From f548eeedd59a1d7d99deb8864c7d11947271e426 Mon Sep 17 00:00:00 2001 From: Nathan TeBlunthuis Date: Fri, 27 Mar 2020 16:52:19 -0700 Subject: [PATCH] expand wikidata search to get keywords from google trends --- .../src/compile_transliterated_phrases.sh | 13 +++- transliterations/src/wikidata_api_calls.py | 2 +- transliterations/src/wikidata_search.py | 44 ++++++++--- .../src/wikidata_transliterations.py | 73 ++++++++++++------- 4 files changed, 92 insertions(+), 40 deletions(-) diff --git a/transliterations/src/compile_transliterated_phrases.sh b/transliterations/src/compile_transliterated_phrases.sh index 8af82f7..09f3bb5 100755 --- a/transliterations/src/compile_transliterated_phrases.sh +++ b/transliterations/src/compile_transliterated_phrases.sh @@ -2,5 +2,14 @@ # For now these scripts don't accept command line arguments. It's an MVP -python3 wikidata_search.py -python3 wikidata_transliterations.py +echo "Reading Google trends" +python3 collect_trends.py + +echo "Searching for Wikidata entities using base_terms.txt" +python3 wikidata_search.py ../data/input/base_terms.txt --output ../data/output/wikidata_search_results.csv + +echo "Searching for Wikidata entities using Google trends" +python3 wikidata_search.py ../data/output/related_searches_rising.csv ../data/output/related_searches_top.csv --use-gtrends --output ../data/output/wikidata_search_results_from_gtrends.csv + +echo "Finding transliterations from Wikidata using sparql" +python3 wikidata_transliterations.py ../data/output/wikidata_search_results_from_gtrends.csv ../data/output/wikidata_search_results.csv --topN 10 20 --output ../data/output/wikidata_entity_labels.csv diff --git a/transliterations/src/wikidata_api_calls.py b/transliterations/src/wikidata_api_calls.py index f2fac88..212ed4f 100644 --- a/transliterations/src/wikidata_api_calls.py +++ b/transliterations/src/wikidata_api_calls.py @@ -29,7 +29,7 @@ def search_wikidata(session, term, *args, **kwargs): return results def run_sparql_query(q): - results = requests.get("https://query.wikidata.org/bigdata/namespace/wdq/sparql",params={"query":q,"format":"json"}) + results = requests.get("https://query.wikidata.org/bigdata/namespace/wdq/sparql",params={"format":"json","query":q}) time.sleep(2) return results diff --git a/transliterations/src/wikidata_search.py b/transliterations/src/wikidata_search.py index 792d61f..e774f68 100644 --- a/transliterations/src/wikidata_search.py +++ b/transliterations/src/wikidata_search.py @@ -3,14 +3,17 @@ from os import path from sys import stdout from wikidata_api_calls import search_wikidata, get_wikidata_api import csv +from itertools import chain class Wikidata_ResultSet: def __init__(self): self.results = [] def extend(self, term, results): - self.results.extend([Wikidata_Result(term, result, i) - for i, result in enumerate(results)]) + self.results.append( + (Wikidata_Result(term, result, i) + for i, result in enumerate(results)) + ) def to_csv(self, outfile=None): if outfile is None: @@ -18,10 +21,9 @@ class Wikidata_ResultSet: else: of = open(outfile,'w',newline='') - writer = csv.writer(of) writer.writerow(Wikidata_Result.__slots__) - writer.writerows(map(Wikidata_Result.to_list, self.results)) + writer.writerows(map(Wikidata_Result.to_list, chain(* self.results))) class Wikidata_Result: @@ -46,20 +48,44 @@ class Wikidata_Result: self.search_position, self.timestamp] -def run_wikidata_searches(terms_file = '../data/input/base_terms.txt', outfile="../data/output/wikidata_search_results.csv"): - +def run_wikidata_searches(terms): + api = get_wikidata_api() resultset = Wikidata_ResultSet() - for term in open(terms_file,'r'): - api = get_wikidata_api() + for term in terms: search_results = search_wikidata(api, term) resultset.extend(term, search_results) + return resultset + +def read_google_trends_files(terms_files): + def _read_file(infile): + return csv.DictReader(open(infile,'r',newline='')) + for row in chain(* [_read_file(terms_file) for terms_file in terms_files]): + yield row['query'] + + +def trawl_google_trends(terms_files, outfile = None): + terms = read_google_trends_files(terms_files) + resultset = run_wikidata_searches(terms) resultset.to_csv(outfile) +def trawl_base_terms(infiles, outfile = None): + terms = chain(* (open(infile,'r') for infile in infiles)) + resultset = run_wikidata_searches(terms) + resultset.to_csv(outfile) ## search each of the base terms in wikidata # store unique entities found in the search results, the position in the search result, and the date if __name__ == "__main__": - run_wikidata_searches() + import argparse + parser = argparse.ArgumentParser("Search wikidata for entities related to a set of terms.") + parser.add_argument('inputs', type=str, nargs='+', help='one or more files to read') + parser.add_argument('--use-gtrends', action='store_true', help = 'toggle whether the input is the output from google trends') + parser.add_argument('--output', type=str, help='an output file. defaults to stdout') + args = parser.parse_args() + if args.use_gtrends: + trawl_google_trends(args.inputs, args.output) + else: + trawl_base_terms(args.inputs, args.output) diff --git a/transliterations/src/wikidata_transliterations.py b/transliterations/src/wikidata_transliterations.py index e947422..d878354 100644 --- a/transliterations/src/wikidata_transliterations.py +++ b/transliterations/src/wikidata_transliterations.py @@ -6,11 +6,11 @@ from json import JSONDecodeError class LabelData: __slots__ = ['entityid','label','langcode','is_alt'] - def __init__(self, wd_res, entityid, is_alt): + def __init__(self, wd_res, is_alt): obj = wd_res.get('label',None) self.label = obj.get('value',None) self.langcode = obj.get('xml:lang',None) - self.entityid = entityid + self.entityid = wd_res.get('entity',None).get('value',None) self.is_alt = is_alt def to_list(self): @@ -19,8 +19,7 @@ class LabelData: self.langcode, self.is_alt] - -def GetAllLabels(in_csv, outfile, topN): +def GetAllLabels(in_csvs, outfile, topNs): def load_entity_ids(in_csv, topN=5): with open(in_csv,'r',newline='') as infile: @@ -29,9 +28,9 @@ def GetAllLabels(in_csv, outfile, topN): if int(row['search_position']) < topN: yield row["entityid"] - ids = set(load_entity_ids(in_csv, topN)) + ids = set(chain(* map(lambda in_csv, topN: load_entity_ids(in_csv, topN), in_csvs, topNs))) - labeldata = chain(* map(GetEntityLabels, ids)) + labeldata = GetEntityLabels(ids) with open(outfile, 'w', newline='') as of: writer = csv.writer(of) @@ -39,41 +38,59 @@ def GetAllLabels(in_csv, outfile, topN): writer.writerows(map(LabelData.to_list,labeldata)) -def GetEntityLabels(entityid): +def GetEntityLabels(entityids): - def run_query_and_parse(query, entityid, is_alt): - results = run_sparql_query(query % entityid) + def run_query_and_parse(query, is_alt): + results = run_sparql_query(query) try: jobj = results.json() + res = jobj.get('results',None) if res is not None: res = res.get('bindings',None) if res is None: raise requests.APIError(f"got invalid response from wikidata for {query % entityid}") + for info in res: - yield LabelData(info, entityid, is_alt) + yield LabelData(info, is_alt) except JSONDecodeError as e: print(e) - print(query % entityid) + print(query) - - label_base_query = """ - SELECT DISTINCT ?label WHERE { - wd:%s rdfs:label ?label; - }""" - - altLabel_base_query = """ - SELECT DISTINCT ?label WHERE { - wd:%s skos:altLabel ?label; - }""" - - label_results = run_query_and_parse(label_base_query, entityid, is_alt=False) - - altLabel_results = run_query_and_parse(altLabel_base_query, entityid, is_alt=True) - - return chain(label_results, altLabel_results) + def prep_query(query, prop, entityids): + values = ' '.join(('wd:{0}'.format(id) for id in entityids)) + return query.format(prop, values) + + base_query = """ + SELECT DISTINCT ?entity ?label WHERE {{ + ?entity {0} ?label; + VALUES ?entity {{ {1} }} + }}""" + + # we can't get all the entities at once. how about 100 at a time? + chunksize = 100 + entityids = (id for id in entityids) + chunk = list(islice(entityids, chunksize)) + calls = [] + while len(chunk) > 0: + label_query = prep_query(base_query, "rdfs:label", chunk) + altLabel_query = prep_query(base_query, "skos:altLabel", chunk) + label_results = run_query_and_parse(label_query, is_alt=False) + altLabel_results = run_query_and_parse(altLabel_query, is_alt=True) + calls.extend([label_results, altLabel_results]) + chunk = list(islice(entityids, chunksize)) + + return chain(*calls) if __name__ == "__main__": - GetAllLabels("../data/output/wikidata_search_results.csv","../data/output/wikidata_entity_labels.csv", topN=20) + import argparse + parser = argparse.ArgumentParser("Use wikidata to find transliterations of terms") + parser.add_argument('inputs', type=str, nargs='+', help='one or more files to read. the inputs are generated by wikidata_search.py') + parser.add_argument('--topN', type=int, nargs='+', help='limit number of wikidata search results to use, can pass one arg for each source.') + parser.add_argument('--output', type=str, help='an output file. defaults to stdout',default=20) + + args = parser.parse_args() + + GetAllLabels(args.inputs, args.output, topNs=args.topN) -- 2.39.5