Comments on: Recommendation Engines: A Must for News Sites

By: The Cash Is In The Topics | Monday Note

The Cash Is In The Topics | Monday Note — Mon, 12 Oct 2009 15:06:08 +0000

[...] few months ago, I mentioned the importance of recommendation engines comparable to the ones used by e-commerce sites. These engines can be algorithm-based — [...]

By: The Long Tail: Coming Up Short. | Monday Note

The Long Tail: Coming Up Short. | Monday Note — Sun, 27 Sep 2009 18:11:59 +0000

[...] than we should not develop better recommendation engines: as I explained in an earlier article (see Recommendation Engines: A Must for News Sites ), I think recommendation engines hold great potential to increase the number of pageviews for the [...]

By: Opening the News | Monday Note

Opening the News | Monday Note — Sun, 22 Mar 2009 16:56:37 +0000

[...] value from the “long tail” of archives, either through recommendation engines (see Monday Note #73) or through well disseminated [...]

By: Lisa

Lisa — Thu, 19 Feb 2009 16:38:06 +0000

Recommendation engines are important for news sites. Period.

You site different examples of implicit recommendation in your post, all of which make sense and will resonate with different groups of users. The Amazon approach is actually a catch-all for passive recommendation since it combines a number of means to push additional items to users based on:

your prior purchases
your prior searches
others who have purchased items similar to you
your explicit wishlist
… and so on

The issue for most news organizations comes down to bandwidth issues - not having enough resources internally to build and support a system that can read peripheral metadata about their content. It is a challenge they face. However, implementing such a system is a necessity in order to create/increase user stickiness as well as loyalty. Users don’t want to invest their time in discovering additional news, if it is not in front of them, they will likely leave your site and revert back to a search engine as a means of new discovery. (Something we like to think does not happen anymore, but does in large numbers.)

I have seen a number of good apps out there, some open-source, others require payment, but all will require a point person internally to program, QA, maintain, etc. In the end, it comes down to a business decision… prioritizing what you think will get you to where you ultimately want to be.

By: Mike Orren

Mike Orren — Wed, 18 Feb 2009 23:15:14 +0000

My original comment appears stuck in moderation (probably because of links), but we’ve got a live passive recommendation engine on a Dallas, TX news site that does much of what you describe here. We use it to deliver individually customized news and advertising.

The vector you describe is the missing piece– some good ideas here that we’ll look at incorporating.

By: John Einar Sandvand

John Einar Sandvand — Wed, 18 Feb 2009 17:41:40 +0000

Thank you for a very interesting article! I think you are right in pointing out that news sites should use recommendation engines. Far too long news organizations have assumed - wrongly, in my opinion - that we are very competent in structuring information. We are not. Instead news organizations typically only organize their material along two dimensions: Date of publication (recent is big) and which main section of the site/paper content belongs to. All other meta information - like geo coordinates, tags, etc - we have been poor at adding.

But how to do this now? Are there any automatic recommendation engines out there which would work perfect for news sites? Anyone has suggestions?

By: Mike Orren

Mike Orren — Wed, 18 Feb 2009 03:10:42 +0000

Frédéric , we’re not exactly to your spec, but we’re pretty close with a passive filtering engine that’s been running for a year on our Dallas, TX-based site and that now has an integrated ad server:

http://www.pegasusnews.com/about/dailyyou/

http://www.pegasusnews.com/blogs/pegasusnewsblog/2009/feb/02/adserver/

It’s still far from perfect, but it’s a lot of fun to be able to give contest prizes to readers who haven’t even entered, simply based on what we know they like:

http://www.pegasusnews.com/blogs/pegasusnewsblog/2008/feb/05/uncontest/

By: Aimée

Aimée — Mon, 16 Feb 2009 16:33:08 +0000

Hi Fréd, great article and so true, take a look at the site Lepoint.fr, we implemented a tool (antidot) that gives readers (automatically) links to other articles - no tagging, it’s an algorythm. Of course we can always do more, but this is a good start, and economically viable !
Aimée
ps and as nicolas says, all of this does take some kind of investment - either human or technological - and the hardest part (as you know, I know) is figuring out how to pay for it !

By: Frédéric

Frédéric — Mon, 16 Feb 2009 14:03:55 +0000

Tagging is not the only way to make a recommendation engine work properly, you can use logs of IP and URL from servers to make a “People who read this article also read…” fonction.
However, there is one rule : The more datas the better the recommandation is. Corollary: It’s a bit difficult to make such algorythms work on a single blog.

PS: happy that you open your blog to comments!

By: Systemy rekomendacji newsów « Wiktor Cegła

Systemy rekomendacji newsów « Wiktor Cegła — Mon, 16 Feb 2009 10:45:07 +0000

[...] Otagowane Marketing internetowy, Nowe trendy, serwisy internetowe, treści No właśnie. Dlaczego nikt nie robi rekomendacji wiadomości i treści na podobnej zasadzie, jak Amazon? Monday Note rozbija temat na drobne. [...]

By: nicolas.

nicolas. — Mon, 16 Feb 2009 09:13:05 +0000

FF,

Although you’re right overall, I totally agree with Stan that recommendation engines require more than good will.

Changing a website from shovelware (even if it’s realtime, full of links and well indexed) into a library does require a shift in perspective. The value proposition has to change: A media isn’t there to add new stories every day; rather, it should organize all the content of its niche (while updating it in real time, that goes without saying).

Changing opinions is necessary but insufficient. Hiring a tagger would costs around 25k€ a year (or consider this expense the time spent by j’lists tagging). Add development costs and the bill easily reaches 30,000€. Who can afford that today?

Do you have data on how much Amazon spends on its recommendation engine, for instance?

By: Stan

Stan — Mon, 16 Feb 2009 08:51:20 +0000

Let’s start with your article ! I see no tag around it. The reason is : it’s a real job to ‘tag’ stuff like article or news.
Like you, editors need to hire new profiles instead of having more and more journalists. They should have a ‘library’ department with professionnal archivists among other techies. Editors should understand that Internet is not only a ‘realtime’ media but over all a (The) big library. Editors should understand that their content need non-journalist work on it to raise their value. Stan.