Emetrics Summit Synthesis - Part 1 - Site Search Query Analysis

Posted by Marshall on May 13, 2007 | Link It

This is my first post to synthesize the information I listened to or read about at Emetrics Summit, earlier this week in San Francisco:

I sat in on Alex Langshur's session - Measuring ROI when you don't sell on May 7th, and noticed a lot of the analysis was based on datamining site search query logs (which I've done at IBM and for my own clients, outside IBM). The presentation is not online yet but there's other case studies on Langshur's site.

For example - Analyzing Intent (aka alignment of intent) is Search Analytics where you can compare keywords by category over time to see what went up / down, as well as seasonal variations.  

However, Rich Wiggins and Lou Rosenfeld presented Listening to the Search Box: Search Analytics in a Nutshell, pretty much at the same time Langshur was presenting (in another suite). and the content of both presentations overlapped, judging from the copies I have in the Emetrics Presentation Handbook.  Rosenfeld noted, and I have noted the same in my own experience, that:

    • ~45% of search referral (from Search Engines) are variants of a company's name.  This is certainly true of Branded businesses, like the Architects I have dealt with, including the one Architect client I still have, www.Mascord.com.
    • There's a lot of Long Tail data in local search queries

Rosenfeld suggested that you data-mine search query logs and map much of your long tail data to popular pages in your site that you think go with the query.  The problem with that … you need special tools to do it as the massive amount of long tail data makes doing this manually way too tedious for humans to do (and be sane for long) - I'm talking about massive amounts of queries where there are only one or two searches internally on that phase - and it's unique - meaning only one person executed that search.

Both presentations mirror what I heard at Search Engine Strategies last month, in NYC, where I spoke with someone who data-mines Search Query Logs to determine intent of Searchers and if the content of a site matches what people are searching for - and if not, changing or improving content in specific ways to increase customer satisfaction.

As Rosenfeld pointed out - there's often little time and few useful tools to parse Search Query Logs in order to generate reports - and that the work of analysis is both hard and often boring (because the right tools have not been developed - I firmly believe that).

My takeaway from both sessions (even though I only attended one of them - I have the material from both and familiarity with the subject) is that Site Search query logs don't get the attention they deserve in any of the organizations I have worked with and most companies don't realistically know what to do with the information - even if they were able to data-mine it, segment it, and match up the relevant pages.

I'd even go so far as to suggest vendor solutions that learn and match up the right pages (or create content on the fly out of XML data) over trying to manually determine what the right content on your site is for a query that was unique and executed one time for someone's specific need at the Long Tail.

In other words, for a large site, I think you'd need an automated tool with some human oversight - but doing it manually is almost unthinkable - based on the sheer volume of data that needs to be analyzed and remapped to the right pages.

 



2 Responses

These are the current comments for "Emetrics Summit Synthesis - Part 1 - Site Search Query Analysis"

05/14/07 @ 7:40 am

Hi Marshall; thanks for mentioning our talk (and sorry you couldn’t make it). I think you may have misinterpreted the slides though: we’re suggesting creating best bet search results (i.e., manually matching quality results with search queries) for only short head or highly common queries, not long tail queries, which, I agree, would be a huge amount of work.

As far as long tail analysis goes, we are talking with more people who are doing it, as they feel like they’ve already derived all the benefit they can get from analyzing common queries. The way to do this is to look for patterns or clusters in the long tail through random sampling, which, though not ideal, is a bit more manageable.

The good news is that the tools are getting better. We’ve made a couple simple and free tools available on our “book in progress” web site (our book on on-site search analytics will be out later this year).

Thanks again; hope to meet you next time.

cheers



05/14/07 @ 7:49 am

PS If anyone’s interested, here’s a very similar presentation to the one Rich Wiggins and I gave at Emetrics, viewable through SlideShare (which is pretty cool in and of itself)>



Post a Response

Name (required)

Email (required, not published)

Website (optional)

Note: The following tags are approved for comments on this blog:
<a href=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <del> <strong>





Subscribe

RSS Subscribe View my FriendFeed Current Subscribers