Tuesday, February 07, 2006

Don Dodge interviews Findory

Don Dodge posted an interview with me about Findory.

Don was Director of Engineering at AltaVista. He is now Director of Business Development in Microsoft's Emerging Business team.

Don's weblog is a good read. Don't miss this old post where he talks about AltaVista and Google.

6 comments:

rr said...

Google's news recommendations seems to be getting better since the time you first blogged about it. It now removes the articles I read under the recommendations column(I'm not sure if the article is removed when clicked on in another column).

And it is pretty relevant. From my experience (I use Google news alot), it seems to work like the following: News is recommended to you based on the subjects or themes you regularly read, and on what others reading similar subjects are reading, and on popularity. It seems to rank first by closest matches to what you read, then broader subject matches, then popular matches among similar users. For example if I regularly read about google, I get recommendations on google news first, then I get recommendations on other tech news(apple, dell, yahoo, microsoft), then I get some science stuff. Then at the bottom of the recommendations are some news articles not tech related but seem to be popular among those who have read similar articles as I (By the way it also seems to weigh in my personalized columns - the columns you are allowed to create beyond the default. Alot of my lower ranked recommendations were related to some of my personalized columns that I don't read as much.)

Greg Linden said...

Hi, RR. I just tried it again. I clicked on two articles on Google "Dell tools up with Google" and "Google unites IM with e-mail".

The recommended articles didn't change. They remain completely unrelated articles, I would guess top stories. Specifically, they are "Cartoons: Crisis deepens", "Royal command falls on deaf ears", and "BHP wheat deal soft bribe, says Cole".

It is odd that your experience differs. I'm not sure what to conclude, but it certainly doesn't work the same way as Findory.

rr said...

I found that the recommended page have to be reloaded. When i hit the browser back button, it does not change, only when I reload the recommended page.

By the way, I don't think it works exactly like Findory either, but it does recommend new things to me after I've read an article. Well, it's probably more accurate to say each time the recommended page is reloaded, it recommends something new other than what I've read(I notice that sometimes related lower ranked news are moved up, and sometimes the same story from a different source is recommended).

All the top recommendations (like top 5) are all related to what I usually search and read on. The bottom recommendations seem to be either related to my personalized columns, and or popular stories.

If it does not change for you, I don't know why. I use Google news alot, maybe that makes a difference.

Greg Linden said...

Thanks, RR. I was hitting reload. That's not the problem.

It could be that Google News needs a lot of data to personalize, as you suggest. I had only read five articles about Google. However, I just read another ten article about Google and the results still didn't change significantly.

Another possible explanation might be that your personalization is mainly based on explicit customizations of the Google News page. You said that you added a lot of personalized columns (blocks of news for searches like "Google", I would assume) to your page. Perhaps these are responsible for the accuracy of your recommended articles, not your clickstream history.

I suspect the method Google is using is to cluster Google News users, match individual users to a cluster, then recommend popular articles for that cluster. In my experience, that method would produce results like what I'm seeing, overly broad, poorly targeted.

They may be supplementing this with articles from any customizations you have added to the Google News page. If true, that might explain why your results seem better.

Josh said...

I really like Findory, but have one suggestion. Sometimes I get recommendations that I really don't like. I would like to see a button for "not interested" for a news story.

Keep up the good work!

rr said...

Ok, I don't want to keep beating this, but I find it interesting.

The thing is I only have one personalized column for "google", and the other columns are not related (Jamaica, Christianity). So that does not explain why I get other tech recommendations. It has to also be because I usually read the "google", and other tech news on Google news the most. (I got a few recommendations on Jamaica, I occassionally read from that column. I never get recommendations on Christianity, but I don't read articles from that section because the news that comes up is not what I was really looking for in Christian news.)

Also, something interesting today. yesterday I clicked on a health article that was at the bottom of my recommendations(that story about the study that casts doubts on the benefit of low fat diets). Now today, these are my recommendations - "email firms charge for first class deliveries", "Africa scrambles to limit bird flu", "french face transplant woman", "no, no nano", study on alzheimer. As you can see more health related stuff came to the top of the reommendations.
I clicked on the "no, no nano" 4th article recommended; and now,"microsoft anounces pricing for OneCare Live" takes its place.

So, there is something else going on here. I agree with your conclusion, but it also seems to take into account clickstream and search history(remember the personalized columns are really just a search query when clicked on). Maybe it gives more weight to the conclusions you came to, and less to clickstream and search history.

One last thing - I started this out saying that the service seems to be getting better. When I first saw your post about it and tried it, the recommendations were really broad. I started to notice last week that it was increasingly matching my interests, and so I started using it more, and it keeps giving good recommendations(things I'm interested in, but would not normally seek out).

Sorry for going on about this so much. It's just an interesting topic.