage, such as MongoDB. The structure of RSS is actually non-relational, I suppose) Second of all: is your data clean? If not then you might need to clean it from any boilerplate (such as HTML code) Then you can process it with some tools. There are some good NLP tools available such as Gate. You may have a look at them. You can do great deal of things there: - detect some entities (companies? products?) and do some classification of documents - you can detect some events (iPhone announcements, etc) - if you have time & date (hope you have) you can do some trending topics analysis (what was hot in June 2010) - probably you can't sell the data as the content of articles is not yours, but you may sell some derived data (analysis, etc)

Natural Language processing n-gram studies Or just some kind of webservice where people can look this up

Thought of that, but I need to add it to a 50GB+ cloud SQL storage which cost a few bucks... :)

Write an algorithm to match articles from different sources that are about the same story. That way you can autohide news that you've supposedly already read from another source. Clustering news from different sources would be a killer feature for me :)

Depending on how far back the data goes, you could try to spot language trends trough time or make a pretty graph of the average article length trough time. I'm not that inspiring.

I hope this'll get a follow up, I'm sure someone can think of something awesome to do with it.

If you decide to open source it (copyrights?), use Bittorrent: it's the perfect tool for this job.

How big is the DB dump?

I know I would be interested in downloading it, and just poking it for interesting stats...

50-60 GB

I suggest you change your privacy policy (i am sure you have one) and tell the users that you will open the database for researchers, and make the open database a donation based service, this way you can pay to host it and might get some good extra money.

Sounds tempting!

Are they all open? If so, I'd contact wikimedia and see what use can they give them.

Do you have the rights to distribute the articles in your database?

Donate it to http://Archive.org. They can reconstruct disappeared web sites from it, preserving the web's past.

If you think it could be useful to someone, try to sell it

How much do you offer? ;)

I'm not that someone :D

Maybe I can offer you hosting for free if this is your problem.. pls email me on ahmed(at)svwebdev.com

I can sponsor the S3 storage but with limited IO

I guess it would need more that just storage, but also an API so you can query it. Maybe a JQuery/JSONP layer ontop of a Lucene index?

make it open!

a bunch of interesting things. It is a nice NLP corpus. Put a dump on S3 and make it public

love it...

Reply on news.ycombinator.com