MySQL finding the most similar user - mysql

Question

Let's say I have a table that looks like this:
Mark - Green
Mark - Blue
Mark - Red
Adam - Yellow
Andrew - Red
Andrew - Green
And my objective is to compare the user "Mark" with all the other users in the database, to find out which other user he is most similar to. In this case, he would be most similar to Andrew (2/3 matches), and least similar to Adam (0/3) matches. After I've found out which user is most similar to Mark I would like to extract the entires that Andrew got but Mark doesn't.
Would this be possible in MySQL? I appreciate all help, thank you guys!
Edit: OVERWHELMED by all the good help! THANK you SO MUCH guys! I will be sure to check out all of your contributions!

Answer 1

The following query attempts to list all the users with the number of matches to Mark. It basically joins the table with Mark's entries and counts the common entries for all users.
SELECT ours.user, theirs.user, count(*) as `score`
FROM tableName as `theirs`, (SELECT *
FROM tableName
WHERE user = 'Mark') as `ours`
WHERE theirs.user != 'Mark' AND
theirs.color = ours.color
GROUP BY theirs.user
ORDER BY score DESC
The query, however, wouldn't work if there's duplicate data (i.e. one person picks the same color twice). But that shouldn't be a problem as you mention in the comments that it wouldn't occur.
The query can be modified to show the score for all users:
SELECT ours.user as `myUser`, theirs.user as `theirUser`, count(*) as `score`
FROM tableName as `ours`, tableName as `theirs`
WHERE theirs.user != ours.user AND
theirs.color = ours.color
GROUP BY ours.user, theirs.user
ORDER BY score DESC
Let Q be the above query that gives you the most similar user. Once you have that user, you can use it to show the distinct entries between them. Here's what we're trying to do:
SELECT *
FROM tableName as theirs
WHERE user = 'Andrew'
AND NOT EXISTS (SELECT 1
FROM tableName as ours
WHERE ours.user = 'Mark'
AND ours.color = theirs.color)
Replacing the inputs Andrew and Mark from Q:
SELECT similar.myUser, theirs.user, theirs.color
FROM tableName as theirs JOIN (Q) as similar
ON theirs.user = similar.theirUser
WHERE NOT EXISTS (SELECT 1
FROM tableName as ours
WHERE ours.user = similar.myUser
AND ours.color = theirs.color)
Here's the final query up and running. Hope this makes sense.

Answer 2

Use FULLTEXT INDEXES. And your query will be like:
SELECT * FROM user WHERE MATCH (name,color) AGAINST ('Mark blue');
Or simplest way, is using LIKE search
SELECT * FROM user WHERE name LIKE '%Mike%' OR color = 'blue'
You can choose which way more suitable for you

Answer 3

select
name,
sum(case when t2.cnt > t1.cnt then t1.cnt else t2.cnt end) matches
from (
select name, color, count(*) cnt
from table
where name <> 'Mark'
group by name, color
) t1 left join (
select color, count(*) cnt
from table
where name = 'Mark'
group by color
) t2 on t2.color = t1.color
group by name
order by matches desc
The derived table t1 contains the # of colors each user (except Mark) has, t2 contains the same for Mark. The tables are then left joined on the color and the smaller of the 2 counts is taken i.e. if Amy has 2 reds and Mark has 1 red, then 1 is taken as the number of matches. Finally group by name and return the largest sum of matches.

Answer 4

select match.name, count(*) as count
from table
join table as match
on match.name <> table.name
and table.name = 'mark'
and match.color = table.color
group by match.name
order by count(*) desc

Answer 5

Below query returns matching score between name and matching_name and the maximum score it could get, so that you know what % value your matching has.
This code counts duplicate values in color column as only one, so that if you have record Mark - Red twice, it will only count as 1.
select
foo.name, foo.matching_name, count(*) AS matching_score, goo.color_no AS max_score
from
(
select
distinct a.name, a.color, b.name AS matching_name
from
(
select name, color from yourtable
) a
left join yourtable b on a.color = b.color and a.name <> b.name
where b.name is not null
) foo
left join ( select name, count(distinct color) AS color_no from yourtable group by name ) goo
on foo.name = goo.name
group by foo.name, foo.matching_name
Attaching SQLFiddle to preview the output.

Answer 6

This should get you close. The complexity comes from the fact that you allow each user to pick each color multiple times and require that each otherwise-identical pair be matched in the other user you're comparing to. Therefore, we're really interested in knowing how many total color picks a user per color and how that number compares to the compared-to users' count for that same color.
First, we create a derived relation that does the simple math for us (counting up each users' number of picks by color):
CREATE VIEW UserColorCounts (User, Color, TimesSeen)
AS SELECT User, Color, COUNT(*) FROM YourTable GROUP BY User, Color
Second, we need some kind of relation that compares each color-count for the primary user to the color-counts of each secondary user:
CREATE VIEW UserColorMatches (User, OtherUser, Color, TimesSeen, TimesMatched)
AS SELECT P.User, S.User, P.Color, P.TimesSeen, LEAST(P.TimesSeen, S.TimesSeen)
FROM UserColorCounts P LEFT OUTER JOIN UserColorCounts S
ON P.Color = S.Color AND P.User <> S.User
Finally, we total up the color-counts for each primary user and compare against the matching color counts for each secondary user:
SELECT User, OtherUser, SUM(TimesMatched) AS Matched, SUM(TimesSeen) AS OutOf
FROM UserColorMatches WHERE OtherUser IS NOT NULL
GROUP BY User, OtherUser