Unknown duplicates from querying a nested JSON - json

Question

I would like to do text search in a JSON object in a table.
I have a table called Audio that is structured like below:
id| keyword | transcript | user_id | company_id | client_id
-----------------------------------------------------------
This is the JSON data structure of transcript:
{"transcript": [
{"duration": 2390.0,
"interval": [140.0, 2530.0],
"speaker": "Speaker_2",
"words": [
{"p": 0, "s": 0, "e": 320, "c": 0.545, "w": "This"},
{"p": 1, "s": 320, "e": 620, "c": 0.825, "w": "call"},
{"p": 2, "s": 620, "e": 780, "c": 0.909, "w": "is"},
{"p": 3, "s": 780, "e": 1010, "c": 0.853, "w": "being"},
{"p": 4, "s": 1010, "e": 1250, "c": 0.814, "w": "recorded"}
]
},
{"duration": 4360.0,
"interval": [3280.0, 7640.0],
"speaker": "Speaker_1",
"words": [
{"p": 5, "s": 5000, "e": 5020, "c": 0.079, "w": "as"},
{"p": 6, "s": 5020, "e": 5100, "c": 0.238, "w": "a"},
{"p": 7, "s": 5100, "e": 5409, "c": 0.689, "w": "group"},
{"p": 8, "s": 5410, "e": 5590, "c": 0.802, "w": "called"},
{"p": 9, "s": 5590, "e": 5870, "c": 0.834, "w": "tricks"}
]
},
...
}
What I am trying to do is to do a text search in the "w" field within "words". This is the query that I tried to run:
WITH info_data AS (
SELECT transcript_info->'words' AS info
FROM Audio t, json_array_elements(transcript->'transcript') AS transcript_info)
SELECT info_item->>'w', id
FROM Audio, info_data idata, json_array_elements(idata.info) AS info_item
WHERE info_item->>'w' ilike '%this';
Right now I only have four columns with data and the fifth column is null. And there are five columns in total. However, I got the following result where even the column that doesn't have data results an output:
?column? | id
----------+----
This | 2
This | 5
This | 1
This | 3
This | 4
This | 2
This | 5
I would love to know what the problem of my query is and whether there are more efficient way in doing this.

Answer 1

The problem is that you make a cartesian join between table Audio on the one hand and info_data and info_item on the other hand (there is an implicit lateral join between these latter two) here:
FROM Audio, info_data idata, json_array_elements(idata.info) AS info_item
You can solve this by adding Audio.id to the CTE and then adding WHERE Audio.id = info_data.id.
It is doubtful that this is the most efficient solution (CTEs rarely are). If you just want to get those rows where the word "this" is a word in the transcript, then you are most likely better off like this:
SELECT DISTINCT id
FROM (
SELECT id, transcript_info->'words' AS info
FROM Audio, json_array_elements(transcript->'transcript') AS transcript_info) AS t,
json_array_elements(info) AS words
WHERE words->>'w' ILIKE 'this';
Note that the % in the pattern string is very inefficient. Since very few words in the English language other than "this" end with the same, I have taken the liberty of removing it.