How to generate all n-grams in Hive

Question

I'd like to create a list of n-grams using HiveQL. My idea was to use a regex with a lookahead and the split function - this does not work, though:

select split('This is my sentence', '(\\S+) +(?=(\\S+))');

The input is a column of the form

|sentence                 |
|-------------------------|
|This is my sentence      |
|This is another sentence |

The output is supposed to be:

["This is","is my","my sentence"]
["This is","is another","another sentence"]

There is an n-grams udf in Hive but the function directly calculates the frequency of the n-grams - I'd like to have a list of all the n-grams instead.

Thanks a lot in advance!

Answer 1

This is maybe not the most optimal but quite working solution. Split sentence by delimiter (in my example it is one or more space or comma), then explode and join to get n-grams, then assemble array of n-grams using collect_set (if you need unique n-grams) or collect_list:

with src as 
(
select source_data.sentence, words.pos, words.word
  from
      (--Replace this subquery (source_data) with your table
       select stack (2,
                     'This is my sentence', 
                     'This is another sentence'
                     ) as sentence
      ) source_data 
        --split and explode words
        lateral view posexplode(split(sentence, '[ ,]+')) words as pos, word
)

select s1.sentence, collect_set(concat_ws(' ',s1.word, s2.word)) as ngrams 
      from src s1 
           inner join src s2 on s1.sentence=s2.sentence and s1.pos+1=s2.pos              
group by s1.sentence;

Result:

OK
This is another sentence        ["This is","is another","another sentence"]
This is my sentence             ["This is","is my","my sentence"]
Time taken: 67.832 seconds, Fetched: 2 row(s)