How to convert fields to JSON in Postgresql

Question

I have a table with the following schema (postgresql 14):

message   sentiment     classification
any text  positive    mobile, communication

message are only string, phrases. sentiment is a string, only one word classification are string but can have 1 to many word comma separated

I would like to create a json field with these columns, like this:

{"msg":"any text", "sentiment":"positive","classification":["mobile,"communication"]}

Also, if possible, is there a way to consider the classification this way:

{"msg":"any text", "sentiment":"positive","classification 1":"mobile","classification 2" communication"}

Answer 1

Use jsonb_build_object and concatenate the columns you want

SELECT 
  jsonb_build_object(
      'msg',message,
      'sentiment',sentiment,
      'classification',       
          string_to_array(classification,',')) 

FROM mytable;

Demo: db<>fiddle

The second output is definitely not trivial. The SQL code would be much larger and harder to maintain - not to mention that parsing such file also requires a little more effort.

Answer 2

The first part of question is easy - Postgres provides functions for splitting string and converting to json:

with t(message,   sentiment,     classification) as (values
('any text','positive','mobile, communication')
)
select row_to_json(x.*)
from (
  select t.message
       , t.sentiment
       , array_to_json(string_to_array(t.classification, ', ')) as classification
  from t
) x

The second part is harder - your want json to have variable number of attributes, mixed of grouped and nongrouped data. I suggest to unwind all attributes and then assemble them back (note the numbered CTE is actually not needed if your real table has id - I just needed some column to group by):

with t(message,   sentiment,     classification) as (values
('any text','positive','mobile, communication')
)
, numbered (id, message,   sentiment,     classification) as (
  select row_number() over (order by null)
       , t.*
  from t
)
, extracted (id,message,sentiment,classification,index) as (
  select n.id
       , n.message
       , n.sentiment
       , l.c
       , l.i
  from numbered n
  join lateral unnest(string_to_array(n.classification, ', ')) with ordinality l(c,i) on true
), unioned (id, attribute, value) as (
  select id, concat('classification ', index::text), classification
  from extracted
  union all
  select id, 'message', message
  from numbered
  union all
  select id, 'sentiment', sentiment
  from numbered
)
select json_object_agg(attribute, value)
from unioned
group by id;

DB fiddle

Answer 3

You can use a cte to handle the flattening of the classification attributes and then perform the necessary grouping in the main queries for each problem component:

with cte(r, m, s, k) as (
   select row_number() over (order by t.message), t.message, t.sentiment, v.* from tbl t 
   cross join json_array_elements(array_to_json(string_to_array(t.classification, ', '))) v
)
-- first part --
select json_build_object('msg', t1.message, 'sentiment', t1.sentiment, 'classification', string_to_array(t1.classification, ', ')) from tbl t1
-- second part --
select jsonb_build_object('msg', t1.m, 'sentiment', t1.s)||('{'||t1.g||'}')::jsonb 
from (select c.m, c.s, array_to_string(array_agg('"classification '||c.r||'":'||c.k), ', ') g 
      from cte c group by c.m, c.s) t1