def show_layers_effect(query_set, searcher, layers_combs):
'''
Shows performance for different layers configuration.
'''
queries = load_query_set(query_set, 40)
legend = {'P': 'papers_relev',
'A': 'authors_relev',
'K': 'words_relev',
'V': 'venues_relev'}
params = {'age_relev': 0.0,
'query_relev': 0.0,
'ctx_relev': 0.0}
for layers in layers_combs :
print "%s\t" % layers,
for layer in legend:
params[legend[layer]] = 1.0 if (layer in layers) else 0.0
searcher.set_params(**params)
get_search_metrics(queries, searcher, show=True, force=True,
results_file=("%s/results/layers_effect/%s.p" % (config.DATA, layers)))
print
def save_scholar_results(query_set, n):
'''
For each given query, request on google scholar, search for textually similar
entries on the index and show them to the user for confirmation.
'''
from scholar_api import ScholarQuerier, SearchScholarQuery
queries = load_query_set(query_set)
querier = ScholarQuerier()
scholar_query = SearchScholarQuery()
# Folder to store saved results
# from_folder = config.QUERY_SETS_PATH + "manual"
save_folder = "%s/scholar/%s" % (config.DATA, config.DATASET)
for query, query_id, _, _, _, _ in queries:
# Only searches if file doesn't already exists
file_path = "%s/%s.txt" % (save_folder, query_id)
if os.path.exists(file_path):
continue
time.sleep(1)
print "\nProcessing '%s'" % query,
scholar_query.set_words(query)
# We stop requesting new pages once we found at least n
start = 0
titles = []
while (len(titles) < n):
scholar_query.set_start(start)
querier.send_query(scholar_query)
# Check if we got a captcha as response
if (len(querier.articles) == 0):
if (start == 0):
raise Exception(
"Request probably got blocked due to overload.")
else:
# If we got 0 article after some requests it may be that
# all articles were fetched. Skip to next query.
break
# Get only titles and try to find entries in our dataset for them
for article in querier.articles:
titles.append(article['title'].strip('. '))
# Set correct pagination
start += 20
# Write to file
with open(file_path, 'a') as file:
print >> file, query
for title in titles:
print >> file, "%s" % (title.encode("UTF-8"))
def save_layers_results_query_set(query_set) : queries = load_query_set(query_set, 10) queries = [query for query,_,_,_,_,_ in queries] layers = ['paper', 'author', 'venue', 'ngram'] searcher = Searcher(**PARAMS) for layer in layers : folder = "%s/results/layers/%s/%s/" % (config.DATA, config.DATASET, layer) get_layer_results(queries, searcher, folder, layer)
def save_aminer_results(query_set):
queries = load_query_set(query_set, limit=100)
for query, pub_id, _, _, _, _ in queries:
# folder = config.DATA + "aminer/" + query_set
folder = "%s/aminer/%s" % (config.DATA, config.DATASET)
if (not os.path.exists(folder)):
os.makedirs(folder)
# Only searches if file doesn't already exists
file_path = "%s/%s.txt" % (folder, pub_id)
if os.path.exists(file_path):
continue
titles = search_aminer(query, 20)
print "%d\t%s" % (len(titles), query)
with open(file_path, 'w') as file:
print >> file, query
for title in titles:
print >> file, "%s" % (title.encode("UTF-8"))
def main() :
# for qs in ["manual", "surveys", "testing"] :
# print "\n%s" % qs
# time_diversity(["MultiLayered", "TopCited(G)"], qs)
# sys.exit()
# config.IN_MODELS_FOLDER = config.DATA + "topic_models/%s_%d_%d"
## check_topics_effect(Searcher(**PARAMS), "manual")
# show_layers_effect("manual", Searcher(**PARAMS), ["PAK", "PAT", "PAKT"][:1])
# save_layers_results_queries(["citation recommendation",
# "author recommendation",
# "link prediction"],
# folder="/var/tmp/results")
# save_layer_results("manual")
# query_set = 'manual'
# query_set = 'surveys'
# query_set = 'tuning'
# query_set = 'testing'
# queries = load_query_set(query_set, 200)
# vary_parameters(Searcher(**PARAMS), queries, 'age_relev', [0.0, 0.001, 0.01, 0.1, 0.25, 0.5, 1.0])
# vary_parameters(Searcher(**PARAMS), queries, 'ctx_relev', [0.0, 0.1, 0.25, 0.5, 1.0, 2.0, 5.0, 10.0, 20.0])
# vary_parameters(Searcher(**PARAMS), queries, 'query_relev', [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.8])
# layers = ["P","PA","PV","PK","PAV","PAK","PKV","PAKV"]
# show_layers_effect(queries, Searcher(**BEST_PARAMS), layers)
# show_attenuators_effect(queries, Searcher(**PARAMS))
# vary_rho_values(Searcher(**PARAMS), queries, 'papers_relev', [0.05, 0.1, 0.25, 0.5, 0.75, 0.85, 0.9, 1.0])
# vary_rho_values(Searcher(**PARAMS), queries, 'authors_relev', [0.0, 0.1, 0.25, 0.5, 0.75, 0.9])
# vary_rho_values(Searcher(**PARAMS), queries, 'topics_relev', [0.0, 0.1, 0.25, 0.5, 0.75, 0.9])
# vary_rho_values(Searcher(**PARAMS), queries, 'words_relev', [0.0, 0.1, 0.25, 0.5, 0.75, 0.9])
# vary_rho_values(Searcher(**BEST_PARAMS), queries, 'venues_relev', [0.0, 0.05, 0.1, 0.25, 0.5])
# get_other_layers(load_query_set('manual', 10), Searcher(**BEST_PARAMS), "ngram")
query_sets = [
# 'manual',
# 'surveys',
# 'tuning',
'testing'
]
searchers = [
Searcher(**PARAMS),
# Searcher(**config.PARAMS),
# PageRankSubgraphSearcher(**PARAMS),
# TopCitedSubgraphSearcher(**PARAMS),
# TopCitedGlobalSearcher(),
# TFIDFSearcher(),
# BM25Searcher(),
# CiteRankSearcher(tau=2.6),
# PageRankFilterBeforeSearcher(),
# PageRankFilterAfterSearcher(),
# GoogleScholarSearcher(),
# ArnetMinerSearcher(),
#MengSearcher(),
# CiteseerSearcher("eval/citeseer"),
# WeightedTopCitedSubgraphSearcher(**PARAMS)
]
for query_set in query_sets :
log.info("Running '%s' query set.\n" % query_set)
queries = load_query_set(query_set, 200)
for s in searchers :
print "%s\t" % s.name(),
# print "\nRunning %s with %d queries from %s set..." % \
# (s.name(), len(queries), query_set)
if s.name() == "MultiLayered":
s.set_params(**{
'K': 20,
'H': 1,
'papers_relev': 0.25,
'authors_relev': 0.25,
'words_relev': 0.25,
'topics_relev' : 0.0,
'venues_relev': 0.25,
'alpha': 0.3,
'query_relev': 0.3,
'age_relev': 0.01,
'ctx_relev': 0.5})
if s.name() == "TopCited(G)": # TopCitedSubgraphSearcher
s.set_params(**{
'K': 20,
'H': 1,
'papers_relev': 0.25,
'authors_relev': 0.25,
'words_relev': 0.25,
'topics_relev' : 0.0,
'venues_relev': 0.25,
'alpha': 0.3,
'query_relev': 0.3,
'age_relev': 0.01,
'ctx_relev': 0.5})
if s.name() == "WeightedTopCited(G)":
s.set_params(**{
'K': 20,
'H': 1,
'query_relev': 0.15, # 0.15
'age_relev': 0.01, # 0.01
'ctx_relev': 0.8, # 0.6 (manual), 0.8
'beta': 0.1}) # 0.1
rfile = get_results_file(query_set, s.name())
get_search_metrics(queries, s, force=True, results_file=rfile)
del s
def check_topics_effect(searcher, query_set) : queries = load_query_set(query_set, 10) get_search_metrics(queries, searcher, force=True)
maps = []
for query, pub_id, _year, actual_docs, _rels, _titles in queries:
top = self.searcher.search(query,
limit=20,
exclude=set([pub_id]),
force=False)
maps.append(apk(actual_docs, top, k=20))
return np.mean(maps)
if __name__ == '__main__':
query_set = 'manual'
queries = load_query_set(query_set, 30)
se = SearchEvaluator(Searcher(**config.PARAMS))
bocv = BayesianOptCV(
estimator=se,
param_bounds={
'K': (5, 50),
'papers_relev': (0.001, 1.0),
'authors_relev': (0.001, 1.0),
'venues_relev': (0.001, 1.0),
'words_relev': (0.001, 1.0),
'alpha': (0.1, 0.9),
'age_relev': (0.001, 1.0),
'query_relev': (0.001, 1.0),
'ctx_relev': (0.01, 10.0)
},