MongoDB Full and Partial Text Search

Env:

• MongoDB (3.2.0) with Mongoose

Collection:

• users

Text Index creation:

  BasicDBObject keys = new BasicDBObject();
  keys.put("name","text");

  BasicDBObject options = new BasicDBObject();
  options.put("name", "userTextSearch");
  options.put("unique", Boolean.FALSE);
  options.put("background", Boolean.TRUE);
  
  userCollection.createIndex(keys, options); // using MongoTemplate

Document:

• {"name":"LEONEL"}

Queries:

• db.users.find( { "$text" : { "$search" : "LEONEL" } } ) => FOUND
• db.users.find( { "$text" : { "$search" : "leonel" } } ) => FOUND (search caseSensitive is false)
• db.users.find( { "$text" : { "$search" : "LEONÉL" } } ) => FOUND (search with diacriticSensitive is false)
• db.users.find( { "$text" : { "$search" : "LEONE" } } ) => FOUND (Partial search)
• db.users.find( { "$text" : { "$search" : "LEO" } } ) => NOT FOUND (Partial search)
• db.users.find( { "$text" : { "$search" : "L" } } ) => NOT FOUND (Partial search)

Any idea why I get 0 results using as query "LEO" or "L"?

Regex with Text Index Search is not allowed.

db.getCollection('users')
     .find( { "$text" : { "$search" : "/LEO/i", 
                          "$caseSensitive": false, 
                          "$diacriticSensitive": false }} )
     .count() // 0 results

db.getCollection('users')
     .find( { "$text" : { "$search" : "LEO", 
                          "$caseSensitive": false, 
                          "$diacriticSensitive": false }} )
.count() // 0 results

MongoDB Documentation:

• Text Search
• $text
• Text Indexes
• Improve Text Indexes to support partial word match

Solution 1:

As at MongoDB 3.4, the text search feature is designed to support case-insensitive searches on text content with language-specific rules for stopwords and stemming. Stemming rules for supported languages are based on standard algorithms which generally handle common verbs and nouns but are unaware of proper nouns.

There is no explicit support for partial or fuzzy matches, but terms that stem to a similar result may appear to be working as such. For example: "taste", "tastes", and tasteful" all stem to "tast". Try the Snowball Stemming Demo page to experiment with more words and stemming algorithms.

Your results that match are all variations on the same word "LEONEL", and vary only by case and diacritic. Unless "LEONEL" can be stemmed to something shorter by the rules of your selected language, these are the only type of variations that will match.

If you want to do efficient partial matches you'll need to take a different approach. For some helpful ideas see:

• Efficient Techniques for Fuzzy and Partial matching in MongoDB by John Page
• Efficient Partial Keyword Searches by James Tan

There is a relevant improvement request you can watch/upvote in the MongoDB issue tracker: SERVER-15090: Improve Text Indexes to support partial word match.

Solution 2:

As Mongo currently does not supports partial search by default...

I created a simple static method.

import mongoose from 'mongoose'

const PostSchema = new mongoose.Schema({
    title: { type: String, default: '', trim: true },
    body: { type: String, default: '', trim: true },
});

PostSchema.index({ title: "text", body: "text",},
    { weights: { title: 5, body: 3, } })

PostSchema.statics = {
    searchPartial: function(q, callback) {
        return this.find({
            $or: [
                { "title": new RegExp(q, "gi") },
                { "body": new RegExp(q, "gi") },
            ]
        }, callback);
    },

    searchFull: function (q, callback) {
        return this.find({
            $text: { $search: q, $caseSensitive: false }
        }, callback)
    },

    search: function(q, callback) {
        this.searchFull(q, (err, data) => {
            if (err) return callback(err, data);
            if (!err && data.length) return callback(err, data);
            if (!err && data.length === 0) return this.searchPartial(q, callback);
        });
    },
}

export default mongoose.models.Post || mongoose.model('Post', PostSchema)

How to use:

import Post from '../models/post'

Post.search('Firs', function(err, data) {
   console.log(data);
})

Solution 3:

Without creating index, we could simply use:

db.users.find({ name: /<full_or_partial_text>/i}) (case insensitive)

Solution 4:

If you want to use all the benefits of MongoDB's full-text search AND want partial matches (maybe for auto-complete), the n-gram based approach mentioned by Shrikant Prabhu was the right solution for me. Obviously your mileage may vary, and this might not be practical when indexing huge documents.

In my case I mainly needed the partial matches to work for just the title field (and a few other short fields) of my documents.

I used an edge n-gram approach. What does that mean? In short, you turn a string like "Mississippi River" into a string like "Mis Miss Missi Missis Mississ Mississi Mississip Mississipp Mississippi Riv Rive River".

Inspired by this code by Liu Gen, I came up with this method:

function createEdgeNGrams(str) {
    if (str && str.length > 3) {
        const minGram = 3
        const maxGram = str.length
        
        return str.split(" ").reduce((ngrams, token) => {
            if (token.length > minGram) {   
                for (let i = minGram; i <= maxGram && i <= token.length; ++i) {
                    ngrams = [...ngrams, token.substr(0, i)]
                }
            } else {
                ngrams = [...ngrams, token]
            }
            return ngrams
        }, []).join(" ")
    } 
    
    return str
}

let res = createEdgeNGrams("Mississippi River")
console.log(res)

Now to make use of this in Mongo, I add a searchTitle field to my documents and set its value by converting the actual title field into edge n-grams with the above function. I also create a "text" index for the searchTitle field.

I then exclude the searchTitle field from my search results by using a projection:

db.collection('my-collection')
  .find({ $text: { $search: mySearchTerm } }, { projection: { searchTitle: 0 } })