This solution seems to do the job (stolen almost verbatim from this page). It requires an auxiliary
table, filled with sequential numbers from 1 to at least the expected number of distinct words. This is quite important to check that the auxiliary table is large enough, or results will be wrong (showing no error).
SELECT
SUBSTRING_INDEX(SUBSTRING_INDEX(maintable.comment, ' ', auxiliary.id), ' ', -1) AS word,
COUNT(*) AS frequency
FROM maintable
JOIN auxiliary ON
LENGTH(comment)>0 AND SUBSTRING_INDEX(SUBSTRING_INDEX(comment, ' ', auxiliary.id), ' ', -1)
<> SUBSTRING_INDEX(SUBSTRING_INDEX(comment, ' ', auxiliary.id-1), ' ', -1)
GROUP BY word
HAVING word <> ' '
ORDER BY frequency DESC;
SQL Fiddle
This approach is as inefficient as one can be, because it cannot use any index.
As an alterative, I would use a statistics table that I would keep up-to-date with triggers. Perhaps initialise the stats table with the above.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…