def analyze_article_citations(self, num_of_articles=20, quiet=False):
"""Prints and returns a list of the top 20 most important articles in the
TFEU treaty, as determined by the number of citing cases."""
# Create a mapping of article equivalencies, eg Art 28 TEC == Art 34 TFEU
sameas = self._sameas()
equivs = {}
pred = util.ns['owl'] + "sameAs"
for (s, o) in sameas.subject_objects(URIRef(pred)):
equivs[str(o)] = str(s)
self.log.debug("Defined %s equivalent article references" %
len(equivs))
# Select unique articles citings
store = TripleStore(self.config.storetype, self.config.storelocation,
self.config.storerepository)
sq = """PREFIX eurlex:<http://lagen.nu/eurlex#>
SELECT DISTINCT ?case ?article WHERE {
?case eurlex:cites ?article .
FILTER (regex(str(?article), "^http://lagen.nu/ext/celex/1"))
}"""
cites = store.select(sq, format="python")
citationcount = {}
unmapped = {}
self.log.debug("Going through %s unique citations" % len(cites))
for cite in cites:
article = cite['article'].split("-")[0]
if "12008M" in article:
pass
elif article in equivs:
article = equivs[article]
else:
if article in unmapped:
unmapped[article] += 1
else:
unmapped[article] = 1
article = None
# Keep track of the number of citing cases
if article:
if article in citationcount:
citationcount[article] += 1
else:
citationcount[article] = 1
# Report the most common cites to older treaty articles that
# we have no equivalents for in TFEU
# sorted_unmapped = sorted(unmapped.iteritems(), key=itemgetter(1))[-num_of_articles:]
# if not quiet:
# print "UNMAPPED:"
# pprint(sorted_unmapped)
# Report and return the most cited articles
sorted_citationcount = sorted(iter(list(citationcount.items())),
key=itemgetter(1))[-num_of_articles:]
if not quiet:
print("CITATION COUNTS:")
pprint(sorted_citationcount)
return [x[0] for x in reversed(sorted_citationcount)]
def temp_analyze(self):
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
# sq = self._query_cites('http://lagen.nu/ext/celex/12008E045',self._sameas(),False, True, 2012)
sq = self._query_cites(None, self._sameas(), False, False, 2012)
print(sq)
cites = store.select(sq, format="python")
self.log.debug(
" Citation graph contains %s citations" % (len(cites)))
# remove duplicate citations, self-citations and pinpoints
# in citations
citedict = {}
for cite in cites:
# print repr(cite)
if "-" in cite['obj']:
cite['obj'] = cite['obj'].split("-")[0]
if (cite['subj'] != cite['obj']):
citedict[(cite['subj'], cite['obj'])] = True
self.log.debug(
" Normalized graph contains %s citations" % len(citedict))
degree = {}
for citing, cited in list(citedict.keys()):
if citing not in degree:
degree[citing] = []
if cited not in degree:
degree[cited] = []
degree[cited].append(citing)
return
def temp_analyze(self):
store = TripleStore(self.config.storetype, self.config.storelocation,
self.config.storerepository)
# sq = self._query_cites('http://lagen.nu/ext/celex/12008E045',self._sameas(),False, True, 2012)
sq = self._query_cites(None, self._sameas(), False, False, 2012)
print(sq)
cites = store.select(sq, format="python")
self.log.debug(" Citation graph contains %s citations" %
(len(cites)))
# remove duplicate citations, self-citations and pinpoints
# in citations
citedict = {}
for cite in cites:
# print repr(cite)
if "-" in cite['obj']:
cite['obj'] = cite['obj'].split("-")[0]
if (cite['subj'] != cite['obj']):
citedict[(cite['subj'], cite['obj'])] = True
self.log.debug(" Normalized graph contains %s citations" %
len(citedict))
degree = {}
for citing, cited in list(citedict.keys()):
if citing not in degree:
degree[citing] = []
if cited not in degree:
degree[cited] = []
degree[cited].append(citing)
return
def download_from_triplestore(self):
sq = "SELECT ?something ?references ?uri where ?something ?references ?uri AND NOT ?uri ?references ?anything"
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
with self.store.open_downloaded("biggraph") as fp:
for row in store.select(sq):
fp.write("<%(something)s> <%(references)s> <%(uri)s> .\n")
def download_from_triplestore(self):
sq = "SELECT ?something ?references ?uri where ?something ?references ?uri AND NOT ?uri ?references ?anything"
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
with self.store.open_downloaded("biggraph") as fp:
for row in store.select(sq):
fp.write("<%(something)s> <%(references)s> <%(uri)s> .\n")
def eval_get_goldstandard(self, basefile):
goldstandard = Graph()
goldstandard_rdf = util.relpath(
os.path.dirname(__file__) + "/../res/eut/goldstandard.n3")
goldstandard.load(goldstandard_rdf, format="n3")
pred = util.ns['ir'] + 'isRelevantFor'
res = {}
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
sq_templ = """PREFIX eurlex:<http://lagen.nu/eurlex#>
SELECT ?party ?casenum ?celexnum WHERE {
<%s> eurlex:party ?party ;
eurlex:casenum ?casenum ;
eurlex:celexnum ?celexnum .
}"""
self.log.debug(
"Loading gold standard relevance judgments for %s" % basefile)
for article in self._articles(basefile):
res[article] = []
for o in goldstandard.objects(URIRef(article), URIRef(pred)):
res[article].append(str(o))
# Make sure the case exists and is the case we're looking for
sq = sq_templ % str(o)
parties = store.select(sq, format="python")
if parties:
pass
# self.log.debug(" %s: %s (%s)" %
# (parties[0]['celexnum'],
# parties[0]['casenum'],
# " v ".join([x['party'] for x in parties])))
else:
self.log.warning("Can't find %s in triple store!" % o)
self.log.debug(" Gold standard for %s: %s relevant docs" %
(article, len(res[article])))
res[article].sort()
return res
def eval_get_goldstandard(self, basefile):
goldstandard = Graph()
goldstandard_rdf = util.relpath(
os.path.dirname(__file__) + "/../res/eut/goldstandard.n3")
goldstandard.load(goldstandard_rdf, format="n3")
pred = util.ns['ir'] + 'isRelevantFor'
res = {}
store = TripleStore(self.config.storetype, self.config.storelocation,
self.config.storerepository)
sq_templ = """PREFIX eurlex:<http://lagen.nu/eurlex#>
SELECT ?party ?casenum ?celexnum WHERE {
<%s> eurlex:party ?party ;
eurlex:casenum ?casenum ;
eurlex:celexnum ?celexnum .
}"""
self.log.debug("Loading gold standard relevance judgments for %s" %
basefile)
for article in self._articles(basefile):
res[article] = []
for o in goldstandard.objects(URIRef(article), URIRef(pred)):
res[article].append(str(o))
# Make sure the case exists and is the case we're looking for
sq = sq_templ % str(o)
parties = store.select(sq, format="python")
if parties:
pass
# self.log.debug(" %s: %s (%s)" %
# (parties[0]['celexnum'],
# parties[0]['casenum'],
# " v ".join([x['party'] for x in parties])))
else:
self.log.warning("Can't find %s in triple store!" % o)
self.log.debug(" Gold standard for %s: %s relevant docs" %
(article, len(res[article])))
res[article].sort()
return res
def eval_get_ranked_set(self, basefile, algorithm="pagerank",
age_compensation=False, restrict_cited=True):
# * algorithm: can be "indegree", "hits" or "pagerank".
# * age_compensation: create one graph per year and average to
# compensate for newer cases (that have had less time to gain
# citations)
# * restrict_cited: Use only such citations that exist between
# two cases that both cite the same TFEU article (othewise,
# use all citations from all cases that cite the TFEU
# article, regardless of whether the cited case also cites
# the same TFEU article)
sameas = self._sameas()
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
res = {}
self.log.debug("Creating ranked set (%s,age_compensation=%s,restrict_cited=%s)" %
(algorithm, age_compensation, restrict_cited))
for article in self._articles(basefile):
article_celex = article.split("/")[-1]
self.log.debug(" Creating ranking for %s" % (article_celex))
this_year = datetime.datetime.today().year
if age_compensation:
years = list(range(1954, this_year + 1))
# years = range(this_year-3,this_year) # testing
else:
years = list(range(this_year, this_year + 1))
result_by_years = []
for year in years:
restrict_citing = True # always performs better
if (article, year, restrict_cited) in self._graph_cache:
# self.log.debug("Resuing cached graph (%s) for %s in %s" %
# (restrict_cited, article_celex,year))
graph = self._graph_cache[(article, year, restrict_cited)]
else:
# self.log.debug("Calculating graph for %s in %s" %
# (article_celex,year))
sq = self._query_cites(article, sameas, restrict_citing,
restrict_cited, year)
links = store.select(sq, format="python")
graph = self.eval_build_nx_graph(links)
self._graph_cache[(article, year, restrict_cited)] = graph
self.log.debug(" Citegraph for %s in %s has %s edges, %s nodes" %
(article_celex, year, len(graph.edges()),
len(graph.nodes())))
if len(graph.nodes()) == 0:
continue
ranked = self.eval_rank_graph(graph, algorithm)
result_by_years.append({})
for result, score in ranked:
result_by_years[-1][result] = score
if age_compensation:
compensated_ranking = {}
for d, score in ranked: # the last result set
# cut out the year part of the URI
celex = d.split("/")[-1]
try:
age = this_year + 1 - int(
celex[1:5]) # cases decided this year has age 1
# scores = [0,0,0 ... 3,4,8,22]
scores = [result_by_year[d]
for result_by_year
in result_by_years
if d in result_by_year]
avg_score = sum(scores) / float(age)
# self.log.debug("Result %s (age %s, avg score %s) %r" %
# (d,age,avg_score,scores))
compensated_ranking[d] = avg_score
except ValueError:
continue
# return just a list of results, no scores
if age_compensation:
res[article] = [result for result in sorted(
compensated_ranking, key=compensated_ranking.__getitem__, reverse=True)]
else:
res[article] = [result[0] for result in ranked]
return res
def analyze_citation_graphs(self, articles=None):
# Basic setup
# articles = self._articles('tfeu')[-1:]
if not articles:
articles = [None]
if None not in articles:
articles.append(None)
this_year = datetime.datetime.today().year
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
sameas = self._sameas()
distributions = []
# For each article (and also for no article = the entire citation graph)
for article in articles:
# Get a list of all eligble cases (needed for proper degree distribution)
sq = self._query_cases(article, sameas)
# print sq
cases = {}
caserows = store.select(sq, format="python")
for r in caserows:
cases[r['subj']] = 0
self.log.info(
"Creating graphs for %s (%s cases)" % (article, len(cases)))
# Step 1. SPARQL the graph on the form ?citing ?cited
# (optionally restricting on citing a particular article)
if article:
sq = self._query_cites(
article, sameas, True, False, this_year + 1)
else:
sq = self._query_cites(
None, sameas, False, False, this_year + 1)
cites = store.select(sq, format="python")
self.log.debug(
" Citation graph contains %s citations" % (len(cites)))
# remove duplicate citations, self-citations and pinpoints
# in citations
citedict = {}
missingcases = {}
for cite in cites:
# print repr(cite)
if "-" in cite['obj']:
cite['obj'] = cite['obj'].split("-")[0]
if not cite['obj'] in cases:
# print "Case %s (cited in %s) does not exist!\n" % (cite['obj'],
# cite['subj'])
missingcases[cite['obj']] = True
continue
if (cite['subj'] != cite['obj']):
citedict[(cite['subj'], cite['obj'])] = True
self.log.debug(
" Normalized graph contains %s citations (%s cited cases not found)" %
(len(citedict), len(missingcases)))
# pprint(missingcases.keys()[:10])
# Step 2. Dotify the list (maybe the direction of arrows from
# cited to citing can improve results?) to create a citation
# graph
self.analyse_citegraph_graphviz(list(citedict.keys()), article)
# Step 3. Create a degree distribution plot
degree, distribution = self.analyze_citegraph_degree_distribution(
cases, list(citedict.keys()), article)
if article:
distributions.append([article, distribution])
# Step 4. Create a citation/age scatterplot (or rather hexbin)
self.analyze_citegraph_citation_age_plot(
list(citedict.keys()), degree, distribution, article)
# Step 5. Create a combined degree distribution graph of the
# distinct citation networks. Also add the degree distribution
# of gold standard cases
self.analyze_citegraph_combined_degree_distribution(distributions)
def analyze_article_citations(self, num_of_articles=20, quiet=False):
"""Prints and returns a list of the top 20 most important articles in the
TFEU treaty, as determined by the number of citing cases."""
# Create a mapping of article equivalencies, eg Art 28 TEC == Art 34 TFEU
sameas = self._sameas()
equivs = {}
pred = util.ns['owl'] + "sameAs"
for (s, o) in sameas.subject_objects(URIRef(pred)):
equivs[str(o)] = str(s)
self.log.debug(
"Defined %s equivalent article references" % len(equivs))
# Select unique articles citings
store = TripleStore(self.config.storetype,
self.config.storelocation,
self.config.storerepository)
sq = """PREFIX eurlex:<http://lagen.nu/eurlex#>
SELECT DISTINCT ?case ?article WHERE {
?case eurlex:cites ?article .
FILTER (regex(str(?article), "^http://lagen.nu/ext/celex/1"))
}"""
cites = store.select(sq, format="python")
citationcount = {}
unmapped = {}
self.log.debug("Going through %s unique citations" % len(cites))
for cite in cites:
article = cite['article'].split("-")[0]
if "12008M" in article:
pass
elif article in equivs:
article = equivs[article]
else:
if article in unmapped:
unmapped[article] += 1
else:
unmapped[article] = 1
article = None
# Keep track of the number of citing cases
if article:
if article in citationcount:
citationcount[article] += 1
else:
citationcount[article] = 1
# Report the most common cites to older treaty articles that
# we have no equivalents for in TFEU
# sorted_unmapped = sorted(unmapped.iteritems(), key=itemgetter(1))[-num_of_articles:]
# if not quiet:
# print "UNMAPPED:"
# pprint(sorted_unmapped)
# Report and return the most cited articles
sorted_citationcount = sorted(iter(list(
citationcount.items())), key=itemgetter(1))[-num_of_articles:]
if not quiet:
print("CITATION COUNTS:")
pprint(sorted_citationcount)
return [x[0] for x in reversed(sorted_citationcount)]
def eval_get_ranked_set(self,
basefile,
algorithm="pagerank",
age_compensation=False,
restrict_cited=True):
# * algorithm: can be "indegree", "hits" or "pagerank".
# * age_compensation: create one graph per year and average to
# compensate for newer cases (that have had less time to gain
# citations)
# * restrict_cited: Use only such citations that exist between
# two cases that both cite the same TFEU article (othewise,
# use all citations from all cases that cite the TFEU
# article, regardless of whether the cited case also cites
# the same TFEU article)
sameas = self._sameas()
store = TripleStore(self.config.storetype, self.config.storelocation,
self.config.storerepository)
res = {}
self.log.debug(
"Creating ranked set (%s,age_compensation=%s,restrict_cited=%s)" %
(algorithm, age_compensation, restrict_cited))
for article in self._articles(basefile):
article_celex = article.split("/")[-1]
self.log.debug(" Creating ranking for %s" % (article_celex))
this_year = datetime.datetime.today().year
if age_compensation:
years = list(range(1954, this_year + 1))
# years = range(this_year-3,this_year) # testing
else:
years = list(range(this_year, this_year + 1))
result_by_years = []
for year in years:
restrict_citing = True # always performs better
if (article, year, restrict_cited) in self._graph_cache:
# self.log.debug("Resuing cached graph (%s) for %s in %s" %
# (restrict_cited, article_celex,year))
graph = self._graph_cache[(article, year, restrict_cited)]
else:
# self.log.debug("Calculating graph for %s in %s" %
# (article_celex,year))
sq = self._query_cites(article, sameas, restrict_citing,
restrict_cited, year)
links = store.select(sq, format="python")
graph = self.eval_build_nx_graph(links)
self._graph_cache[(article, year, restrict_cited)] = graph
self.log.debug(
" Citegraph for %s in %s has %s edges, %s nodes" %
(article_celex, year, len(
graph.edges()), len(graph.nodes())))
if len(graph.nodes()) == 0:
continue
ranked = self.eval_rank_graph(graph, algorithm)
result_by_years.append({})
for result, score in ranked:
result_by_years[-1][result] = score
if age_compensation:
compensated_ranking = {}
for d, score in ranked: # the last result set
# cut out the year part of the URI
celex = d.split("/")[-1]
try:
age = this_year + 1 - int(
celex[1:5]) # cases decided this year has age 1
# scores = [0,0,0 ... 3,4,8,22]
scores = [
result_by_year[d]
for result_by_year in result_by_years
if d in result_by_year
]
avg_score = sum(scores) / float(age)
# self.log.debug("Result %s (age %s, avg score %s) %r" %
# (d,age,avg_score,scores))
compensated_ranking[d] = avg_score
except ValueError:
continue
# return just a list of results, no scores
if age_compensation:
res[article] = [
result
for result in sorted(compensated_ranking,
key=compensated_ranking.__getitem__,
reverse=True)
]
else:
res[article] = [result[0] for result in ranked]
return res
def analyze_citation_graphs(self, articles=None):
# Basic setup
# articles = self._articles('tfeu')[-1:]
if not articles:
articles = [None]
if None not in articles:
articles.append(None)
this_year = datetime.datetime.today().year
store = TripleStore(self.config.storetype, self.config.storelocation,
self.config.storerepository)
sameas = self._sameas()
distributions = []
# For each article (and also for no article = the entire citation graph)
for article in articles:
# Get a list of all eligble cases (needed for proper degree distribution)
sq = self._query_cases(article, sameas)
# print sq
cases = {}
caserows = store.select(sq, format="python")
for r in caserows:
cases[r['subj']] = 0
self.log.info("Creating graphs for %s (%s cases)" %
(article, len(cases)))
# Step 1. SPARQL the graph on the form ?citing ?cited
# (optionally restricting on citing a particular article)
if article:
sq = self._query_cites(article, sameas, True, False,
this_year + 1)
else:
sq = self._query_cites(None, sameas, False, False,
this_year + 1)
cites = store.select(sq, format="python")
self.log.debug(" Citation graph contains %s citations" %
(len(cites)))
# remove duplicate citations, self-citations and pinpoints
# in citations
citedict = {}
missingcases = {}
for cite in cites:
# print repr(cite)
if "-" in cite['obj']:
cite['obj'] = cite['obj'].split("-")[0]
if not cite['obj'] in cases:
# print "Case %s (cited in %s) does not exist!\n" % (cite['obj'],
# cite['subj'])
missingcases[cite['obj']] = True
continue
if (cite['subj'] != cite['obj']):
citedict[(cite['subj'], cite['obj'])] = True
self.log.debug(
" Normalized graph contains %s citations (%s cited cases not found)"
% (len(citedict), len(missingcases)))
# pprint(missingcases.keys()[:10])
# Step 2. Dotify the list (maybe the direction of arrows from
# cited to citing can improve results?) to create a citation
# graph
self.analyse_citegraph_graphviz(list(citedict.keys()), article)
# Step 3. Create a degree distribution plot
degree, distribution = self.analyze_citegraph_degree_distribution(
cases, list(citedict.keys()), article)
if article:
distributions.append([article, distribution])
# Step 4. Create a citation/age scatterplot (or rather hexbin)
self.analyze_citegraph_citation_age_plot(list(citedict.keys()),
degree, distribution,
article)
# Step 5. Create a combined degree distribution graph of the
# distinct citation networks. Also add the degree distribution
# of gold standard cases
self.analyze_citegraph_combined_degree_distribution(distributions)