Elasticsearch Query String Quote_Analyzer

Elasticsearch Query String Quote_Analyzer

Recently, I was fortunate enough to professionally work on an ASP.NET MVC project again. I worked in tandem with a team leader to switch the application so it used Elasticsearch (ES) instead of Google Search Appliance (GSA). Surprisingly, the switch went really well. There were only a few "enhancements" that needed to be done after the switch.

One such enhancement was something called stopwords. When users searched for words like "a", "an" and "the", they were getting back way too many hits. My team leader, created an analyzer that used the stopwords feature to get rid of this annoying behavior [1]. I then modified the ASP.NET MVC code for the project to apply the said analyzer when necessary. This got rid of that problem, but later when we decided to allow more advanced queries, a different issue popped up.

Initially, basic queries were useful for the ASP.NET MVC app we had worked on. At some point, however, the users needed additional functionality. An easy way to allow this was to enable the Query String Query [2] feature of ES. It allows interesting queries to be constructed using boolean roles, Match Phrase (aka: Quoted) queries [3] and other goodies.

So we started leveraging the Query String component of ES. It worked great but we ran into a small problem. If our users searched for phrases that contained stopwords, they would get way too many hits since the stopwords would get removed from the phrase. As an example, if a user searched for "War of the Worlds" on our system, the query would be altered to "War Worlds" since the analyzer would remove the stopwords "of" and "the". This modified query would then match all sorts of combinations beyond what was intended since stop words between "War" and "Words" in the "War Worlds" phrase would be ignored. Even though "War of the Worlds" was originally specified, as long as our original analyzer was running on the phrase, the "War of the Worlds" could also match "War in the World", "War over the World", etc... because "of", "the", "in" and "over" are all stopwords that would be ignored.

Well, we didn't want to turn the analyzer completely off so I searched for days and hours looking for a solution. I was trying to figure out how to use the stopwords feature to filter out silly queries when stopwords are not needed but yet allow them for Match Phrase queries. I was thinking the solution would be very hard and complicated. It turns out the solution was quite simple. If you look back at [2], you will notice that ES allows you to specify two analyzers in the JSON body. One of them is the regular analyzer (which is intuitively named "analzyer") and the other one is an analyzer that only applies to quoted or Match Phrase queries (this one is named "quote_analyzer").

To keep excluding stopwords for basic queries but not exclude stopwords for quoted queries, you simply continue specifying your regular analzyer via the "analyzer" keyword but set the quote_analzyer to something sensible like the standard analyzer. An example of how the JSON for this might look appears below:

                GET /_search
                    "query": {
                        "query_string" : {
                            "query" : "\"War of the Worlds\" OR "\"Star Trek\""
                    "analyzer": "your_analyzer_to_remove_stopwords",
                    "quote_analyzer": "standard"

Pretty simple. I'm suprised it took me so long to figure out that all I had to do was change one line to adjust quoted searches so they didn't exclude stopwords.


1. Elasticsearch Reference [7.0] - Analysis

2. Elasticsearch Reference [7.0] - Query String Query

3. Elasticsearch Reference [7.0] - Match Phrase Query