Inverted index: Difference between revisions
(added ocaml) |
mNo edit summary |
||
Line 248: | Line 248: | ||
let words_of_file in_file = |
let words_of_file in_file = |
||
let str = load_file in_file in |
let str = load_file in_file in |
||
let words = Str.split (Str.regexp " |
let words = Str.split (Str.regexp "\b") str in |
||
let words = uniq words in |
let words = uniq words in |
||
(words) |
(words) |
Revision as of 20:30, 14 August 2010
You are encouraged to solve this task according to the task description, using any language you may know.
An Inverted Index is a data structure used to create full text search.
Given a set of text files, implement a program to create an inverted index. Also create a user interface to do a search using that inverted index which returns a list of files that contain the query term / terms. The search index can be in memory.
AutoHotkey
<lang AutoHotkey>; http://www.autohotkey.com/forum/viewtopic.php?t=41479 inputbox, files, files, file pattern such as c:\files\*.txt
word2docs := object() ; autohotkey_L is needed.
stime := A_tickcount Loop, %files%, 0,1 {
tooltip,%A_index% / 500 wordList := WordsIn(A_LoopFileFullPath) InvertedIndex(wordList, A_loopFileFullpath)
}
tooltip msgbox, % "total time " (A_tickcount-stime)/1000
gosub, search return
search: Loop {
InputBox, keyword , input single keyword only msgbox, % foundDocs := findword(keyword)
} return
WordsIn(docpath) {
FileRead, content, %docpath% spos = 1 Loop { if !(spos := Regexmatch(content, "[a-zA-Z]{2,}",match, spos)) break spos += strlen(match) this_wordList .= match "`n" } Sort, this_wordList, U return this_wordList
}
InvertedIndex(byref words, docpath) {
global word2docs
loop, parse, words, `n,`r { if A_loopField = continue word2docs[A_loopField] := word2docs[A_loopField] docpath "`n" }
}
findWord(word2find) {
global word2docs
if (word2docs[word2find] = "") return "" else return word2docs[word2find]
}</lang>
Factor
<lang factor>USING: assocs fry io.encodings.utf8 io.files kernel sequences sets splitting vectors ; IN: rosettacode.inverted-index
- file-words ( file -- assoc )
utf8 file-contents " ,;:!?.()[]{}\n\r" split harvest ;
- add-to-file-list ( files file -- files )
over [ swap [ adjoin ] keep ] [ nip 1vector ] if ;
- add-to-index ( words index file -- )
'[ _ [ _ add-to-file-list ] change-at ] each ;
- (index-files) ( files index -- )
[ [ [ file-words ] keep ] dip swap add-to-index ] curry each ;
- index-files ( files -- index )
H{ } clone [ (index-files) ] keep ;
- query ( terms index -- files )
[ at ] curry map [ ] [ intersect ] map-reduce ;
</lang>
Example use : <lang>( scratchpad ) { "f1" "f2" "f3" } index-files
--- Data stack: H{ { "a" ~vector~ } { "is" ~vector~ } { "what" ~vector~ } { ... ( scratchpad ) { "what" "is" "it" } swap query . V{ "f1" "f2" } </lang>
Haskell
<lang haskell>import Control.Monad import Data.Char (isAlpha, toLower) import qualified Data.Map as M import qualified Data.IntSet as S import System.Environment (getArgs)
main = do
(files, _ : q) <- liftM (break (== "--")) getArgs buildII files >>= mapM_ putStrLn . queryII q
data IIndex = IIndex
[FilePath] -- Files in the index (M.Map String S.IntSet) -- Maps word to indices of the list deriving Show
buildII :: [FilePath] -> IO IIndex buildII files =
liftM (IIndex files . foldl f M.empty . zip [0..]) $ mapM readFile files where f m (i, s) = foldl g m $ map (lowercase . filter isAlpha) $ words s where g m word = M.insertWith S.union word (S.singleton i) m
queryII :: [String] -> IIndex -> [FilePath] queryII q (IIndex files m) =
map (files !!) $ S.toList $ intersections $ map (\word -> M.findWithDefault S.empty (lowercase word) m) q
intersections [] = S.empty intersections xs = foldl1 S.intersection xs
lowercase = map toLower</lang>
An example of use, assuming the program is named iindex
and there exist files t0
, t1
, and t2
with contents "It is what it is.", "What is it?", and "It is a banana.":
$ iindex t0 t1 t2 -- what is it t0 t1
J
This just implements the required spec, with a simplistic definition for what a word is, and with no support for stop words, nor for phrase searching.
<lang J>require'files regex strings'
rxutf8 0 NB. support latin1 searches for this example, instead of utf8 files=:words=:buckets=: wordre=: rxcomp '[\w]+' parse=: ,@:rxfrom~ wordre&rxmatches
invert=: verb define
files=: files,todo=. ~.y-.files >invert1 each todo
)
invert1=: verb define
file=. files i.<y words=: ~.words,contents=. ~.parse tolower fread jpath y ind=. words i. contents buckets=: buckets,(1+words -&# buckets)#a: #buckets=: (file,~each ind{buckets) ind}buckets
)
search=: verb define
hits=. buckets{~words i.~.parse tolower y files {~ >([-.-.)each/hits
)</lang>
Example use:
<lang J> invert '~help/primer/cut.htm';'~help/primer/end.htm';'~help/primer/gui.htm'
>search 'finally learning'
~help/primer/end.htm ~help/primer/gui.htm
>search 'argument'
~help/primer/cut.htm ~help/primer/gui.htm
>search 'around'
~help/primer/gui.htm</lang>
OCaml
We store the inverted index data in the file "data.inv" using the sexplib library, so we compile with:
ocamlc -c \ -pp "camlp4o -I `ocamlc -where`/type-conv \ -I `ocamlc -where`/sexplib \ pa_type_conv.cmo pa_sexp_conv.cmo" \ unix.cma bigarray.cma nums.cma -I +sexplib sexplib.cma str.cma \ inv.ml ocamlc -o inv.byte unix.cma bigarray.cma nums.cma -I +sexplib sexplib.cma str.cma inv.cmo
<lang ocaml>TYPE_CONV_PATH "Inverted_index"
type files = string array with sexp type inverted_index = (string * int list) list with sexp
type t = files * inverted_index with sexp
open Sexplib
let data_file = "data.inv" let data_path = Filename.concat Filename.temp_dir_name data_file
let get_inv_index() =
if not(Sys.file_exists data_path) then ([| |], []) else let files, se = t_of_sexp(Sexp.load_sexp data_path) in (files, se)
let load_file f =
let ic = open_in f in let n = in_channel_length ic in let s = String.create n in really_input ic s 0 n; close_in ic; (s)
let array_push ar v =
let len = Array.length ar in Array.init (succ len) (fun i -> if i < len then Array.unsafe_get ar i else v), len
let uniq lst =
let h = Hashtbl.create (List.length lst) in List.iter (fun x -> Hashtbl.replace h x ()) lst; Hashtbl.fold (fun x () xs -> x :: xs) h []
let combine words i inv_index =
let h = Hashtbl.create (List.length inv_index) in List.iter (fun (w, from) -> Hashtbl.replace h w from) inv_index; List.iter (fun w -> if Hashtbl.mem h w then begin let from = Hashtbl.find h w in Hashtbl.replace h w (i::from) end else Hashtbl.add h w [i] ) words; Hashtbl.fold (fun w from acc -> (w, from) :: acc) h []
let words_of_file in_file =
let str = load_file in_file in let words = Str.split (Str.regexp "\b") str in let words = uniq words in (words)
let index_file in_file =
let words = words_of_file in_file in let files, inv_index = get_inv_index() in let files, i = array_push files in_file in let inv_index = combine words i inv_index in let se = sexp_of_t (files, inv_index) in Sexp.save data_path se
let search_word word =
let files, inv_index = get_inv_index() in try let is_in = List.assoc word inv_index in List.iter (fun i -> print_endline files.(i)) is_in with Not_found -> print_endline "# Not Found"
let usage() =
Printf.printf "Usage: %s \ --index-file <file.txt> / \ --search-word <word>\n%!" Sys.argv.(0); exit 1
let () =
let cmd, arg = try (Sys.argv.(1), Sys.argv.(2)) with _ -> usage() in match cmd, arg with | "--index-file", in_file -> index_file in_file | "--search-word", word -> search_word word | _ -> usage()</lang>
Perl
<lang perl>use Set::Object 'set';
- given an array of files, returns the index
sub createindex {
my @files = @_;
my %iindex;
foreach my $file (@files) {
open(F, "<", $file) or die "Can't read file $file: $!"; while(<F>) {
s/\A\W+//;
foreach my $w (map {lc} grep {length() >= 3} split /\W+/) { if ( exists($iindex{$w}) ) { $iindex{$w}->insert($file); } else { $iindex{$w} = set($file); } } } close(F);
} return %iindex;
}
- given an index, search for words
sub search_words_with_index {
my %idx = %{shift()}; my @words = @_; my $res = set(); foreach my $w (map {lc} @words) {
$w =~ s/\W+//g; # strip non-words chars
length $w < 3 and next; exists $idx{$w} or return set(); $res = $res->is_null ? set(@{$idx{$w}}) : $res * $idx{$w}; # set intersection } return @$res;
}
- TESTING
- USAGE: invidx.pl the,list,of,words file1 file2 .. fileN
my @searchwords = split /,/, shift;
- first arg is a comma-separated list of words to search for
print "$_\n"
foreach search_words_with_index({createindex(@ARGV)}, @searchwords);</lang>
PicoLisp
Assuming three files "file1", "file2" and "file3":
$ cat file1 it is what it is $ cat file2 what is it $ cat file3 it is a banana
we can read them into a binary tree in the global variable '*MyIndex' <lang PicoLisp>(off *MyIndex)
(use Word
(for File '("file1" "file2" "file3") (in File (while (skip) (if (idx '*MyIndex (setq Word (till " ^I^J^M" T)) T) (push1 (car @) File) (set Word (cons File)) ) ) ) ) )
(de searchFor @
(apply sect (extract '((Word) (val (car (idx '*MyIndex Word)))) (rest) ) ) )</lang>
Output:
: (searchFor "what" "is" "it") -> ("file2" "file1") : (searchFor "a" "banana") -> ("file3") : (searchFor "it" "is") -> ("file3" "file2" "file1")
Python
Simple inverted index
First the simple inverted index from here together with an implementation of a search for (multiple) terms from that index. <lang python> This implements: http://en.wikipedia.org/wiki/Inverted_index of 28/07/10
from pprint import pprint as pp from glob import glob try: reduce except: from functools import reduce try: raw_input except: raw_input = input
def parsetexts(fileglob='InvertedIndex/T*.txt'):
texts, words = {}, set() for txtfile in glob(fileglob): with open(txtfile, 'r') as f: txt = f.read().split() words |= set(txt) texts[txtfile.split('\\')[-1]] = txt return texts, words
def termsearch(terms): # Searches simple inverted index
return reduce(set.intersection, (invindex[term] for term in terms), set(texts.keys()))
texts, words = parsetexts() print('\nTexts') pp(texts) print('\nWords') pp(sorted(words))
invindex = {word:set(txt
for txt, wrds in texts.items() if word in wrds) for word in words}
print('\nInverted Index') pp({k:sorted(v) for k,v in invindex.items()})
terms = ["what", "is", "it"] print('\nTerm Search for: ' + repr(terms)) pp(sorted(termsearch(terms)))</lang>
Sample Output
Texts {'T0.txt': ['it', 'is', 'what', 'it', 'is'], 'T1.txt': ['what', 'is', 'it'], 'T2.txt': ['it', 'is', 'a', 'banana']} Words ['a', 'banana', 'is', 'it', 'what'] Inverted Index {'a': ['T2.txt'], 'banana': ['T2.txt'], 'is': ['T0.txt', 'T1.txt', 'T2.txt'], 'it': ['T0.txt', 'T1.txt', 'T2.txt'], 'what': ['T0.txt', 'T1.txt']} Term Search for: ['what', 'is', 'it'] ['T0.txt', 'T1.txt']
Full inverted index
There is a re-write of the termsearch
function to work off this type of index, as well as a new phrasesearch
function
The phrasesearch
function will return multiple matches in a text, and goes on to show how this can be used to pick the text with most matches.
It is assumed that the following code is added to the end of the code for the simple case above and so shares its file opening and parsing results <lang python>from collections import Counter
def termsearch(terms): # Searches full inverted index
if not set(terms).issubset(words): return set() return reduce(set.intersection, (set(x[0] for x in txtindx) for term, txtindx in finvindex.items() if term in terms), set(texts.keys()) )
def phrasesearch(phrase):
wordsinphrase = phrase.strip().strip('"').split() if not set(wordsinphrase).issubset(words): return set() #firstword, *otherwords = wordsinphrase # Only Python 3 firstword, otherwords = wordsinphrase[0], wordsinphrase[1:] found = [] for txt in termsearch(wordsinphrase): # Possible text files for firstindx in (indx for t,indx in finvindex[firstword] if t == txt): # Over all positions of the first word of the phrase in this txt if all( (txt, firstindx+1 + otherindx) in finvindex[otherword] for otherindx, otherword in enumerate(otherwords) ): found.append(txt) return found
finvindex = {word:set((txt, wrdindx)
for txt, wrds in texts.items() for wrdindx in (i for i,w in enumerate(wrds) if word==w) if word in wrds) for word in words}
print('\nFull Inverted Index') pp({k:sorted(v) for k,v in finvindex.items()})
print('\nTerm Search on full inverted index for: ' + repr(terms)) pp(sorted(termsearch(terms)))
phrase = '"what is it"' print('\nPhrase Search for: ' + phrase) print(phrasesearch(phrase))
- Show multiple match capability
phrase = '"it is"' print('\nPhrase Search for: ' + phrase) ans = phrasesearch(phrase) print(ans) ans = Counter(ans) print(' The phrase is found most commonly in text: ' + repr(ans.most_common(1)[0][0]))</lang>
Sample Output
Full Inverted Index {'a': [('T2.txt', 2)], 'banana': [('T2.txt', 3)], 'is': [('T0.txt', 1), ('T0.txt', 4), ('T1.txt', 1), ('T2.txt', 1)], 'it': [('T0.txt', 0), ('T0.txt', 3), ('T1.txt', 2), ('T2.txt', 0)], 'what': [('T0.txt', 2), ('T1.txt', 0)]} Term Search on full inverted index for: ['what', 'is', 'it'] ['T0.txt', 'T1.txt'] Phrase Search for: "what is it" ['T1.txt'] Phrase Search for: "it is" ['T0.txt', 'T0.txt', 'T2.txt'] The phrase is found most commonly in text: 'T0.txt'
Tcl
<lang tcl>package require Tcl 8.5 proc wordsInString str {
# We define "words" to be "maximal sequences of 'word' characters". # The other possible definition is to use 'non-space' characters. regexp -all -inline {\w+} $str
}
- Adds a document to the index. The index is a map from words to a map
- from filenames to lists of word locations.
proc addDocumentToIndex {filename} {
global index set f [open $filename] set data [read $f] close $f
set i 0 array set localidx {} foreach word [wordsInString $data] {
lappend localidx($word) $i incr i
}
# Transcribe into global index foreach {word places} [array get localidx] {
dict set index($word) $filename $places
}
}
- How to use the index to find files containing a word
proc findFilesForWord {word} {
global index if {[info exists index($word)]} {
return [dict keys $index($word)]
}
}
- How to use the index to find files containing all words from a list.
- Note that this does not use the locations within the file.
proc findFilesWithAllWords {words} {
set files [findFilesForWord [lindex $words 0]] foreach w [lrange $words 1 end] {
set wf [findFilesForWord $w] set newfiles {} foreach f $files { if {$f in $wf} {lappend newfiles $f} } set files $newfiles
} return $files
}
- How to use the index to find a sequence of words in a file.
proc findFilesWithWordSequence {words} {
global index set files {} foreach w $words {
if {![info exist index($w)]} { return }
} dict for {file places} $index([lindex $words 0]) {
if {$file in $files} continue foreach start $places { set gotStart 1 foreach w [lrange $words 1 end] { incr start set gotNext 0 foreach {f ps} $index($w) { if {$f ne $file} continue foreach p $ps { if {$p == $start} { set gotNext 1 break } } if {$gotNext} break } if {!$gotNext} { set gotStart 0 break } } if {$gotStart} { lappend files $file break } }
} return $files
}</lang> For the GUI: <lang tcl>package require Tk pack [labelframe .files -text Files] -side left -fill y pack [listbox .files.list -listvariable files] pack [button .files.add -command AddFile -text "Add File to Index"] pack [labelframe .found -text Found] -side right -fill y pack [listbox .found.list -listvariable found] -fill x pack [entry .found.entry -textvariable terms] -fill x pack [button .found.findAll -command FindAll \ -text "Find File with All"] -side left pack [button .found.findSeq -command FindSeq \ -text "Find File with Sequence"] -side right
- The actions invoked by various GUI buttons
proc AddFile {} {
global files set f [tk_getOpenFile] if {$f ne ""} {
addDocumentToIndex $f lappend files $f
}
} proc FindAll {} {
global found terms set words [wordsInString $terms] set fs [findFilesWithAllWords $words] lappend found "Searching for files with all $terms" {*}$fs \
"---------------------" } proc FindSeq {} {
global found terms set words [wordsInString $terms] set fs [findFilesWithWordSequence $words] lappend found "Searching for files with \"$terms\"" {*}$fs \
"---------------------" }</lang>