What Does $100 Million Buy You? A Semantic Search Engine That Works.

The technical folks behind Monster.com invited me to visit last week. I somehow couldn’t convince them to show me any Superbowl ads but they did demo their cool new search engine. It’s based on technology they acquired when they bought Trovix in 2008. What can it do?

  • Understand the meaning of words: The search engine knows the difference between “development” in the fundraising context and “development” in the software context.
  • Appreciate the relationships between words: A custom ontology fortifies the search engine. The ontology rolls up skills like auditing into the larger category of finance. It differentiates between a top ranked school and a lower ranked school. It understands that years spent working as a prosecutor should count towards a candidate’s overall legal experience.
  • Cut text-heavy resumes into nimble content components: Recruiters can use the power resume search to compare candidates side-by-side, because the search mixes and normalizes the information into simple, clean categories like “Experience,” “Education,” and “Skills”.

Information access vendors like Exalead are hot on the concept of “search-based applications” (SBAs). SBAs are purpose-built information access applications that integrate diverse information from multiple sources. Given that Monster.com is one of the most visited sites on the Web, and a top-tier vertical search engine, what does Monster’s experience tell us about SBAs?

  • Search can be a competitive advantage… but it takes management commitment and serious resources. Monster.com invested in search because search success is crucial to its bottom line. But building a search app is not for the faint of heart. It took over 120 FTEs, 18 months and $30m to integrate, scale and customize the Trovix technology.
  • It’s important to focus on the user and his/her tasks. Monster execs claim they didn’t build a search engine, they built a match engine. Their goal is to match candidates to jobs. Consider the difference between a match engine, a “discovery” engine (as seen on PubMed), and a “recommendation” engine (such as Yelp). These engines have different relevance models to support distinct information needs. A specialized, focused business model is a pre-requisite for a successful SBA.
  • It takes time and traffic to optimize relevance. With millions of queries to interpret and over a dozen analytics staffers, Monster has a lot of data crunching ahead. Human beings must edit and govern the ontologies, and amend the relevance algorithms to match user expectations. Just another example of the operational scale required to pull off vertical search success.

If you have a story to tell about deploying search technology, we would like to hear about it.