Mongo aggregation $group conditional $eq for string field - mongodb

Question

I use Mongo v2.2.0.
I wrote the query but the main issue is $arrayElemAt. Standard replacement with $unwind-$first doesn't work for me and I suppose that better solution exists. I have a restriction to run this aggregation pipeline as a single operation instead of running a query for positive and negative data and later merge results in a code. I need to apply a $sort, $limit and $skip for the resulting query to restrict count of words to be used for filtering records from other collection and combine data from both collections in Java code.
Aggregation query:
[
{
$match: {
"merchantId": ObjectId("59520e6ccc7a701fbed31f94"),
"date": {
"$gte": NumberLong(1389644800000),
"$lt": NumberLong(1502409599999)
},
"isbn": "a123",
}
},
{
$project: {
"word": 1,
"sentence": 1,
"type": 1,
"date": 1
}
},
{
$sort: {
"date": -1
}
},
{
$group: {
"_id": {
"word": "$word",
"type": "$type"
},
"date": {
$max: "$date"
},
"sentence": {
$first: "$sentence"
},
"sentenceCount": {
"$sum": 1
}
},
},
{
$group: {
"_id": "$_id.word",
"word": { $first: "$_id.word"},
"positiveCount": {$sum: {$cond: [{$eq: ["$_id.type", "positive"]}, "$sentenceCount", 0]}},
"count": {$sum: "$sentenceCount"},
"positiveSentence": {
"$push": {
"$cond": [{$eq: ["$_id.type", "positive"]}, "$sentence", "$noval"]
}
},
"negativeSentence": {
"$push": {
"$cond": [{$eq: ["$_id.type", "negative"]}, "$sentence", "$noval"]
}
}
}
},
{
$project: {
"_id": 0,
"word": 1,
"sentimentPercentage": {$cond: [{$eq: ["$count", 0]}, 0, {$multiply: [{$divide: ["$positiveCount", "$count"]}, 100]}]},
"positiveSentence": {$arrayElemAt: ["$positiveSentence", 0]},
"negativeSentence": {$arrayElemAt: ["$negativeSentence", 0]},
}
},
{
$sort: {
sentimentPercentage: -1
}
},
{
$limit: 50
}
]
Collection document "schema":
{
"_id" : ObjectId("59887424e4b099e00724aa44"),
"merchantId" : ObjectId("59520e6ccc7a701fbed31f94"),
"isbn" : "a123",
"sentence" : "Great, friendly service.",
"word" : "service",
"type" : "positive",
"date" : NumberLong(1466809200000),
}
Expected output:
{
"word" : "expectations",
"sentimentPercentage" : 100.0,
"positiveSentence" : "The service exceeded our expectations."
},
{
"word" : "representative",
"sentimentPercentage" : 87.5,
"positiveSentence" : "Excellent local representative, met the flight and gave us all the relevant information to ensure a great holiday.",
"negativeSentence" : "The representative at resort was poor."
},
{
"word" : "seats",
"sentimentPercentage" : 0.0,
"negativeSentence" : "Long delay and pre booked seats were lost ."
}
Please, could you advise me how to replace $arrayElemAt operator or even better how to optimise this query to the desired output using just features of Mongo <=2.2.0?

Answer 1

This appears to give me reasonable results. I think it will not work properly, though, in cases where you have no positive or no negative sentence because of the $unwind stage which does not support the preserveNullAndEmptyArrays parameter in v2.2...
db.getCollection('test').aggregate([
{
$project: {
"word": 1,
"sentence": 1,
"type": 1,
"date": 1
}
},
{
$sort: {
"date": -1
}
},
{
$group: {
"_id": {
"word": "$word",
"type": "$type"
},
"date": {
$max: "$date"
},
"sentence": {
$first: "$sentence"
},
"sentenceCount": {
"$sum": 1
}
},
},
{
$group: {
"_id": "$_id.word",
"word": { $first: "$_id.word"},
"positiveCount": {$sum: {$cond: [{$eq: ["$_id.type", "positive"]}, "$sentenceCount", 0]}},
"count": {$sum: "$sentenceCount"},
"positiveSentence": {
"$push": {
"$cond": [{$eq: ["$_id.type", "positive"]}, "$sentence", "$noval"]
}
},
"negativeSentence": {
"$push": {
"$cond": [{$eq: ["$_id.type", "negative"]}, "$sentence", "$noval"]
}
}
}
},
{ $unwind: "$positiveSentence" },
{ $group:
{
"_id": "$_id",
"word": { $first: "$word" },
"count": { $first: "$count" },
"positiveCount": { $first: "$positiveCount" },
"positiveSentence": { $first: "$positiveSentence" },
"negativeSentence": { $first: "$negativeSentence" },
}
},
{ $unwind: "$negativeSentence" },
{ $group:
{
"_id": "$_id",
"word": { $first: "$word" },
"count": { $first: "$count" },
"positiveCount": { $first: "$positiveCount" },
"positiveSentence": { $first: "$positiveSentence" },
"negativeSentence": { $first: "$negativeSentence" },
}
},
{
$project: {
"_id": 0,
"word": 1,
"sentimentPercentage": {$cond: [{$eq: ["$count", 0]}, 0, {$multiply: [{$divide: ["$positiveCount", "$count"]}, 100]}]},
"positiveSentence": 1,
"negativeSentence": 1
}
}
])
You might be able to simplify this further, e.g. get rid of the first projection and grouping stage. I can perhaps look into that in a few hours if you'd like me to.