Busting A Buzzword: Semantic Search

If you’re one of those billion or so people who thinks that Google “knows” what you’re searching for, this may be of interest to you.  Let’s take a search such as “local florist purple lady slippers.”  I might enter that search because I want to buy some Lady Slippers (a rare flower.)  My top results show local florists (none of which mention Lady Slippers but pay a lot for Google Ad Words) and two obscure references to US Forest service website about how to grow these rare flowers, neither of which mention florist.

What Do You Mean?

Apart from the fact that the Google algorithm is heavily influenced by advertising dollars, it’s that some of the words in the search are extremely rare while others are very common, so the model returns documents that have a lot of the common phrase, even if they entirely lack the rare words.   Wouldn’t it be better if I could somehow say, “Hey, regardless of whether there are a thousand mentions of the first topic, I want to see documents that have both topics mentioned and that would have a better chance of finding a florist that has these flowers?  Imagine how different a search result would be if you could control which of these topics are important, to what extent and whether they are required or optional.  The results would change dramatically.

Remember when you’d enter a keyword and Google would simply return whatever it “thought” you meant? Of course, this went just about as well as trying to interpret what you “thought” your partner said when you lost connection just before you walked into the grocery store.  Or what you “thought:” a date meant by “I’ll call you next week.” You can easily misinterpret the results and are left tweaking your phrasing or Boolean search to every extent to try to explain what you really meant to the search engine and get the right results.

Today, Google operates on a mix of semantic search, user history, loosely associated adwords and other information they’ve secretly gathered about you so they can make some assumptions about what you mean and try to assess intent. Of course, we use Google for search a lot more often than, say, our ATS.

Semantic – The “Secret Sauce”

If you’ve been in the recruiting industry for awhile now, you’ve heard of semantic search. It’s the “super secret sauce” behind job boards and used to be all the rage. The first time I really remember hearing about it was when I worked at Monster.com, and they released their 6sense product ( ). The whole deal was that it simplified search. The intent, of course, was simply to improve accuracy by trying to make sense of what you really mean when you type one thing or another.

But there’s a catch, and it’s that semantic search isn’t programmable by you – a fundamental flaw in the whole idea, if you ask me. Semantic search builds on a library of shallow word associations, formal grammar rules, and inputs from technicians with no recruiting experience.  Those techies tried to build this complex matrix of assumptions that keyword A= keywords B, C, and D, sanitizing all nuance in the process, using the lowest common denominators and resulting in a lot of false positives that often make it harder to find that needle in the haystack. Business intelligence regularly departs from standard use terms, as does a vast array of other topics, such as sports, art, street jargon and a world of colloquialisms.

Java or Java?

The reality is that it’s really hard to cover every instance. For example, a semantic search might know that Java isn’t coffee, but it might not know that Java is also a small town in Georgia. Inevitably a keyword match would bring back all the Java – not some of the Java PLUS the related terms. Since you are now thinking about coffee, grab that morning cup of joe. I’ll wait.

Princeton University’s has a search capability that allows you to query the standard semantic library. Are you surprised that when you search for “Programmer” there’s no synet for “Developer?”   Google “knows” a lot less than you may think.   

Now that you’re caffeinated, you’re thinking “Wait a minute.  Smart search isn’t so smart then, is it?”   You probably knew that already if you’ve done a search lately.   Of course, that doesn’t stop your job board rep from coming in and selling it like it’s the best thing since Google. The more you know, the more you can debunk this snake oil sales tactic.  At the end of the day, you can’t control the results, and you can’t evaluate them because you just don’t know why you got such results. 

What we really need, and I only know one company that does this (shout out to TalentBrowser, powered by DataScava , and founders Janet Dwyer and John Harney) is a completely customizable white box  “profile” search built on input and personalized rules that you the user control, not a black box semantic search engine that thinks it knows what you “really mean.”  Profile search allows you to specify many individual topics in a search, with thresholds (minimums) to be met by each topic. This twofold process bubbles the best candidates right to the top.

So this year, when the sales people come stalking – and we all know they will – don’t fall for the same ol’ semantic search “innovation” sales tactics. Don’t let the buzzword bingo fool you. Semantic is the same old shit, different job board.  All semantic tools use the same shallow library of terms. At best, it’s a rookie recruiter.

By Katrina Kibben

RecruitingDaily contributing writer and editor.  I am a storyteller. A tactical problem solver. A curious mind. A data nerd. With that unique filter, I work to craft messages that strategically improve the perceptions and experiences of our clients, the people they employ and the candidates they wish to attract. I methodically review and collect research and insights to offer solution-based recommendations that meet the one-off, and not so one-off, recruiting and employer branding problems of today's global employers.