Sorting by relevance with MongoDB - mongodb

Question

I have a collection of documents in the following form:
{ _id: ObjectId(...)
, title: "foo"
, tags: ["bar", "baz", "qux"]
}
The query should find all documents with any of these tags. I currently use this query:
{ "tags": { "$in": ["bar", "hello"] } }
And it works; all documents tagged "bar" or "hello" are returned.
However, I want to sort by relevance, i.e. the more matching tags the earlier the document should occur in the result. For example, a document tagged ["bar", "hello", "baz"] should be higher in the results than a document tagged ["bar", "baz", "boo"] for the query ["bar", "hello"]. How can I achieve this?

Answer 1

MapReduce and doing it client-side is going to be too slow - you should use the aggregation framework (new in MongoDB 2.2).
It might look something like this:
db.collection.aggregate([
{ $match : { "tags": { "$in": ["bar", "hello"] } } },
{ $unwind : "$tags" },
{ $match : { "tags": { "$in": ["bar", "hello"] } } },
{ $group : { _id: "$title", numRelTags: { $sum:1 } } },
{ $sort : { numRelTags : -1 } }
// optionally
, { $limit : 10 }
])
Note the first and third pipeline members look identical, this is intentional and needed. Here is what the steps do:
pass on only documents which have tag "bar" or "hello" in them.
unwind the tags array (meaning split into one document per tags element
pass on only tags exactly "bar" or "hello" (i.e. discard the rest of the tags)
group by title (it could be also by "$_id" or any other combination of original document
adding up how many tags (of "bar" and "hello") it had
sort in descending order by number of relevant tags
(optionally) limit the returned set to top 10.

Answer 2

You could potentially use MapReduce for something like that. You'd process each document in the Map step, figuring out how many tags match the query, and assign a score. Then you could sort based on that score.
http://www.mongodb.org/display/DOCS/MapReduce

Answer 3

Something that complex should be done after querying. Either server-side through db.eval (if your client supports this) or just clientside. Here's an example for what you're looking for.
It will retreive all posts with the tags you specified, then sorts them according to the amount of matches.
remove the db.eva( part and translate it to the language your client uses to query to get the clientside effect (
db.eval(function () {
var tags = ["a","b","c"];
return db.posts.find({tags:{$in:tags}}).toArray().sort(function(a,b){
var matches_a = 0;
var matches_b = 0;
a.tags.forEach(function (tag) {
for (t in tags) {
if (tag == t) {
matches_a++;
} else {
matches_b++;
}
}
});
b.tags.forEach(function(tag) {
for (t in tags) {
if (tag == t) {
matches_b++;
} else {
matches_a++;
}
}
});
return matches_a - matches_b;
});
});