Provide query instead of index name when generating recommendations

[pratikshya]

Would it be possible to give a query instead of index name alone when generating the recommendations? For example, in the code below:
curl -XPOST localhost:9200/_river/movielens_items_from_user/_meta -d '{
"type": "taste",
"action": "recommended_items_from_user",
"num_of_items": 10,
"data_model": {
"cache": {
"weight": "100m"
}
},
"index_info": {
"index": "movielens"
}
}'

Instead of providing the index_info, would it be also possible to provide some query which will bring subset of the actual dataset? I do not want to recommend items for user from the entire dataset but subset of it. And having said that I do not want to have one index per user as well. Is it possible to use query as source of data instead of index itself?