Sprixi – useful image search?

It’s trite, but it’s true – the best products to build are those that people want.  Sage words from Paul Graham, yet I’m always surprised at how many products out there are solutions looking for problems.  That’s why I was particularly pleased to see my good friend Andrew Goldstiver’s new startup, Sprixi — a site for finding useful images.

But what are “useful” images?  First let’s consider the state of image search currently.  Firstly, good image search matters a lot to people, as made clear by the positive reception of Bing image search after years of Google image search languishing.  But what are the common use cases for image search?  Trying to show someone something you don’t have a picture of (“Here Steve, this is what a Huntsman spider looks like in Australia”)?  Trying to confirm what something looks like (“Ahh, so that’s what a Darwin stubby is”)?  Looking for something to use for a blog post or an assignment (“Where do I find a picture of a periodontal probe“)?

It’s the last use case that’s always been frustrating.  While writing my thesis I remember emailing dozens of people for permission to use their images.  Then there’s the size issues – most images online are thumbnails, with few print quality images – without using filters these are difficult to surface.  Finally, there’s relevance — finding the best image can be a chore.

So how to solve that?  Sprixi does this a few different ways:

1. Great interface. Much slicker to use than other image search engines.
2. Fair-use image crawling.  Sprixi aggregates all the best sources of images which can be re-used.
3. User contributed content.  Users can upload their own images.
4. Crowd-sourced relevance.  While images have a built-in relevance, Sprixi allows users to score photos so that the most representative photo floats to the top of the list.
5. Web 3.0 – like many new sites, Sprixi is part of the semantic web, by having an understanding of different concepts (also see Adioso and Trovix for other examples).

Sprixi learning a new topic

Sprixi learning a new topic

Rating an image

Rating an image

Using an image - notice the "choose" link on the left

Using an image - notice the "choose" link on the left

As an example, let’s say I’m doing a presentation on user research, perhaps the ethnographic dental studies I completed for my thesis.  If I need a generic picture of a dentist to illustrate a slide, I simply search for “dentist” on Sprixi.  There’s an existing topic with sorted and unsorted images. I can then browse through the photos, rate them myself, and when I find one I like, I choose the size and click the “use” button (which is some very innovative use of horizontal real-estate for those of us on small screens), giving me the option to download or embed a link.  If Sprixi doesn’t understand the topic you’re search for it will create a pool of photos based on aggregation on the fly.

As a later engaged member of the community, I could also upload a copy of any photos of dentists I took as part of my studies for other people to use.  I could then populate a new concept, such as “participatory design” with some shots of researchers interacting with practitioners.  Now suddenly there is a site which has useful pictures of the participatory design process.  Trying this search on Google, Bing and Flickr revealed reasonable results, but with lots of cruft.  Sprixi wants to get rid of the cruft.  According to Andrew, it’s not explicitly necessary to rate images manually either, as Sprixi does some implicit rating through site interaction.

I was a little disappointed that Sprixi didn’t pick up on more obscure topics (e.g. periodontal probes) with its aggregation, but I’m sure this will continue to improve.   Managing contributions (and the sign-up process in general) seems to be a work in progress.

I do wonder how big the user base for a product like this will be — however there are an awful lot of bloggers, uni students and Powerpoint presentation creators out there.  I am still interested to see how Sprixi plans to monetize and grow beyond the few sites its crawling at the moment.  If it gets traction though it could easily become a fantastic repository for free-use images.

Give Sprixi a try and let Andrew know what you think.

Google continues to disappoint in search quality

At first I thought it was just me who noticed how poor Google has been lately, but happily some folks from Hacker News confirmed I wasn’t going crazy.  Today I had an odd query which got me suitably terrible results.

I was writing an email and wrote as part of my sentence “based on recency”, and Outlook (really Word) underlined it to indicate a spelling error.  I felt sure that “recency” was a real word (and as I type this, Firefox keeps insisting it does not exist either), so I ran a Google search for [recency microsoft word misspelled].  Note that I put Microsoft in there to ensure I would get back results that talked about Microsoft Word and not the word recency in general.  Anyway, the first result I got back was this:

Google search for recency

So not only has it changed my search from recency to Regency, without actually asking me, it also changed the stemming from misspelled to misspellings.  I wondered if it would also start dropping terms from my search which drives me crazy, and was one of the original reasons I switched from Altavista to Google.  Sure enough, results 6 and 8 were both missing “recency”, the thing I was most interested about.

Google's other problems with "recency"

Yes Google, you do make it easier for people who aren’t savvy, but can you at least give me an account flag to not have you mess with my results?  I had at least 7 fantastic years, and my more mundane queries still work fine, but more and more I find myself frustrated.  One last example, which I’ve come across a few times.  I heard something about a campaign at The University of Queensland called “Bring back the Red Room”, which was the university bar while I was studying there.  Wondering if it’d been shut down and what the story was, I ran an appropriate search:

Red Room Google Search

Why is the first link something that doesn’t even mention the Red Room?  Neither link provides any useful information to the query at hand.  Here’s where it got interesting – when I was typing a new query, [red room], it automatically suggested [red room st lucia].  Clicking on this returned this:

Red Room St Lucia search

…answering the question about what happened while still in the search results.  So why such a difference in results?

The problem is essentially that Google is becoming a tool for people to find destination sites, rather than information.  I am guessing that the majority of users of Google actually do find this useful, but it is alienating a core component of the Google faithful, creating a potential opening in the search engine market.  I look forward to seeing what might come of it.

Will Craigslist’s dominance continue?

One thing has always puzzled me — is Craigslist successful because of its simplicity?  Or is it because it was first?

I’ve heard the “you can’t be successful unless you’re first” line bandied about a few times, but have never truly believed it.  It was a reason given why not to create a social networking component on Trovix, which I argued against with the example of Facebook versus Myspace, or Google versus Yahoo.  Based on examples with sites that big (and examples of smaller companies continuing to thrive in a niche, such as Vimeo versus YouTube) I’ve always believed that a superior product and/or a compelling value proposition are key ingredients to a product’s success, and while some inertia can be gained from being first (particularly with network effects), it is not insurmountable with appropriate marketing (which is why Trovix now has a social networking component).

But what surprised me today was seeing the success of eBay’s kijiji in Canada.  A hat tip to kitcar for the link.

Kijiji versus Craigslist in Canda

Red line is kijiji and blue line is Craigslist.

In Canada, Craigslist was slow to take off.  As such eBay put serious effort into marketing to gain that all important network effect, and now Craigslist is slowly becoming irrelevant.  The most amazing part is seeing the actual moment the network effect takes hold in late 2007.  If Craigslist was like other companies, they would do what eBay in Australia did.  Back in the late 90s, eBay was king in the US.  However in Australia, some local imitators jumped in, including sold.com.au (which now redirects to ebay).  All of these were handily beating eBay, until a well-coordinated marketing campaign put eBay back in front.  They weren’t as effective in New Zealand though:

Trademe.co.nz versus Ebay.co.nz

Blue line is trademe.co.nz, red line is eBay.co.nz (insufficient traffic to appear on the graph).

The final thing worth noting is that kijiji does in fact have a foothold in the US.  If growth continues (currently a healthy 10% per month compared to 4% for Craigslist) and Craigslist remains stagnant, it’s hard to imagine, but they could eventually win out:

Craigslist versus Kijiji

Red line is craigslist and blue line is kijiji.

There’s a very good example of this sort of slow growth playing out right now with Gmail and the other webmail providers.  It’s slow, but it’s seemingly inevitable that Gmail will be the number one email provider:

Gmail is the orange line, bright blue is AOL and dark blue is Hotmail.

So – back to the original question.  Is it better to have a feature rich site or a more usable but limited site?  There’s unfortunately no easy answer — it all comes down to whether users who want to use your site can get done what they want to get done.  Craigslist has been “good enough” for a long time, and inertia continues to carry them.  However given enough incentive (a site with better features and just enough of an audience to make it worth your while) users can and will swap.  It’ll be interesting to see how kijiji continues to fare in coming years.

Information overload

One important usability principle is to manage complexity for the user.  This means both the UI itself, and contraining the data to prevent overload.  Brandon Walkin wrote a nice article about managing complex UIs, but didn’t talk about managing complex data.

Usually Google does a great job in this field.  Their onebox concept allows them to float up what they think will be most relevant to you in a small snippet.  If they want to give you extra results, they just give you a few and clearly delineate the rest of your results.  Google Finance does a really nice job of different zoom levels of data:

Google Finance zoomed in

and when zoomed out:

Google Finance zoomed out

It’s hard to tell the difference — which is deliberate.  Google put in some very nice smoothing algorithms as well as doing a decent job on adjusting the scale and news for the stock.

One thing that has surprised me is the increasing complexity to Google Maps.  I’m not sure why they keep adding unverified user submitted datas and non-relevant photos, but they do.  Now they provide so much information it’s essentially unusable:

Google Maps

I’ve also noticed lately that if I search for a particular destination, I usually get a large cluster of destination points, as well as an arrow actually pointing to my location (example below is of Massachusetts State House).

Massachusetts State House

The top x data points is usually a good rule of thumb (depending on the size of the display between 10 and 25), or just show 1 if there is a high degree of confidence.  I think Yelp does a pretty nice job with their map – giving you the option of retaining the original data, or updating the top 10, which keeps it nicely manageable.

Google doesn’t stop iterating

One thing you can say about Google is its search results page has been known for its consistency and simplicity.  The most radical addition was the poorly-received (by bloggers anyway) search wiki.  Most changes have been behind the scenes — there has been a lot of additional complexity being continually added to the html source over the years, and plenty of tweaks with how the results are presented, their ordering, and the OneBox.  Google had their unbranded playground, SearchMash, which recently was decommissioned, and they also bucket-tested very select features, but very little actually changed on Google.com.  The site is certainly different to how it was 5 years ago, but it has been so gradual, it’s like watching someone you know age.  If you look at a photo you can see the difference, but otherwise they look the same.

I’m not sure if it’s a direct response to the Microsoft/Yahoo partnership, but in the past few weeks Google appears to be taking broad steps towards changes that now benefit them rather than the consumer, while also pushing out new features.  There was the recently revealed Caffeine, creating a strong connection between new features and the Google brand.  They recently changed the landing page for Google.com, when signed into a Google account so that it defaults to iGoogle rather than the vanilla search page (giving me yet another location to have an unattended chat window open).  The latter is most definitely an attempt by Google to keep users within their sites and expose them to new products (currently Google Latitude and Calendar are featured prominently on mine, even though I had previously customised iGoogle).  Now Google have rolled out some tweaks to their search result page (which I am pretty sure are not being bucket-tested, as I’m sure I saw them tested a couple of years (!!) ago):

Click for a larger view

Click for a larger view

On the left in the red box is something I’m very surprised to see from Google and I was just thinking about the other week.  Filters!  Filters are a surprisingly difficult usability problem, and something I’ve spent a lot of time on.  I’m pleased to see Google has put them on the left (something I believe is more natural to the user), and it’s also good to see they start minimized, but are still noticeable enough so that when you need them they are there.  In this example, I was keen to see more timely details about the AT&T iPhone tethering, and was about to adjust my query when I noticed the filter box and was able to simply ask for results in the last week.

The green box on the right is a relatively subtle change.  My guess is that given the prevalence of wide-screen monitors these days, the clicks on the ads were trending downwards (or perhaps their gaze-tracking machine noticed less eyeballs).  As such they have brought the ads to come sit right next to your results.

However, the fact is that neither of these would be that interesting on any other site.  My original title for this entry was “Google doesn’t stop innovating”, but Google has a tried and true formula, and I think it’s far more accurate to say that they are relentless iterators (at least when it comes to the UI).  Now, if only they would implement the custom re-ordering of the navigation bar based on usage that I suggested to them…

The Bay Area response to foreclosure

I spotted this article a little while ago, discussing how one man responded to having his house foreclosed.

The front yard of Williams’ home is strewn with boxes, furniture and trash cans. There’s even some of the home’s air conditioning duct work lying on the lawn. That’s not the only part of the property left in shambles. The inside of the house is just as messy.

And just in case anyone needed an explanation for his actions, Williams also allegedly left the bank a big note –– using spray paint. The words painted on a wall near the front door are hard to make out but it appears to declare: “Brought to you by Deutsche Bank… Eat it.”

My favourite part though was the response from Burbed.com (who I’ve declared my amusement of here, here, and here):

Makes sense. Milpitas isn’t in the Real Bay Area. Full of savages. I hear they’re starting to eat babies because house prices are so low. 


Microsoft Live Search – trying to increase their referrer percentage artificially?

I’ve noticed a lot of my traffic supposedly is coming from Microsoft Live.  However, looking at the referring URLs, something fishy is obviously going on:

  01 May, Fri, 09:10:38    http://search.live.com/results.aspx?q=authentication  
  01 May, Fri, 09:19:00    http://search.live.com/results.aspx?q=virgin  
  01 May, Fri, 10:51:20    http://search.live.com/results.aspx?q=usability  
  02 May, Sat, 15:55:08    http://search.live.com/results.aspx?q=since  
  02 May, Sat, 17:43:10    http://search.live.com/results.aspx?q=about  
  03 May, Sun, 11:49:51    http://search.live.com/results.aspx?q=issues  

I can only conclude they’re trying to increase their referrer percentages to try and increase ad spending.  I’d be interested to know if anyone else has noticed this, or can perhaps explain it.

Google doesn’t trust us with our search queries anymore

A trend I’ve noticed lately is that Google selectively ignores and “reimagines” my search terms.

(actually, it also habitually drops my search entirely. There are bugs in both the new Google toolbar, and a bug in iGoogle that goes back at least a year, where my first search takes me back to a blank Google page in any version of Firefox)

For example, if I am looking for a verb in particular, it will often change its tense.  For example, searching for “carmack suggest” has the first result matching “carmack suggested”.

Another example is suggesting all sorts of “alternatives” in your main page of results.  Check out the results for “first person“.

3 results for “first person”, followed by three results for “third person”, then a suggestion for “first person narrative”, followed by the rest of the results.

These irritations are manageable.  The search that inspired this post tonight was the search for “irobot lighthouse instructions”.  I put in fresh batteries to my lighthouse, and the light didn’t come on, and there was no obvious ‘on’ switch.  So I had a look online to see if I could find an instruction manual (much more convenient than digging through old papers and/or boxes).

1. iRobot: 500 Series: Virtual Wall® Lighthouse

– 11:11pm

The Virtual Wall Lighthouse uses an invisible infrared signal to help iRobot Roomba® achieve the most efficient and thorough room-to-room cleaning.
store.irobot.com/product/index.jsp?productId=2821601&cp=2842706&parentPage=subcategory – 55k – CachedSimilar pages

2. iRobot: Redesign Root: 500 Series: Virtual Wall® Lighthouse

The Virtual Wall Lighthouse uses an invisible infrared signal to help iRobot Roomba® achieve the most efficient and thorough room-to-room cleaning.
store.irobot.com/product/index.jsp?productId=2821601&cp=3322463.3358508 – 55k – CachedSimilar pages
More results from store.irobot.com »

3. Irobot Lighthouse – Compare Prices, Reviews and Buy at NexTag …

Irobot Lighthouse – 6 results like the Sunbeam iRobot Roomba Virtual Wall Lighthouse, iRobot Roomba 500 Series Virtual Wall Lighthouse – 80201,
www.nextag.com/irobotlighthouse/search-html – 73k – CachedSimilar pages


10. iRobot Roomba 535 Robotic Vacuum with Lighthouse Technology | TV News

iRobot Roomba 535 Robotic Vacuum with Lighthouse Technology – Very great deal! Everything instructions said it would do, it did, recharges fast and
www.gosutrailers.com/2009/03/irobot-roomba-535-robotic-vacuum-with-lighthouse-technology.html – 26k – CachedSimilar pages

Searches related to: irobot lighthouse instructions

irobot roomba 535

irobot looj

irobot roomba 560

irobot roomba 570

irobot vacuum cleaner

irobot blu ray

Tip: These results do not include the word “instructions”. Show results that include “instructions”.

I like that tip at the bottom, where of course you’re not looking, basically saying “we’re ignoring what you asked for”.

If what I’m searching for doesn’t turn up decent results, at least show me that if you can figure out what I mean (in this case I should’ve searched for “manual”).  “irobot blu ray”?  I wondered if I was the only one to be so frustrated with Google lately, so of course I turned to Google Blog Search, and it tried to redirect me to port 9.  What is going on?!  Suspiciously, the first result for “google poor search results” is an article about Wikia.

Between the poor search results, numerous bugs on Gmail (I enjoy sending all my emails -1 minutes ago) and Google Reader (scrolling still doesn’t work properly), the awful search wiki, and never mind the numerous reports recently of Google stealing people’s money, my opinion of Google’s previously stellar product quality is starting to get quite tarnished.

PCMag.com top 20 job sites

PC Magazine listed its top 20 job sites today.  There were three Monster sites on there: Monster (#11), Trovix (#17) and USAJobs (#19, although they don’t appear to be in any particular order).

The bit I noticed of course was:

Trovix‘s free search engine makes the job-search process more personalized. Users input their work experience and qualifications and the site matches results to what info they have given. Trovix also has an innovative feature called Job Map, which allows you to type in your location and see on Google Maps how many jobs are available in your area.

My one bit of production-facing JavaScript ever.  I’m happy it got a mention.