/
job.py
505 lines (444 loc) · 16.3 KB
/
job.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
import math
import string
from misc import *
import nltk
import MySQLdb as mdb
from nltk.corpus import wordnet as wn
from nltk.corpus import stopwords
from nltk.stem.wordnet import WordNetLemmatizer as wnl
from nltk.collocations import *
from collections import defaultdict
from lxml import html
from lxml import etree
class Job:
'''
This class contains information pertaining to a job, i.e. keywords, ads,
similar jobs, etc. It also contains many functions that can be applied to
the job data.
Functions:
calc_fd() - calculates aggregate frequency distributions
calc_all_fd() - calculates individual frequency distributions
extract_keywords(ref_m, ref_b, ref_t) - extracts keywords
calc_job_relevance(jobs) - finds similar jobs
fetch_salary(place) - fetches salary info
calc_keyword_freq(job2) - calculates keyword frequencies for job2 keywords
in self.ads
'''
def __init__(self, name, n_ad, n_key):
self.name = name
self.ads = []
self.keywords = []
self.simjobs = []
self.fd_all = []
self.statesal = {}
self.n_ad = n_ad
self.n_key = n_key
def kill(self):
self.ads = []
self.keywords = []
self.simjobs = []
self.fd_all = []
self.statesal = {}
def calc_fd(self):
"""
This function calculates the frequency distributions of unigrams, bigrams
and trigrams from all the scraped ads (combined). It also finds the bigrams
and trigrams that are likely combinations of words (i.e. those that make
sense).
Output:
fd_mono - unigram frequency distribution.
fd_bi - bigram frequency distribution.
fd_tri - trigram frequency distribution.
goodbi - bigrams that have high measure of pointwise mutual information
goodtri - trigrams that have high measure of pointwise mutual information
"""
all_mono = []
all_bi = []
all_tri = []
lmt = wnl()
combined_ads = []
#Concatenante unigrams, bigrams and trigrams from different ads together so
#that we don't need to make frequency distributions for each one (we only care
#about the collective anyway).
for ad in self.ads:
combined_ads = combined_ads+ad
btemp = nltk.bigrams(ad)
all_bi += btemp
ttemp = nltk.trigrams(ad)
all_tri += ttemp
#Lemmatize unigrams (this prevents things like cat and cats from being
#counted as different words.
mono_stem = [lmt.lemmatize(w) for w in ad]
all_mono += mono_stem
#Do some collocation analysis using pmi - pointwise mutual information.
#This measures how likely it is that a bigram/trigram of words actually make
#sense together.
bigram_measures = nltk.collocations.BigramAssocMeasures()
trigram_measures = nltk.collocations.TrigramAssocMeasures()
finder = BigramCollocationFinder.from_words(combined_ads)
#Only keep grams that occur >25 times and pmi>100
finder.apply_freq_filter(0.5*self.n_ad)
goodbi = finder.nbest(bigram_measures.pmi, 100)
finder = TrigramCollocationFinder.from_words(combined_ads)
finder.apply_freq_filter(0.5*self.n_ad)
goodtri = finder.nbest(trigram_measures.pmi, 100)
#Calculate frequency distributions.
fd_mono = nltk.FreqDist(all_mono)
fd_bi = nltk.FreqDist(all_bi)
fd_bi = KEY_TUPLE_TO_LIST(fd_bi)
fd_tri = nltk.FreqDist(all_tri)
fd_tri = KEY_TUPLE_TO_LIST(fd_tri)
goodbi = TUPLE_TO_LIST(goodbi)
goodtri = TUPLE_TO_LIST(goodtri)
return fd_mono, fd_bi, fd_tri, goodbi, goodtri
def calc_all_fd(self):
"""
This function creates a mega-frequency distribution dictionary. It takes
a list of frequency distributions (from each ad) for each job and puts all
of this info into a single dictionary keyed to job name.
Result:
self.fd_all - the mega-dictionary described above.
"""
fd_all = []
lmt = wnl()
for ad in self.ads:
btemp = nltk.bigrams(ad)
fd_btemp = nltk.FreqDist(btemp)
fd_btemp = KEY_TUPLE_TO_LIST(fd_btemp)
ttemp = nltk.trigrams(ad)
fd_ttemp = nltk.FreqDist(ttemp)
fd_ttemp = KEY_TUPLE_TO_LIST(fd_ttemp)
mono_stem = [lmt.lemmatize(w) for w in ad]
fd_mtemp = nltk.FreqDist(mono_stem)
fd_all.append( dict(fd_mtemp.items()+fd_btemp.items()+fd_ttemp.items()) )
self.fd_all = fd_all
def extract_keywords(self, rm, rb, rt):
"""
This function drives the keyword extraction.
Input:
rm - reference table unigram frequency distribution.
rb - reference table bigram frequency distribution.
rt - reference table trigram frequency distribution.
Output:
dictionary of top keywords.
"""
def calc_relevance(fd, ref, n):
"""
This function constructs the term frequencies necessary to calculate the
relevance score.
Input:
fd - frequency distribution for ads for a specific job.
ref - frequency distributions for reference ads.
n - set to 1 for unigrams, 2 to bigrams, 3 for trigrams.
Output:
Rel - dictionary of relevance scores for each keyword.
"""
C_query_tot = sum(fd.values())*1.
C_query_avg = C_query_tot/len(fd)
C_ref_tot = sum(ref.values())*1.
C_ref_avg = C_ref_tot/len(ref)
R = {}
for term, count in fd.items():
#For some reason, unigrams get split into individual characters some times, so
#set a maxsplit.
k = string.split(term,maxsplit=n)
C_query = count
f_query = C_query/C_query_avg
C_ref = ref.get(term,-1)
if(C_ref <= 10):
f_ref = 1. if(IS_WORD(term)==1) else 100000
else:
f_ref = C_ref/C_ref_avg
rel = relevance(f_query,f_ref,C_query,0.6)
R[term] = rel
return R
def relevance(fq, fr, Cq, x):
"""
This function calculates the relevance score of a keyword.
Input:
fq - term frequecy of the keyword in the ads for a specific job.
fr - term frequency of the keyword in the reference table.
Cq - number of times the keyword appears in the ads for a specific job.
x - some extra floating term.
Output:
relevance score.
"""
return (x*math.log(fq/fr)+(1.-x))*Cq
def remove_junk(tm, tb, tt, gb, gt):
"""
This code cleans the keyword list (i.e. removes similar words, etc.)
Input:
tm - dictionary of unigram relevance scores.
tb - dictionary of bigram relevance scores.
tt - dictionary of trigram relevance scores.
gb - bigrams with high pmi scores.
gt - trigrams with high pmi scores.
Output:
tm, tb and tt cleaned.
"""
#Filter out all of the bigrams/trigrams with low pmi scores (i.e. only keep the
#ones in gb and gt.
tempb = {}
for term in gb: tempb[term] = tb.get(term,0)
tb = tempb
tempt = {}
for term in gt: tempt[term] = tt.get(term,0)
tt = tempt
#Get rid of terms that contain the job name.
job_name = self.name
if(' ' in str(job_name)):
sjn = job_name.split()
for w in sjn:
tm.pop(w,None)
if(len(sjn)==2):
tb.pop(job_name,None)
elif(len(sjn)==3):
tt.pop(job_name,None)
else:
tm.pop(job_name,None)
#Set the stopwords + other words that seem to come up often but obviously make
#no sense.
sw = stopwords.words('english')+\
['yes', 'no', 'your', 'youll', 'benefits', 'go', 'river', 'amp',\
'us', 'e', 'permit','requires','work','types', 'dot', 'without',\
'plus', 'must', 'way', 'new', 'job', 'click', 'http', 'winning',\
'/', 'intended', 'youre', 'location', 'conditions', 'sized',\
'use', 'may', 'june', 'year', 'o', 'g', 'n', 'take', 'right',\
'term', 'always', 'existing', 'onto', 'youve', 'experience',\
'really', 'ensure', 'difference', 'ensures', 'v', 'years', 'onto']
monopop = []
bipop = []
tripop = []
#Remove stop words (or terms containing stop words).
for key in tm.keys():
if(key in sw):
monopop.append(key)
for key in tb.keys():
k = key.split()
for w in k:
if(w in sw):
bipop.append(key)
break
for key in tt.keys():
k = key.split()
for w in k:
if(w in sw):
tripop.append(key)
break
for p in set(monopop): tm.pop(p,None)
for p in set(bipop): tb.pop(p,None)
for p in set(tripop): tt.pop(p,None)
#Take care of trigrams with duplicating bigrams (i.e. banker residential
#brokerage and residential brokerage company, etc.): remove the one with the
#lower relevance score (or if equal, just keep one).
monopop = []
bipop = []
tripop = []
maxr = {}
for term, rel in tt.items():
k = term.split()
c1 = k[0]+' '+k[1]
c2 = k[1]+' '+k[2]
if(maxr.get(c1,0) <= rel): maxr[c1] = term
if(maxr.get(c2,0) <= rel): maxr[c2] = term
for term, rel in tt.items():
k = term.split()
c1 = k[0]+' '+k[1]
c2 = k[1]+' '+k[2]
if(maxr.get(c1,0) != term and maxr.get(c2,0) != term):
tripop.append(term)
#Do some lemmatizing on the units making up bigrams and trigrams to get rid of
#similar words or bigrams contained in trigrams (unigrams contained in bigrams).
lmt = wnl()
for term, rel in tt.items():
k = term.split()
c1 = k[0]+' '+k[1]
c2 = k[1]+' '+k[2]
if(math.fabs(rel-tb.get(c1,1e5)) < 140. and rel > 70.): bipop.append(c1)
if(math.fabs(rel-tb.get(c2,1e5)) < 140. and rel > 70.): bipop.append(c2)
if(c1 == job_name or c2 == job_name): tripop.append(term)
kl = [lmt.lemmatize(w) for w in k]
if( (math.fabs(rel-tm.get(kl[0],rel))<140. or \
math.fabs(rel-tm.get(kl[1],rel))<140. or \
math.fabs(rel-tm.get(kl[2],rel))<140.) and rel > 34. ):
monopop += kl
for term, rel in tb.items():
k = term.split()
kl = [lmt.lemmatize(w) for w in k]
if( (math.fabs(rel-tm.get(kl[0],rel))<70. or \
math.fabs(rel-tm.get(kl[1],rel))<70.) and rel > 17. ):
monopop += kl
#Do some stemming on unigrams to get rid of similar words.
stemmer = nltk.PorterStemmer()
for t1, r1 in tm.items():
if(r1 < 17.): continue
st1 = stemmer.stem(t1)
for t2, r2 in tm.items():
if(t1 == t2 or r2 < 17.): continue
st2 = stemmer.stem(t2)
if(r1 < r2): small = t1
else: small = t2
if( (st1 == st2) or (st1 == st2+"e") or (st1+"e" == st2) or \
(st1[:-1] == st2) or (st1 == st2[:-1]) or \
(st1[:-1] == st2[:-1]) ):
monopop.append(small)
for p in set(monopop): tm.pop(p,None)
for p in set(bipop): tb.pop(p,None)
for p in set(tripop): tt.pop(p,None)
return tm, tb, tt
fd_mono, fd_bi, fd_tri, gbi, gtri = self.calc_fd()
R_mono = calc_relevance(fd_mono,rm,1)
R_bi = calc_relevance(fd_bi,rb,2)
R_tri = calc_relevance(fd_tri,rt,3)
top_mono, top_bi, top_tri = remove_junk(R_mono,R_bi,R_tri,gbi,gtri)
top_tri = {}
top = dict(top_mono.items()+top_bi.items()+top_tri.items())
top_words = []
for term, rel in sorted(top.iteritems(),key=lambda(w,f): (f,w)):
top_words.append( (term,rel) )
top_words = top_words[::-1][:self.n_key]
return top_words
def calc_job_relevance(self,jobs):
"""
This function calculates the similarity score based on the Jaccard distance
between the query job and all other jobs.
Input:
jobs - list of Job objects that you want to find sim jobs for.
Output:
dictionary of similar scores for each job (relative to query job q).
"""
def fetch_keyword_counts(q_top, t_top, fd):
"""
This function counts the occurrences of keywords in job ads to obtain
numbers for the Jaccard distance calculation.
Input:
q_top - top keywords from job 1.
t_top - top keywords from job 2.
fd - list of frequency distributions for each ad (job1 + job2).
Output:
# of ads (job1+job2) where a keyword for job1 occurred.
# of ads (job1+job2) where a keyword for job2 occurred.
# of ads (job1+job2) where pairs of keywords (job1,job2) occur.
"""
n = len(q_top)
qkeyword_count = defaultdict(int)
qkeyword_count.fromkeys(q_top,0)
tkeyword_count = defaultdict(int)
tkeyword_count.fromkeys(t_top,0)
qtkeyword_count = defaultdict(int)
for w1 in q_top:
for w2 in t_top:
qtkeyword_count[(w1,w2)] = 0
for ad in fd:
for w1 in q_top:
v1 = ad.get(w1,0)
if(v1!=0): qkeyword_count[w1] += 1
for w2 in t_top:
v2 = ad.get(w2,0)
if(v2!=0): tkeyword_count[w2] += 1
for w1 in q_top:
v1 = ad.get(w1,0)
for w2 in t_top:
v2 = ad.get(w2,0)
if(v1!=0 and v2!=0): qtkeyword_count[(w1,w2)] += 1
return qkeyword_count, tkeyword_count, qtkeyword_count
def calc_Jaccard(cq, ct, cqt):
"""
This function calculates the Jaccard distance between the keywords for
2 different jobs.
Input:
cq - # of ads (job1+job2) where a keyword for job1 occurred.
ct - # of ads (job1+job2) where a keyword for job2 occurred.
cqt - # of ads (job1+job2) where pairs of keywords (job1,job2) occur.
Output:
Jaccard distance = # where pair occurred / (total # where either
occurred)
"""
n = len(cq)
Jaccard = {}
for kq, vq in cq.items():
for kt, vt in ct.items():
Jaccard[(kq,kt)] = cqt[(kq,kt)]*1./(vq+vt-cqt[(kq,kt)])
return Jaccard
qkeywords = self.keywords
qfd_all = self.fd_all
nk = len(qkeywords)
nd = nk*nk*1.
avgJaccard = {}
for job in jobs:
tkeywords = job.keywords
tfd_all = job.fd_all
cq, ct, cqt = fetch_keyword_counts(qkeywords,tkeywords,qfd_all+tfd_all)
J = calc_Jaccard(cq, ct, cqt)
avgJ = 0.
skill_pair = J.keys()
for skills in skill_pair: avgJ += J[skills]
avgJ = avgJ/nd
avgJaccard[job.name] = avgJ
return avgJaccard
def fetch_salary(self,place):
"""
This code fetches salary information for a job and place.
Input:
place - location.
Output:
salary - the salary (as a string, i.e. with $ and ,).
"""
job_name = self.name.split()
job_name = '+'.join(job_name)
city_name = place.split()
city_name = '+'.join(city_name)
home = "http://www.indeed.com/salary?q1=\""
page = OPEN_URL(home+job_name+"\"&l1="+city_name)
h = html.document_fromstring(page)
meat = h.xpath('.//span[@style="display:block;"]')[0]
text = etree.tostring(meat)
for i in range(0,len(text)):
if(text[i] == "$"): bdex = i
if(text[i] == "<" and text[i+1] == "/"): edex = i
salary = text[bdex:edex]
if(salary == ''): salary = 0
return salary
def calc_keyword_freq(self, job2):
"""
This function calculates which of job2's keywords are already known by
someone with job1 and which skills they should work on to transition to
job2.
Input:
job2 - list of Job objects for similar job(s)
Output:
dictionary of tuple(occurrence rates for each keyword, 0 or 2), where 0
indicates a common keyword and 2 indicates that it is more applicable to
job2.
"""
kw = job2.keywords
q1_fd = self.fd_all
q2_fd = job2.fd_all
len1 = len(q1_fd)*1.
cdict1 = {}
len2 = len(q2_fd)*1.
cdict2 = {}
#First calculate how often a keyword from job2 occurs in job ads from job1
#as well as how often it occurs in job ads from job2.
for k in kw:
count = 0
for fd in q1_fd:
if(fd.get(k,0)!=0): count+=1
cdict1[k] = count/len1
count=0
for fd in q2_fd:
if(fd.get(k,0)!=0): count+=1
cdict2[k] = count/len2
#If the keyword occurs in more than 40% of job1 ads, then people in job1
#probably already know it. So it is "common" to both jobs (these keywords get
#the 0 label). Otherwise, the keywords get the 2 label. The occurrence rate
#of the keyword for job2 is recorded to size the circles on the results.html
#page.
cdict = {}
ndiff = 0
for k in kw:
if(cdict1[k] > 0.4): cdict[k] = [cdict2[k],0]
else:
cdict[k] = [cdict2[k],2]
ndiff += 1
return cdict, ndiff