Analysis Screen

Once you’ve defined a field type in your Schema, and specified the analysis steps that you want applied to it, you should test it out to make sure that it behaves the way you expect it to.

Luckily, there is a very handy page in the Solr Admin UI that lets you do just that. You can invoke the analyzer for any text field, provide sample input, and display the resulting token stream.

For example, let’s look at some of the "Text" field types available in the bin/solr -e techproducts example configuration, and use the Analysis Screen (http://localhost:8983/solr/#/techproducts/analysis) to compare how the tokens produced at index time for the sentence "Running an Analyzer" match up with a slightly different query text of "run my analyzer".

We can begin with text_ws - one of the most simplified Text field types available:

This shows a simple output of only the tokens produced by each step of analysis (a single step in this case). The tokenizer is shown with an abbreviation, hover or click on this to see the full name.

If we enable Verbose Output by checking the box, more details are shown:

By looking at the start and end positions for each term, we can see that the only thing this field type does is tokenize text on whitespace. Notice in this image that the term "Running" has a start position of 0 and an end position of 7, while "an" has a start position of 8 and an end position of 10, and "Analyzer" starts at 11 and ends at 19. If the whitespace between the terms was also included, the count would be 21; since it is 19, we know that whitespace has been removed from this query.

Note also that the indexed terms and the query terms are still very different. "Running" doesn’t match "run", "Analyzer" doesn’t match "analyzer" (to a computer), and obviously "an" and "my" are totally different words. If our objective is to allow queries like "run my analyzer" to match indexed text like "Running an Analyzer" then we will need to pick a different field type with index and query time text analysis that does more processing of the inputs.

In particular we want:

For our next attempt, let’s try the text_general field type:

With the verbose output enabled, we can see how each stage of our new analyzers modify the tokens they receive before passing them on to the next stage. As we scroll down to the final output, we can see that we do start to get a match on "analyzer" from each input string, thanks to the "LCF" stage, which if you hover over with your mouse, you’ll see is the LowerCaseFilter:

The text_general field type is designed to be generally useful for any language, and it has definitely gotten us closer to our objective than text_ws from our first example by solving the problem of case sensitivity. It’s still not quite what we are looking for because we don’t see stemming or stopword rules being applied. So now let us try the text_en field type:

Now we can see the "SF" (StopFilter) stage of the analyzers solving the problem of removing Stop Words ("an"), and as we scroll down, we also see the "PSF" (PorterStemFilter) stage apply stemming rules suitable for our English language input, such that the terms produced by our "index analyzer" and the terms produced by our "query analyzer" match the way we expect.

At this point, we can continue to experiment with additional inputs, verifying that our analyzers produce matching tokens when we expect them to match, and disparate tokens when we do not expect them to match, as we iterate and tweak our field type configuration.