Word clouds with phrases, not just single words

A classic word cloud counts single words, so "customer service" becomes two unrelated words, "customer" and "service", and "waiting time" turns into "waiting" and "time". WordCloud+ counts phrases of two, three and four words alongside single words and ranks them all together, so an idea that people express as a phrase appears as that phrase.

There is nothing to switch on. Paste your text into Create a word cloud and the free preview already includes phrases. The settings below, in the full editor, let you tune how phrases are found and how big they are drawn.

How WordCloud+ finds a phrase

A run of two to four words becomes a phrase when all of these are true:

• It occurs at least twice. Single words count from their first use, but a phrase has to repeat to be treated as one.
• The words sit next to each other in the same passage. WordCloud+ splits your text into passages at line breaks, commas, full stops, question marks and exclamation marks, and a phrase never crosses one of those. Putting each survey response or comment on its own line helps.
• It does not start or end with a stop word. "the service" and "service was" are never phrases; "service" can be.
• It does not have too many stop words inside. A two-word phrase can contain none. A three- or four-word phrase can contain stop words in the middle, up to the limit you set, so "proof of concept" and "quality of care" can both appear.

Because stop words shape phrases, editing them changes which phrases you get. See How to remove words from a word cloud.

Choose how phrases are scored

The size of each term comes from its score. Open the Settings tab and pick a Scoring Algorithm:

• Combination Counting (the default). When a longer phrase is counted, those occurrences are taken away from the shorter phrases and single words inside it, so "customer service" does not also inflate "customer" and "service". The phrase then gets a bonus for each of its words that also appears on its own, multiplied by the Weighting setting. This is what lets meaningful phrases stand out.
• Deductions. The same taking-away as Combination Counting, without the bonus. Phrases are sized by how often they occur and no more.
• Frequency. Plain counts. Every word and phrase is counted independently, so the same occurrence of "customer service" counts towards "customer", "service" and "customer service". Single words tend to dominate.

Make phrases bigger or smaller

With Combination Counting selected, the Weighting slider under Combination Counting on the Settings tab runs from 0 to 5 and starts at 2. Raise it and phrases grow relative to single words; set it to 0 and phrases get no bonus at all, which behaves like Deductions. The slider is greyed out for the other two algorithms.

If you would rather see mostly single words, choose Frequency. Common words keep their full counts and will usually outrank the phrases that contain them.

Allow stop words inside longer phrases

Two further settings on the Settings tab control the stop word limit inside longer phrases:

• Three word phrases, Allowed Stop Words: 0 or 1 (default 1). With 0, "proof of concept" is no longer a phrase.
• Four word phrases, Allowed Stop Words: 0, 1 or 2 (default 1). With 2, phrases such as "waiting on the phone" can appear.

Lower limits give tighter, more content-heavy phrases. Higher limits keep more natural wording at the risk of some filler.

Check what the phrases are based on

The Summary tab lists every term with its Score and number of Occurrences, and Download saves that table as a CSV file. Choose Examine on a row, or click a phrase in the cloud, then open References to see every passage the phrase came from. It is worth a look before you present a cloud: a large phrase should reflect what people said, not a repeated heading or signature.

Keep a phrase together yourself

Some phrases matter even when they fail the rules above, such as a product name used once or a phrase interrupted by a comma. Use the Replace tab to join the words with a hyphen: Find net promoter score, Replace net-promoter-score. The hyphenated form counts as one word, so it appears whenever it occurs. You can also merge different wordings of the same idea into one phrase. How to combine and rename words in a word cloud walks through it.

Keep the cloud readable

Phrases take more room than single words. If the cloud feels crowded, lower the Maximum term count from the term count button in the Cloud toolbar (2 to 200, 50 by default), or widen the canvas with the dimensions button.

Want to see which phrases your own text produces? Create a word cloud from it now; the preview is free and needs no account.