How much of an impact will Semantic Search have?
Do we really need a better search for the Internet? And when it exists will anyone choose to use it?
There has been a lot of press lately about challengers to the current keyword search kings: Google Yahoo and Microsoft. There are a variety of different approaches being taken but a lot of the debate has centered around which technology can best return relevant results. Is it the incumbent keyword search the best we’ll ever have? How about the very retro human powered search like ChaCha or Mahalo? Powerset is banking that semantic search engines are the way to go.
At first I was skeptical about semantic search but after meeting with the folks at Powerset during their public unveiling I started coming around. They certainly had a good number of compelling queries that Google and Yahoo don’t handle with any great degree of accuracy. The ability to determine the intent of a query and to do more than word matching is impressive. Ranking based on semantic data sounds very useful. Hey maybe semantic search will be massively superior!
I’ve been thinking about these issues over the last couple of weeks and more cracks keep appearing. Just for fun I went back through my Google History (it has to be good for something, right?) and looked for places where I thought semantic search would be superior. I found that my searches generally took on a couple of forms and yours may differ.
Spell check and definitions
Over 15% of my searches were either a quick way to check the spelling of a word or used as a dictionary lookup. Sometimes it’s stupid things.. like determining if “margarita” has an ‘h’ in it since I knew the pizza-type “margherita” does. I can tell you right now that I’m not going to type “Does margarita have an h in it?” into a search engine when I can make Google’s auto-suggest feature do basic spell check for me. Sorry Powerset. I also know that Google has a handy “define” feature that saves me the time of loading up a dictionary site or changing the browser’s search box away from Google.
Shortcut to a site
Roughly 30% of my searches are for sites where searching Google beat typing in the actual address for non-bookmarked sites. If I need to visit MSDN it’s easier for me to type MSDN into Google than to type out “.microsoft.com” in the address bar. For many blogs and other sites out there I don’t have to remember if it’s a .net, .com, .us, .whatever. Keyword search to the rescue.
Just get a me a generic pointer to more
Most of the rest of my searches aren’t particularly targeted. Band names, people, products, etc. Typing is “What/Who is” just wouldn’t be helpful to the semantic search engine or myself. Firefox’s ability to highlight text and search Google is the origin for many of these searches. With the addition of Google’s Universal Search I’m often getting the image, video or news that I’m interested in by default.
Math and Unit Conversion
Google is my calculator. I suppose that’s all I need to say about that.
Error messages
I work with a lot of different software apps with varying degrees of stability. I also tend to have to track down a lot of random errors that seem to pop up when I’m learing new frameworks like Django or trying something new in RoR. Searching for error messages is unfortunately a habit of mine. This may be an area where semantic search can work quite well when looking to solve certain problems but understanding the meaning behind code seems to be an impossibly difficult task.
Facts, Recommendations, Recent results, etc.
I have identified only 36 searches out of just over 1800 that I believe have a high chance of being significantly impacted by semantic search. In some instances I’d like to find about what’s new on a particular topic over the last week. Unfortunately Google’s options only lets me limit results to things that are less than 3 months old. Most of these other queries were about fact finding: Who played X in a movie? What is Y’s latest album? When is Z going on tour? How many users does Facebook have? What is the best free screen capture app for the Mac? There were a few more where I was intentionally iterating to remove irrelevant results. Here semantic search could really shine! It’s unfortunate that all but two of these searches happened to be for programming related items. I just don’t search for specific facts all that often or I have different non-public channels where I look for advice (IM, Facebook, etc).
Conclusions
I immediately saw the benefit for about 36 searches, that’s roughly a search per day. 2% of my searches. Not very encouraging. Even if the number was 10 times higher I’m still not certain I’d want to take the time to change my searching behavior. This is especially true if using Powerset involves more typing (for better results) and slower response times (for now). This just means that Powerset won’t be a drop in replacement for my current search engine of choice. I’m not sure they want to be. I’m not sure Powerset knows what they want to be. Answering trivia questions or being a better search engine for Wikipedia may be their calling.
How do your search results compare? Do you see a lot of opportunity for semantic search to make a difference in your everyday searches?
PS: I found it quite difficult to remember the intent of my searches from over a month ago. I would have tried to use a large sample otherwise.
Comments
3 Responses to “How much of an impact will Semantic Search have?”
Leave a Reply
But perhaps you are only attempting those searches that you know will work within the limits of the current engines. Pointless typing “best chinese takeout in downtown” or vast numbes of similar queries into any current engine.
I agree. For those types of searches I tend to hit up specialized sites for answers now but that could change. I just don’t think that it makes up a significant portion of my search needs.
I see that you are interested in semantic search
You should take a look at http://www.linguisticagents.com. It’s a start-up company that has developed a natural language understanding technology that will be used in many applications in addition to search. Let me know what you think - techshrek@gmail.com.