A Semantic Search Use Case: But What about General Business Content with Words and Charts?

September 9, 2022

I am okay with semantic search. The idea is that a relatively standard suite of mathematical procedures delivers “close enough for horse shoes” matches germane to a user’s query. Elastic is now combining key word with some semantic goodness. The idea is that mixing methods delivers more useful results. Is this an accurate statement?

The answer is, “It depends on the use cases.”

How Semantic Search Improves Search Accuracy” explains a use case that is anchored in a technical corpus. Now I don’t want to get crossways with a group of search experts. I would submit that, in general, the vocabulary for scientific, medical, and technical information is more constrained. One does not expect to find “cheugy” or OG* in a write up about octonitrocubane.

In my limited experience, what happens is that a constrained corpus allows the developer of a finding system to use precise taxonomies, and some dinobabies may employ controlled vocabularies like those kicking around old-school commercial databases.

However, what happens when the finding system ingests a range of content objects from tweets, online news services, and TikTok-type content?

The write up says:

One particular advantage of semantic search is the resolution of ambiguous terminology and that all specific subtypes (“children”) of a technical term will be found without the need to mention them in the query explicitly.

Sounds good, particularly for scientific and technical content. What about those pesky charts and graphs? These are often useful, but many times are chock full of fudged data. What about the query, “Octonitrocubane invalid data”? I want to have the search system present links to content which may be in an article. Why? I want to make sure the alleged data set squares with my limited knowledge of statistical principles. Yeah, sorry.

The write up asserts:

A lexical search will deliver back all documents in which “pesticides” is mentioned as the text string “pesticides” plus variants thereof. A semantic search will, in addition to all documents containing the text string “pesticides”, also return documents that contain specific pesticides like bixafen, boscalid, or imazamox.

What about a chemical structure search? I want a document with structure information. Few words, just nifty structures just like the stuff inorganic and organic chemists inhale each day. Sorry about that.

Net net: Writing about search is tough when the specific corpus, the content objects, and the presence of controlled terms in addition to strings in a content object are not spelled out. Without this information, the assertions are a bit fluffy.

And the video thing? The DoD, NIST, and other outfits are making videos. Things that go boom are based on chemistry. Can semantic search find the videos and the results of tests?

Yeah, sure. The PowerPoint deck probably says so. Hands on search experience may not. Search-enabled applications may work better than plain old search jazzed up with close enough for horse shoes methods.

Stephen E Arnold, September 9, 2022

[* OG means original gangster]


Got something to say?

  • Archives

  • Recent Posts

  • Meta