Fast way to determine if a string contains a member of a list of strings

[Karch]

I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of strings.
So, think of it in terms of this: I have a string such as the following:

"Smith said the dog belongs (or did belong) to the secretary, an employee of
the company."

Then the list contains the following (and this list is MUCH larger in the
real situation):

Adams
Alan
Jones
Jacobs
Smith
Thaxton
Vela
Zulu

I would need to stop the processing and return (true) as soon as Smith was
found. On the other hand, if the string was changed to the following, there
would be no match and I would return (false):

"Smitherson said the dog belongs (or did belong) to the secretary, an
employee of the company."

In the given string, do know that the matches should begin at a given point
(zero position), but I need to keep processing until I have exhausted the
candidate string in the list - as shown above - to prevent a false match.

I have played around with some different data structures, such as prefix and
suffix trees, an these work in the case that you have a string that you are
trying to match in a list, not vice versa. The approach is required to be
very performant because it will be evaluated millions of times. I am also
okay with an unsafe code approach that works. I just need the evaluations to
terminate as soon as possible rather than looping through every single item
in the list. Even an IndexOf is too slow.

[amdrit]

I think RegEx will help you out here. As far as not iterating over the
entire list, I've no clue how you can selectively not fall through given the
only match in the list is the last item in the list or when there is no
match at all.
"Karch" <news.microsoft.comwrote in message
news:e9**************@TK2MSFTNGP06.phx.gbl...

>I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of strings.
So, think of it in terms of this: I have a string such as the following:

"Smith said the dog belongs (or did belong) to the secretary, an employee
of the company."

Then the list contains the following (and this list is MUCH larger in the
real situation):

Adams
Alan
Jones
Jacobs
Smith
Thaxton
Vela
Zulu

I would need to stop the processing and return (true) as soon as Smith was
found. On the other hand, if the string was changed to the following,
there would be no match and I would return (false):

"Smitherson said the dog belongs (or did belong) to the secretary, an
employee of the company."

In the given string, do know that the matches should begin at a given
point (zero position), but I need to keep processing until I have
exhausted the candidate string in the list - as shown above - to prevent a
false match.

I have played around with some different data structures, such as prefix
and suffix trees, an these work in the case that you have a string that
you are trying to match in a list, not vice versa. The approach is
required to be very performant because it will be evaluated millions of
times. I am also okay with an unsafe code approach that works. I just need
the evaluations to terminate as soon as possible rather than looping
through every single item in the list. Even an IndexOf is too slow.

[John Daragon]

Karch wrote:

I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of strings.
So, think of it in terms of this: I have a string such as the following:

"Smith said the dog belongs (or did belong) to the secretary, an employee of
the company."

Then the list contains the following (and this list is MUCH larger in the
real situation):

Adams
Alan
Jones
Jacobs
Smith
Thaxton
Vela
Zulu

Could you give us some idea of the likely length of the list? And
whether it's static or dynamic ?

jd

[Peter Duniho]

On Wed, 05 Mar 2008 09:05:22 -0800, Karch <news.microsoft.comwrote:

I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of
strings.

I don't know if RegEx has optimizations for dealing with this sort of
thing. It could, but given how long a search string for RegEx could be if
you concatenate all of your possible matches, maybe it doesn't.

That'd be my first try though. It seems like the actual RegEx search
string would be simple (just "or" all of your possible matches together).
If it performs well enough, there you go.

If not, I'd guess there's a fair amount of research into algorithms like
this, but a fairly basic approach is a state graph. It's memory
intensive, but rewards you with good performance. This came up awhile ago
and I posted some sample code to get someone else started. You can find
that post here:
http://groups.google.com/group/micro...06f696d4500b77

I've referred to this post a couple of other times, and while I've never
had anyone say it actually turned out to be useful, they've never said it
wasn't either. :)

I suppose if I'm going to keep referring people to it, maybe I ought to
clean it up a little bit more. But what's there does work and should be
enough to get you pointed in the right direction.

Pete

[Jack Jackson]

One problem I see with this is that you apparently are not doing
simply a string search, but some kind of word search, without defining
what a word is.

You say the "Smithperson said ..." should not have any matches, so
while the string Smith is a substring, it does not match.

Would it match:
(Smith
(Smith)
Smith,
Smith;
Smith-
smith

etc.

The answer to this will probably constrian your possible algorithms.
If it didn't get too many false hits, doing a string search first and
then re-checking to see if the matches are valid word matches might be
OK.

On Wed, 5 Mar 2008 11:05:22 -0600, "Karch" <news.microsoft.comwrote:

>I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of strings.
So, think of it in terms of this: I have a string such as the following:

"Smith said the dog belongs (or did belong) to the secretary, an employee of
the company."

Then the list contains the following (and this list is MUCH larger in the
real situation):

Adams
Alan
Jones
Jacobs
Smith
Thaxton
Vela
Zulu

I would need to stop the processing and return (true) as soon as Smith was
found. On the other hand, if the string was changed to the following, there
would be no match and I would return (false):

"Smitherson said the dog belongs (or did belong) to the secretary, an
employee of the company."

In the given string, do know that the matches should begin at a given point
(zero position), but I need to keep processing until I have exhausted the
candidate string in the list - as shown above - to prevent a false match.

I have played around with some different data structures, such as prefix and
suffix trees, an these work in the case that you have a string that you are
trying to match in a list, not vice versa. The approach is required to be
very performant because it will be evaluated millions of times. I am also
okay with an unsafe code approach that works. I just need the evaluations to
terminate as soon as possible rather than looping through every single item
in the list. Even an IndexOf is too slow.

[Karch]

The list is static (read into memory during initialization) and could be up
to approximately 1000 items (hence the need for a fast method)

"John Daragon" <jo**@argv.co.ukwrote in message
news:47***********************@news.zen.co.uk...

Karch wrote:

>I need to find the fastest way in terms of storage and searching to
determine if a given string contains one of a member of a list of
strings. So, think of it in terms of this: I have a string such as the
following:

"Smith said the dog belongs (or did belong) to the secretary, an employee
of the company."

Then the list contains the following (and this list is MUCH larger in the
real situation):

Adams
Alan
Jones
Jacobs
Smith
Thaxton
Vela
Zulu

Could you give us some idea of the likely length of the list? And whether
it's static or dynamic ?

jd

[Peter Duniho]

On Thu, 06 Mar 2008 10:12:51 -0800, Karch <news.microsoft.comwrote:

The Hashtable would work great alongside a tokenized string, but the
problem
is that I don't have a distinct delimiter since the string to be matched
can
span multiple words. It's really a question of "do any of these strings
appear in this source string", irrespective of whitespace.

Then how do you distinguish "Smith" from "Smitherson", as in your
example? Can you at least make a requirement that the search pattern will
always begin and end on a space, even if it includes spaces within?

Pete

[Mufaka]

Mufaka wrote:

Karch wrote:

>The Hashtable would work great alongside a tokenized string, but the
problem is that I don't have a distinct delimiter since the string to
be matched can span multiple words. It's really a question of "do any
of these strings appear in this source string", irrespective of
whitespace.

<snip>
This is a little different than your original post, but if you can't
delimit search string, you might try something like the following:

Hash the static list of words
Keep a list of unique lengths for those words
for each unique length, iterate the string pulling out substrings of
that length
see if that substring is in your hashed list of words

It at least reduces the amount of InString's you'd have to do.

I'd be interested in seeing how this performs against large strings.
Here's my test code:

<snip>

>
if (text.Length < length) break;

<snip>
That should be continue instead of break.