Algolia: A New Approach to Relevance?
February 21, 2022
Algolia is a company providing search and retrieval services to a number of companies. A call for résumés provides some interesting assertions about the company, its philosophy, and its goal.
The goal interests me. The posting on the Algolia Web site says:
Our mission is to return relevant search results at all times and give companies the ability to tailor those results to their own specific needs. We are tasked with reimagining a core piece of Algolia’s technology: how search results are ranked. To that end, we are still early in our building, and we are looking for someone who can help us perform experiments and manage the technical aspects of our pilot program, including building clients for users to test our work and tools to evaluate the impact of our changes.
I like the idea of tailoring “search” which is certainly okay if someone knows for that which the individual is looking. I like the idea of ranking because relevance is — to some people — helpful. I like the idea that the company is “early in building.” The right person with the right stuff will make an impact. I like the idea of measuring results, which works reasonably well when the people in the same know that which they need to find.
There are several challenges in delivering or finding better ways to rank search results.
First, today the idea of knowing the corpus and using old-fashioned techniques like precision and recall are not as sexy as capsule network or caps net methods.
Second, users who want to formulate complex search queries like those required to extract semi useful information from Google or a Dataminr feed of social media are rare birds. I heard at one big search outfit that fewer than three percent of queries are a result of complex search statements; for example, site: or filetype:. Serving experts, analysts, and intelligence professionals is different from serving the ingredients for a Sicilian pizza.
Third, the now threadbare truism of lots of data, changing rapidly, and incorporating different content types and a veritable fun house of metadata requires some innovation. So far the best efforts of some bright folks have led to outright failure (Autonomy, Fast Search & Transfer, et al) or recycling endlessly with minor variations the functionality of everyone’s favorite fighter of Amazon, Elastic.
I noted some interesting supportive information in the write up; for instance, the candidate with right stuff must have grit (the sort of effort required to get an advanced degree from MINES ParisTech or Université Paris Saclay or the toughness required to deal with a wealthy family or a generational link to the Capetians. Other ingredients in the “right stuff” trois étoiles cannelés of Bordeaux:
- Trust
- Care
- Candor
- Humility
I am eager to explore the new approach to relevance. But I harbor an abiding affection for a clear explanation of the content indexed and good old Boolean logic. Snorkels, caps nets, and a 21sst century approach to relevance? Meh.
Stephen E Arnold, February 21, 2022
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
