LinkedIn Basic Search is Semantic – LIR Search is NOT

booleanstringsBoolean, LIR 6 Comments

Black-Box-LinkedIn

 

In the previous post Discrepancies in Search: LinkedIn Recruiter vs. Personal we looked at some differences in the search results. While we have no word from @LinkedIn, my guess is that the Basic search does additional “lightly semantic” interpretation of  the search queries, which leads to those differences.

(1) THE REASON BEING THE INDUSTRY

In the first example with “computer games”, it matters that it’s the exact name of an industry. LinkedIn personal account adds the profiles that have that industry to the search results.  That’s a guess backed up by a good number of tests. I find this “light semantic” addition to be quite relevant. As an example, let’s narrow the example search to current company = Apple. People who work at Apple do not necessarily work with computer games; those who do may express that as their industry. If we look for people at Apple with C++ 3D iOS “computer games”, the personal account finds 8 profiles, while LIR finds only 3.

Thanks to Glen Cathey for his comments on my last post and his new blog post with some exploration of what’s going on. (Glen: I just tried to see what the difference might be between a word used in the industry name and as a keyword otherwise – and ran across a search that I don’t even know how to begin to explain! A search for software NOT games/industry: “computer games” finds 28 people in the US – it will find more people in other countries! – and 4,623 results in LIR. Of course, that search itself doesn’t make much sense other than a test search.)

(2) THE REASON BEING OTHER THAN THE INDUSTRY

Responding to Glen’s request to provide examples of searches that do not include searches for industries: it is not hard to find, once you know that, generally, the personal/Basic search is “lightly semantic”.

EXAMPLE ONE. Search for VP Recruiting, Bank of America: personal – 22 results, LIR… 6 results. Guess what, personal account knows that VP is Vice President, LIR doesn’t! Here’s a variation: “VP Recruiting”, current or past, BofA: personal: 9 results. LIR: 2 results.

EXAMPLE TWO. Search for Sr. Manager at Deloitte in the US; personal: 60 results. LIR: 3 results. Personal knows that Sr means Senior; LIR doesn’t.

EXAMPLE THREE. Search for Morgan, Senior Project Manager, NYC Area  9 results, while LIR provides a whooping 358 results. I didn’t find the time for a better example and wanted to point what is going on: the personal decides to use Morgan as the first or the last name only, while LIR finds past and present employees of Morgan Stanley and J.P. Morgan.  In this case (the personal names interpretation), I’d rather the semantic interpretation didn’t happen.

There are other examples of discrepancies I have run into, that I still can’t explain.

Needless to say, LinkedIn Recruiter subscribers don’t expect the differences in the search results to be so significant, in some cases seeing way fewer results in LIR, in other cases seeing many more results, yet in some – roughly the same numbers (as it used to be before Galene). That provides for poor UX, to say the least.

Once again, I seriously recommend searching “on both sides” and perhaps X-ray as well.

 

Comments 6

  1. Ok thanks Irina/Glen. So, inexplicable. That does indeed provide a poor UX. It means you have to do many more variations of a search and still not guarantee to find all results. It renders the advice I’ve been giving LinkedIn users about how to get found quickly redundant. It’s really poor for LIR users who want an all-encompassing service and there will be a backlash. It just looks like LinkedIn’s broken.

  2. This is the answer I was waiting for. I knew Industry would throw off results, but the other factors that come into play between free and LIR regarding logic and order are not defined.

    The major advantage or LIR is the ability to use facets like Industry, Groups, and Years in current company as well as dedicated first, last name, and company filters. If you are using keyword for all things you are using LIR incorrectly.

    My working theory = Marketing Ploy
    vsearch gives you more results pulling from broader criteria. If the results exceed 100, free users will feel like they are missing out on a hidden gem just beyond their grasp. LIR solves this problem with unlimited results and better filters.

    Creating a problem that only they can solve. Sounds like something a marketing person would dream up.

  3. Irina,
    Some people may be surprised to learn that LinkedIn’s free search has been “lightly semantic” for years.

    I recall from some of the earliest Talent Connect events that they mentioned they programmed the search engine to try and recognize first and last names from the keyword search field/main search bar, as well as company names.

    What you might find interesting is that LinkedIn now knows that “V.P.” is also VP and Vice President, but it didn’t always “know” that. I’ve used that as an example in the past during my presentations at LinkedIn events: “V.P.” -VP would return results of V.P. and not VP or Vice President. Apparently they paid attention, because now that search produces 0 results because it “sees” them as the same thing, but that was not always the case.

    I am also fairly confident I can claim credit for the Morgan = JPMorgan semantics. I’ve used JPMC as a classic example during nearly all of my LinkedIn Talent Connect presentations, demonstrating the need to search for the many ways a person can say that they’ve worked for that financial institution on their LinkedIn profile. If you search for JPMC in the keyword field, you will now notice it will also pull JPMorgan Chase – see here: http://linkd.in/1sFZQhN Again, they’ve been paying attention, which I think it a good thing.

    However, their “light semantic search” is ultralight. While LinkedIn has “known” that Sr = Senior for a couple of years now, it also knows Jr is Junior (which I believe is newer), and while there are other examples of semantic search at work, you won’t find too many more.

    For example, it unfortunately doesn’t “know” that Mgr = manager, and there are over 500,000 people who mention Mgr in their title.

  4. Also, for anyone who searches LinkedIn (free or Recruiter), I believe it’s a best practice to use the specific search fields (e.g., company, title, etc.) rather than searching for those kinds of terms in the keyword field – that’s what the specific fields are for. Why leave terms open to uncertain interpretation? You know what you’re looking for, so use the search interface/engine to find it…don’t leave it up to chance by using only the keyword search field, and as I always recommend, use maximum inclusion. Unless you can show me the entire taxonomy, I will simply use my own massively inclusive search which will be much more effective than relying on black box semantic search.

    Irina – as for your software NOT games/industry: “computer games” search example, while I find the wildly varying search results interesting, especially given that LIR returns so many more, I have never been a fan of using industry filters for many reasons (e.g., what about all of the people with experience developing computer games that chose a different industry?), so I’d call this a “corner case.”

    1. Post
      Author

      Glen:

      I appreciate your comments!. However, to have a dialog, we need to talk about the same subject.

      Let’s say, not everyone searches like Glen Cathey. 🙂 It could be, that if people did a series of thorough Boolean searches, using various search facets and all the possible ways to spell out synonyms, the results would be similar on both sides. It’s NOT what my posts are about though.

      My goal is to point to the big discrepancies in the system. I created the search examples with that in mind. Sometimes, my examples are about “corner” cases as well, and that is also on purpose. My interest is on the searches that do *not* work as expected.

      I am trying to create a general awareness about the discrepancies between LIR and personal search results. I have also identified the potential reason, which is – “light semantic” interpretation of the queries on the personal side. This is still a hypothesis though, and it is of interest whether this explains all the differences. I am not sure.

      If the underlying “query pre-processing” is known (as in the case of industries), it is not hard to find workarounds to defeat the interpretations.

      I hope I am making sense. There are still other search cases that I can’t explain; will share later.

      -Irina

  5. Thanks for posting this,

    I enjoy researching “why” and “why not” particularly related to search research.

    It reminds me of the processes of “Google-whacking” starting back 12 & 1/2 years ago and “false positives.” I have heard stories on applications known informally as “cleaners” or “cleaner girls” that modify search results, but I tried and could not find anything specific about who, what, when, and why and gave up this research.

    This may be more like / as the “deep web” and the “dark web” that few search research address publicly for their own reasons. Regardless, it helps me think through the search process – and – that’s a good thing.

    All the Best!

    Ray Towle, SPHR
    # # #

Leave a Reply

Your email address will not be published. Required fields are marked *