Intended for HR professionals, sourcers and recruiters, this presentation explains and explores the concepts of semantics and semantic search, including the 5 levels of semantic search: Conceptual Search, Contextual Search, Grammatical/Natural Language Search, Inferential Search, and Tagging.
2. “The holy grail of search is to understand what
the user wants. Then you’re not matching
words; you’re matching meaning.”
– Amit Singhal, Google
Can applications ever really know what we’re
looking for?
3. Semantics: The study of meaning, inherent at
the levels of words, phrases, and sentences
Semantic Search: Searching beyond the
literal lexical match and into the meaning of
words, phrases, and sentences
5 Levels
4. 1. Skill words/title association, variants, and
misspellings
Director of business development, business
development director, etc.
JDE, JD Edwards, etc.
10Q = SEC reporting
SAP = ERP
JMPC, JP Morgan, JPMorganChase
5. Level 1 Semantic Search can be achieved through:
1. Man
Application of knowledge of synonymous terms
and the relationships between concepts to search
for variants and related terms
2. Machine
Hierarchical or synonymous taxonomies
Semantic Clustering
6. 2. Contextual
Words have different meaning depending on
where they are specifically mentioned in
resumes
Summary, education, recent work experience…
Education vs. address (Harvard Ave.)
7. Level 2 Semantic Search can be achieved through:
1. Man
Innate understanding of contextual references
Field-based search of parsed resumes/profiles
(most recent title, etc.)
2. Machine
Parsing of resumes and profiles
Automated field-based matching (most recent
exp., etc.)
8. 3. Grammatical, natural language search
Targeting sentence-level meaning with noun/verb
combinations
Sentence-level semantics are much more
powerful, predictive, and flexible than word or
phrase level semantics
9. Searching for an identifying specific noun & verb
combinations allows for the ability to target responsibilities
and capabilities, not just keyword presence!
Examples of noun/verb combinations
"3 full life cycle SAP R/3 implementations"
"Carry out wound (pressure ulcer) assessment, recommend
treatment…"
"SOX compliancy weekly internal auditing"
"Perform investment performance and attribution
analysis"
10. Level 3 Semantic Search can be achieved through:
1. Man
Any search engine that supports fixed or
configurable proximity – the ability to control the
distance between search terms
2. Machine
No solution that I am aware of allows for the
automation of specifically targeting of noun/verb
combinations to isolate sentence-level meaning
11. support* NEAR (CEO or CFO or CTO or CIO
or "C-Level" or chief*)
13. “created access database”~7
* PCRecruiter and some other ATS/CRM solutions use Lucene for text search/retrieval
14. 4. Implied skills, experience and responsibilities
Inferential semantic search is a form of Level 3
Talent Mining (Indirect search)
Inferential search involves specifically searching
for what isn't explicitly mentioned – words and
phrases that can imply experience that is not
explicitly stated/present in a resume, LinkedIn
profile, or other source of human capital data
▪ Infer: derive as a conclusion from facts or premises
▪ Imply: to contain potentially, to express indirectly
15. Text-based human capital data (e.g., resumes,
LinkedIn profiles, etc.) is intrinsically limited and
never provides a complete picture
People simply do not mention every detail about their
professional career
Many talented people simply cannot be found via
direct search methods, because their experience isn't
explicitly mentioned anywhere
If the text isn't present, it can't be retrieved!
16. Let's say you need someone who has managed EMC
SAN projects/environments
Realizing that some people will not explicitly
mention EMC or SAN (or any variant) in their
resume/profile, you could search specifically for
data center move, migration and consolidation
experience, because this can imply SAN
experience, and EMC is one of the largest SAN
players*
* This isn't a theoretical example - I achieved a high level placement with a fantastic candidate at EMC using
this exact approach!
17. Let's say you need a Business Analyst with PeopleSoft
experience
After exhausting all search methods using "PeopleSoft"
directly in queries, you could NOT out "PeopleSoft" and
search for the mention of companies that you know use
PeopleSoft
People who have worked at a company that is known to
use PeopleSoft have a probability of experience with
PeopleSoft, even in the absence of explicit mention of
"PeopleSoft"*
* I filled a critical role at Sprint/Nextel using this exact method. The candidate had 3 recent and strong years of
PeopleSoft project experience, and neither PeopleSoft nor any PeopleSoft related terminology was anywhere
in her resume
18. Level 4 Semantic Search can be achieved through:
1. Man
Searching specifically for text that can imply skills
and experience that isn't explicitly mentioned
2. Machine
No solution that I am aware of allows for
inferential semantic search beyond Level 1
conceptual search achieved through synonymous
or hierarchical taxonomies (e.g., GAAP implies
accounting exp.)
19. 5. Human-reviewed and classified
The highest level of semantic search involves meaning
applied by people and the ability to search for human
capital data (resumes, social profiles, etc.) that has
been identified, analyzed and labeled by a human
Searchable tagging allows the retrieval of human
capital data that has been labeled after human
analysis that can include information not actually
present in the document/profile, as well as
"intangibles" such as personality and cultural match
20. Level 5 Semantic Search can be achieved through:
1. Man
Tagging human capital documents, records and
profiles and the ability to search by tags
2. Machine
I'm not aware of any solution that has been
developed to do this, but if I were to design one,
it would involve the ability to automatically
match across human-applied tags