Homec4science
Diffusion invenio-infoscience (maint-2.1) Browse

R3600/invenio/legacy/bibrank/word_searcher.py
dd44fdec50bbmaint-2.1

Last Change
Raw File

word_searcher.py
View Options

## This file is part of Invenio.
## Copyright (C) 2004, 2005, 2006, 2007, 2008, 2010, 2011, 2012 CERN.
##
## Invenio is free software; you can redistribute it and/or
## modify it under the terms of the GNU General Public License as
## published by the Free Software Foundation; either version 2 of the
## License, or (at your option) any later version.
##
## Invenio is distributed in the hope that it will be useful, but
## WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
## General Public License for more details.
##
## You should have received a copy of the GNU General Public License
## along with Invenio; if not, write to the Free Software Foundation, Inc.,
## 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA.
import string
import time
import math
import re
from six import iteritems
from invenio.legacy.dbquery import run_sql, deserialize_via_marshal
from invenio.legacy.bibindex.engine_stemmer import stem
from invenio.legacy.bibindex.engine_stopwords import is_stopword
def find_similar(rank_method_code, recID, hitset, rank_limit_relevance,verbose, methods):
"""Finding terms to use for calculating similarity. Terms are taken from the recid given, returns a list of recids's and relevance,
input:
rank_method_code - the code of the method, from the name field in rnkMETHOD
recID - records to use for find similar
hitset - a list of hits for the query found by search_engine
rank_limit_relevance - show only records with a rank value above this
verbose - verbose value
output:
reclist - a list of sorted records: [[23,34], [344,24], [1,01]]
prefix - what to show before the rank value
postfix - what to show after the rank value
voutput - contains extra information, content dependent on verbose value"""
startCreate = time.time()
global voutput
voutput = ""
if verbose > 0:
voutput += "<br />Running rank method: %s, using find_similar/word_frequency in bibrank_record_sorter<br />" % rank_method_code
rank_limit_relevance = methods[rank_method_code]["default_min_relevance"]
try:
recID = int(recID)
except Exception,e :
return (None, "Warning: Error in record ID, please check that a number is given.", "", voutput)
rec_terms = run_sql("""SELECT termlist FROM %sR WHERE id_bibrec=%%s""" % methods[rank_method_code]["rnkWORD_table"][:-1], (recID,))
if not rec_terms:
return (None, "Warning: Requested record does not seem to exist.", "", voutput)
rec_terms = deserialize_via_marshal(rec_terms[0][0])
#Get all documents using terms from the selected documents
if len(rec_terms) == 0:
return (None, "Warning: Record specified has no content indexed for use with this method.", "", voutput)
else:
terms = "%s" % rec_terms.keys()
terms_recs = dict(run_sql("""SELECT term, hitlist FROM %s WHERE term IN (%s)""" % (methods[rank_method_code]["rnkWORD_table"], terms[1:len(terms) - 1])))
tf_values = {}
#Calculate all term frequencies
for (term, tf) in iteritems(rec_terms):
if len(term) >= methods[rank_method_code]["min_word_length"] and terms_recs.has_key(term) and tf[1] != 0:
tf_values[term] = int((1 + math.log(tf[0])) * tf[1]) #calculate term weigth
tf_values = tf_values.items()
tf_values.sort(lambda x, y: cmp(y[1], x[1])) #sort based on weigth
lwords = []
stime = time.time()
(recdict, rec_termcount) = ({}, {})
for (t, tf) in tf_values: #t=term, tf=term frequency
term_recs = deserialize_via_marshal(terms_recs[t])
if len(tf_values) <= methods[rank_method_code]["max_nr_words_lower"] or (len(term_recs) >= methods[rank_method_code]["min_nr_words_docs"] and (((float(len(term_recs)) / float(methods[rank_method_code]["col_size"])) <= methods[rank_method_code]["max_word_occurence"]) and ((float(len(term_recs)) / float(methods[rank_method_code]["col_size"])) >= methods[rank_method_code]["min_word_occurence"]))): #too complicated...something must be done
lwords.append((t, methods[rank_method_code]["rnkWORD_table"])) #list of terms used
(recdict, rec_termcount) = calculate_record_relevance_findsimilar((t, round(tf, 4)) , term_recs, hitset, recdict, rec_termcount, verbose, "true") #true tells the function to not calculate all unimportant terms
if len(tf_values) > methods[rank_method_code]["max_nr_words_lower"] and (len(lwords) == methods[rank_method_code]["max_nr_words_upper"] or tf < 0):
break
if len(recdict) == 0 or len(lwords) == 0:
return (None, "Could not find similar documents for this query.", "", voutput)
else: #sort if we got something to sort
(reclist, hitset) = sort_record_relevance_findsimilar(recdict, rec_termcount, hitset, rank_limit_relevance, verbose)
if verbose > 0:
voutput += "<br />Number of terms: %s<br />" % run_sql("SELECT count(id) FROM %s" % methods[rank_method_code]["rnkWORD_table"])[0][0]
voutput += "Number of terms to use for query: %s<br />" % len(lwords)
voutput += "Terms: %s<br />" % lwords
voutput += "Current number of recIDs: %s<br />" % (methods[rank_method_code]["col_size"])
voutput += "Prepare time: %s<br />" % (str(time.time() - startCreate))
voutput += "Total time used: %s<br />" % (str(time.time() - startCreate))
rank_method_stat(rank_method_code, reclist, lwords)
return (reclist[:len(reclist)], methods[rank_method_code]["prefix"], methods[rank_method_code]["postfix"], voutput)
def calculate_record_relevance_findsimilar(term, invidx, hitset, recdict, rec_termcount, verbose, quick=None):
"""Calculating the relevance of the documents based on the input, calculates only one word
term - (term, query term factor) the term and its importance in the overall search
invidx - {recid: tf, Gi: norm value} The Gi value is used as a idf value
hitset - a hitset with records that are allowed to be ranked
recdict - contains currently ranked records, is returned with new values
rec_termcount - {recid: count} the number of terms in this record that matches the query
verbose - verbose value
quick - if quick=yes only terms with a positive qtf is used, to limit the number of records to sort"""
(t, qtf) = term
if invidx.has_key("Gi"): #Gi = weigth for this term, created by bibrank_word_indexer
Gi = invidx["Gi"][1]
del invidx["Gi"]
else: #if not existing, bibrank should be run with -R
return (recdict, rec_termcount)
if not quick or (qtf >= 0 or (qtf < 0 and len(recdict) == 0)):
#Only accept records existing in the hitset received from the search engine
for (j, tf) in iteritems(invidx):
if j in hitset: #only include docs found by search_engine based on query
#calculate rank value
recdict[j] = recdict.get(j, 0) + int((1 + math.log(tf[0])) * Gi * tf[1] * qtf)
rec_termcount[j] = rec_termcount.get(j, 0) + 1 #number of terms from query in document
elif quick: #much used term, do not include all records, only use already existing ones
for (j, tf) in iteritems(recdict): #i.e: if doc contains important term, also count unimportant
if invidx.has_key(j):
tf = invidx[j]
recdict[j] = recdict[j] + int((1 + math.log(tf[0])) * Gi * tf[1] * qtf)
rec_termcount[j] = rec_termcount.get(j, 0) + 1 #number of terms from query in document
return (recdict, rec_termcount)
def sort_record_relevance_findsimilar(recdict, rec_termcount, hitset, rank_limit_relevance, verbose):
"""Sorts the dictionary and returns records with a relevance higher than the given value.
recdict - {recid: value} unsorted
rank_limit_relevance - a value > 0 usually
verbose - verbose value"""
startCreate = time.time()
voutput = ""
reclist = []
#Multiply with the number of terms of the total number of terms in the query existing in the records
for j in recdict.keys():
if recdict[j] > 0 and rec_termcount[j] > 1:
recdict[j] = math.log((recdict[j] * rec_termcount[j]))
else:
recdict[j] = 0
hitset -= recdict.keys()
#gives each record a score between 0-100
divideby = max(recdict.values())
for (j, w) in iteritems(recdict):
w = int(w * 100 / divideby)
if w >= rank_limit_relevance:
reclist.append((j, w))
#sort scores
reclist.sort(lambda x, y: cmp(x[1], y[1]))
if verbose > 0:
voutput += "Number of records sorted: %s<br />" % len(reclist)
voutput += "Sort time: %s<br />" % (str(time.time() - startCreate))
return (reclist, hitset)
def word_similarity(rank_method_code, lwords, hitset, rank_limit_relevance, verbose, methods):
"""Ranking a records containing specified words and returns a sorted list.
input:
rank_method_code - the code of the method, from the name field in rnkMETHOD
lwords - a list of words from the query
hitset - a list of hits for the query found by search_engine
rank_limit_relevance - show only records with a rank value above this
verbose - verbose value
output:
reclist - a list of sorted records: [[23,34], [344,24], [1,01]]
prefix - what to show before the rank value
postfix - what to show after the rank value
voutput - contains extra information, content dependent on verbose value"""
voutput = ""
startCreate = time.time()
if verbose > 0:
voutput += "<br />Running rank method: %s, using word_frequency function in bibrank_record_sorter<br />" % rank_method_code
lwords_old = lwords
lwords = []
#Check terms, remove non alphanumeric characters. Use both unstemmed and stemmed version of all terms.
for i in range(0, len(lwords_old)):
term = string.lower(lwords_old[i])
if not methods[rank_method_code]["stopwords"] == "True" or methods[rank_method_code]["stopwords"] and not is_stopword(term):
lwords.append((term, methods[rank_method_code]["rnkWORD_table"]))
terms = string.split(string.lower(re.sub(methods[rank_method_code]["chars_alphanumericseparators"], ' ', term)))
for term in terms:
if methods[rank_method_code].has_key("stemmer"): # stem word
term = stem(string.replace(term, ' ', ''), methods[rank_method_code]["stemmer"])
if lwords_old[i] != term: #add if stemmed word is different than original word
lwords.append((term, methods[rank_method_code]["rnkWORD_table"]))
(recdict, rec_termcount, lrecIDs_remove) = ({}, {}, {})
#For each term, if accepted, get a list of the records using the term
#calculate then relevance for each term before sorting the list of records
for (term, table) in lwords:
term_recs = run_sql("""SELECT term, hitlist FROM %s WHERE term=%%s""" % methods[rank_method_code]["rnkWORD_table"], (term,))
if term_recs: #if term exists in database, use for ranking
term_recs = deserialize_via_marshal(term_recs[0][1])
(recdict, rec_termcount) = calculate_record_relevance((term, int(term_recs["Gi"][1])) , term_recs, hitset, recdict, rec_termcount, verbose, quick=None)
del term_recs
if len(recdict) == 0 or (len(lwords) == 1 and lwords[0] == ""):
return (None, "Records not ranked. The query is not detailed enough, or not enough records found, for ranking to be possible.", "", voutput)
else: #sort if we got something to sort
(reclist, hitset) = sort_record_relevance(recdict, rec_termcount, hitset, rank_limit_relevance, verbose)
#Add any documents not ranked to the end of the list
if hitset:
lrecIDs = list(hitset) #using 2-3mb
reclist = zip(lrecIDs, [0] * len(lrecIDs)) + reclist #using 6mb
if verbose > 0:
voutput += "<br />Current number of recIDs: %s<br />" % (methods[rank_method_code]["col_size"])
voutput += "Number of terms: %s<br />" % run_sql("SELECT count(id) FROM %s" % methods[rank_method_code]["rnkWORD_table"])[0][0]
voutput += "Terms: %s<br />" % lwords
voutput += "Prepare and pre calculate time: %s<br />" % (str(time.time() - startCreate))
voutput += "Total time used: %s<br />" % (str(time.time() - startCreate))
voutput += str(reclist) + "<br />"
rank_method_stat(rank_method_code, reclist, lwords)
return (reclist, methods[rank_method_code]["prefix"], methods[rank_method_code]["postfix"], voutput)
def calculate_record_relevance(term, invidx, hitset, recdict, rec_termcount, verbose, quick=None):
"""Calculating the relevance of the documents based on the input, calculates only one word
term - (term, query term factor) the term and its importance in the overall search
invidx - {recid: tf, Gi: norm value} The Gi value is used as a idf value
hitset - a hitset with records that are allowed to be ranked
recdict - contains currently ranked records, is returned with new values
rec_termcount - {recid: count} the number of terms in this record that matches the query
verbose - verbose value
quick - if quick=yes only terms with a positive qtf is used, to limit the number of records to sort"""
(t, qtf) = term
if invidx.has_key("Gi"):#Gi = weigth for this term, created by bibrank_word_indexer
Gi = invidx["Gi"][1]
del invidx["Gi"]
else: #if not existing, bibrank should be run with -R
return (recdict, rec_termcount)
if not quick or (qtf >= 0 or (qtf < 0 and len(recdict) == 0)):
#Only accept records existing in the hitset received from the search engine
for (j, tf) in iteritems(invidx):
if j in hitset:#only include docs found by search_engine based on query
try: #calculates rank value
recdict[j] = recdict.get(j, 0) + int(math.log(tf[0] * Gi * tf[1] * qtf))
except:
return (recdict, rec_termcount)
rec_termcount[j] = rec_termcount.get(j, 0) + 1 #number of terms from query in document
elif quick: #much used term, do not include all records, only use already existing ones
for (j, tf) in iteritems(recdict): #i.e: if doc contains important term, also count unimportant
if invidx.has_key(j):
tf = invidx[j]
recdict[j] = recdict.get(j, 0) + int(math.log(tf[0] * Gi * tf[1] * qtf))
rec_termcount[j] = rec_termcount.get(j, 0) + 1 #number of terms from query in document
return (recdict, rec_termcount)
def sort_record_relevance(recdict, rec_termcount, hitset, rank_limit_relevance, verbose):
"""Sorts the dictionary and returns records with a relevance higher than the given value.
recdict - {recid: value} unsorted
rank_limit_relevance - a value > 0 usually
verbose - verbose value"""
startCreate = time.time()
voutput = ""
reclist = []
#remove all ranked documents so that unranked can be added to the end
hitset -= recdict.keys()
#gives each record a score between 0-100
divideby = max(recdict.values())
for (j, w) in iteritems(recdict):
w = int(w * 100 / divideby)
if w >= rank_limit_relevance:
reclist.append((j, w))
#sort scores
reclist.sort(lambda x, y: cmp(x[1], y[1]))
if verbose > 0:
voutput += "Number of records sorted: %s<br />" % len(reclist)
voutput += "Sort time: %s<br />" % (str(time.time() - startCreate))
return (reclist, hitset)
def rank_method_stat(rank_method_code, reclist, lwords):
"""Shows some statistics about the searchresult.
rank_method_code - name field from rnkMETHOD
reclist - a list of sorted and ranked records
lwords - the words in the query"""
voutput = ""
if len(reclist) > 20:
j = 20
else:
j = len(reclist)
voutput += "<br />Rank statistics:<br />"
for i in range(1, j + 1):
voutput += "%s,Recid:%s,Score:%s<br />" % (i,reclist[len(reclist) - i][0],reclist[len(reclist) - i][1])
for (term, table) in lwords:
term_recs = run_sql("""SELECT hitlist FROM %s WHERE term=%%s""" % table, (term,))
if term_recs:
term_recs = deserialize_via_marshal(term_recs[0][0])
if term_recs.has_key(reclist[len(reclist) - i][0]):
voutput += "%s-%s / " % (term, term_recs[reclist[len(reclist) - i][0]])
voutput += "<br />"
voutput += "<br />Score variation:<br />"
count = {}
for i in range(0, len(reclist)):
count[reclist[i][1]] = count.get(reclist[i][1], 0) + 1
i = 100
while i >= 0:
if count.has_key(i):
voutput += "%s-%s<br />" % (i, count[i])
i -= 1
#TODO: use Cython instead of psycho
c4science · Help