Searching with punctuation about lifti

[mikegoatly]

Hi @PetesBreenCoding,

The short answer is as things stand, fuzzy matching will do the trick for you - I think the confusion comes from your use of wildcards in the matching. This fiddle has some working code for you to play with:

var index = new FullTextIndexBuilder<int>()
    .WithQueryParser(o => o.AssumeFuzzySearchTerms())
    .Build();

await index.AddAsync(1, "Murphy's law");

Console.WriteLine(index.Search("Murphys").Count());
// Prints "1"

Console.WriteLine(index.Search("Murphy's").Count());
// Prints "1"

But I feel that your question deserves a bit of a deeper dive into what's going on.

By default LIFTI will split words on punctuation, including apostrophes. This means that (rightly or wrongly) "Murphy's law" will actually get tokenized as three words:

• Murphy
• s
• law

When you search for "Murphy's", the query is actually parsed as "Murphy & s" because search terms are also split using the same tokenization. You can see this using the following code (with fuzzy matching on as default):

var query = index.QueryParser.Parse(index.FieldLookup, "Murphy's", index.DefaultTokenizer);
Console.WriteLine(query.Root.ToString());

// Prints "?3,?MURPHY & ?0,?S"

This is searching for documents containing fuzzy matches for "murphy" and "s" (although because "s" is so short, only an exact match would be acceptable - the zero in ?0,? is stating that no edits are allowed)

When you add wildcards into the mix, for your first example you get this:

var query = index.QueryParser.Parse(index.FieldLookup, "*Murphy's*", index.DefaultTokenizer);
Console.WriteLine(query.Root.ToString());
// Prints "*MURPHY & S*"

Which essentially means documents containing any words ending with "murphy" and any words starting with "S". This will by coincidence match your document.

But your second search looks like this:

var query = index.QueryParser.Parse(index.FieldLookup, "*Murphys*", index.DefaultTokenizer);
Console.WriteLine(query.Root.ToString());
// Prints *MURPHYS*

Which means documents containing any words completely containing "murphys" - this won't match in your index.

[PetesBreenCoding]

Thanks very much @mikegoatly. I have been playing around with it a bit, and I have it working pretty well with the fuzzy logic. I also mixed in the wildcard search like so:
?1?query | *query*

This is so I can match on "murphys", "murphy's" but also "mur". I found the default fuzzy matching was returning too many irrelevant results in my database. I'll tweak this over time once I get used to the library a bit more.
Thanks again!

[mikegoatly]

No problem at all, glad to help.

One thing to be aware of is that using a * at the start of your search terms isn't particularly efficient for large indexes at the moment because of the way the index structure has to be recursively scanned to find the first character to match. The queries will be faster if you can just use a wildcard at the end, e.g. mur*, although you may not notice if your index isn't particularly big.

[PetesBreenCoding]

At the moment, this particular index is only 2,000 records, and will probably never be any more than 10,000. I assume they would be considered small numbers?

[mikegoatly]

It'll probably be fine - it was more just something to be aware of. Let me know if you run into any problems though.

[mikegoatly]

Closing this as the question is resolved. Feel free to raise another issue if anything else come up.