Wednesday, November 05, 2008

Resources versus Answers – Asking a Question of Search

{{fr}} La tour Eiffel vue depuis le Champ-de-Mars.Image via WikipediaSearch is very broad in meaning and it is easy to lose sight that search actually consists of two distinct sub-sets of queries. Both sub-sets aim to find something; one is looking for resource and another for an answer. At this time we use the same approach – keywords matched in a document that is ranked for relevancy via some method (human and/or algorithm) – across both of these sub-sets of queries. This works somewhat but we are rapidly approaching the limit of effectiveness for this approach. This limit is Marissa Mayers 80/20 problem of search.

The first sub-set is finding resources (e.g. documents). The current keyword and ranking method works well for this type of query. This is what has fuelled Google’s growth. Keyword and ranking when a user is looking for one or more resources on a topic such as blog posts talking about an election. Where it falls down is answering specific queries such as “How old is the Eiffel Tower?” The user in this case is looking for a fact. Users have gotten around this problem by using the returned resources from a search as the basis to find the answer they are looking for, a human adaptation to a systemic problem.

Finding answers is the second sub-set. While we currently rely on keywords and ranking to navigate to an answer it is cumbersome and not effective. Instead the paradigm of keywords and ranking needs to be tossed out. Finding answers works better with a question and fact. A question (as opposed to queries) allows the system to quantify what fact is being asked about. For example the question “How old is the Eiffel Tower?” focuses the particular answer to be found to the age instead of potentially the location, who built it, what it is made of etc.

Using the question and fact paradigm to find answers creates new approaches to using web services and usefulness of the web to everyday life. This isn’t to say that question and fact will replace keyword and ranking rather it is complimentary and produces better results for a sub-set of search.

Consider the example of finding flights for a holiday. Using keyword and ranking the user would type in something along the lines of “flights cheap [destination]”. The engine would then return a series of web sites that match those keywords. The user then navigates to those pages and then drills through the pages to find the answer to their question. If, however, question and fact is used the user would type in “What is the cheapest flight to [destination] leaving on the 21st of December?” The web then returns the fact that flight y priced at x leaving at 10 am on the 21st is the cheapest flight. How much quicker and easier is that to understand?

For many people the web and search are still too difficult to use. But they know how to ask a question and this opens up the utility of search and the web to a whole range of users that are intimidated by it. It is worth repeating that question and fact will not replace keyword and ranking. There are queries with which question and fact doesn’t work for just as there are queries for which keyword and ranking doesn’t work for. They are complimentary.

Question and fact does have the potential to boost the growth of paid search results. The boost arises from question and fact providing a better signal of the user’s intention and so improves the targeting of advertising that better answers the query. For example a user asks the question “what is the cheapest holiday for a 16 year old girl in Mexico?” it a very reasonable to assume the intent is to find a holiday for a 16 year old girl in Mexico. A keyword and ranking would produce results about holiday’s in Mexico without any knowledge of whom or why he is searching although an assumption could be made that the person is looking for themselves. Interestingly, through in demographics and/or behavioural data and the system will produce completely the wrong answer. Say for example the person is 52 year old male in which case the system is likely to return Mexico holidays for a 52 year old man when his intention was to find a holiday for his 16 year old daughter.

Question and fact will go a long way to addressing the 20% of search remaining. Many web services implement crude methods for asking a question, ones that are frankly laborious and time consuming to use. The key to unlocking the power of question and fact is to make it as easy as possible to ask the questions. The pitfall to implementing question and fact is knowing when to use it. Question and fact works when the question can be answered by a fact e.g. “How old is the Eiffel Tower?” It doesn’t work when the answer is not a fact e.g. “What is the best holiday in Mexico?”

Reblog this post [with Zemanta]