def stacked_query(group, template, joiner, q_type, refresh, stacked, verbose):
"""Auxiliary function to query list of items.
Parameters
----------
group : list of str
Scopus IDs (of authors or sources) for which the stacked query should
be conducted.
template : string.Template()
A string template with one parameter named `fill` which will be used
as search query.
joiner : str
On which the group elements should be joined to fill the query.
q_type : str
Determines the query search that will be used. Allowed values:
"author", "docs".
refresh : bool
Whether the cached files should be refreshed or not.
stacked: bool
If True search for queries as close as possible to the maximum length
QUERY_MAX_LEN. If False, search elements in group one by one.
verbose : bool (optional, default=False)
Whether to print information on the search progress.
Returns
-------
res : list
A list of namedtuples representing publications.
"""
maxlen = 1
if stacked:
maxlen = QUERY_MAX_LEN
queries = create_queries(group, joiner, template, maxlen)
total = len(queries)
print_progress(0, total, verbose)
res = []
for i, q in enumerate(queries):
print_progress(i + 1, total, verbose)
res.extend(long_query(q, q_type, template, refresh))
return res
def search_group_from_sources(self, stacked, verbose, refresh=False):
"""Define groups of authors based on publications from a set of sources.
Parameters
----------
self : sosia.Original
The object of the Scientist to search information for.
verbose : bool (optional, default=False)
Whether to report on the progress of the process.
refresh : bool (optional, default=False)
Whether to refresh cached search files.
Returns
-------
today, then, negative : set
Set of authors publishing in three periods: During the year of
treatment, during years to match on, and during years before the
first publication.
"""
# Filtering variables
min_year = self.first_year - self.year_margin
max_year = self.first_year + self.year_margin
if self.period:
_margin_setter = self.publications_period
else:
_margin_setter = self.publications
max_pubs = max(margin_range(len(_margin_setter), self.pub_margin))
years = list(range(min_year, max_year + 1))
search_years = [min_year - 1]
if not self._ignore_first_id:
search_years.extend(range(min_year, max_year + 1))
search_sources, _ = zip(*self.search_sources)
# Verbose variables
n = len(search_sources)
text = "Searching authors for search_group in {} sources...".format(n)
custom_print(text, verbose)
today = set()
then = set()
negative = set()
if stacked: # Make use of SQL cache
# Year provided (select also based on location)
# Get already cached sources from cache
sources_ay = DataFrame(list(product(search_sources,
[self.active_year])),
columns=["source_id", "year"])
_, _search = sources_in_cache(sources_ay, refresh=refresh, afid=True)
res = query_year(self.active_year,
_search.source_id.tolist(),
refresh,
verbose,
afid=True)
cache_insert(res, table="sources_afids")
sources_ay, _ = sources_in_cache(sources_ay,
refresh=refresh,
afid=True)
# Authors publishing in provided year and locations
mask = None
if self.search_affiliations:
mask = sources_ay.afid.isin(self.search_affiliations)
today = flat_set_from_df(sources_ay, "auids", mask)
# Years before active year
# Get already cached sources from cache
sources_ys = DataFrame(list(product(search_sources, search_years)),
columns=["source_id", "year"])
_, sources_ys_search = sources_in_cache(sources_ys, refresh=refresh)
missing_years = set(sources_ys_search.year.tolist())
# Eventually add information for missing years to cache
for y in missing_years:
mask = sources_ys_search.year == y
_sources_search = sources_ys_search[mask].source_id.tolist()
res = query_year(y, _sources_search, refresh, verbose)
cache_insert(res, table="sources")
# Get full cache
sources_ys, _ = sources_in_cache(sources_ys, refresh=False)
# Authors publishing in year(s) of first publication
if not self._ignore_first_id:
mask = sources_ys.year.between(min_year, max_year, inclusive=True)
then = flat_set_from_df(sources_ys, "auids", mask)
# Authors with publications before
mask = sources_ys.year < min_year
negative = flat_set_from_df(sources_ys, "auids", mask)
else:
auth_count = []
print_progress(0, n, verbose)
for i, source_id in enumerate(search_sources):
info = query_journal(source_id, [self.active_year] + years,
refresh)
today.update(info[str(self.active_year)])
if not self._ignore_first_id:
for y in years:
then.update(info[str(y)])
for y in range(int(min(info.keys())), min_year):
negative.update(info[str(y)])
for y in info:
if int(y) <= self.active_year:
auth_count.extend(info[str(y)])
print_progress(i + 1, n, verbose)
c = Counter(auth_count)
negative.update({a for a, npub in c.items() if npub > max_pubs})
return today, then, negative
def filter_pub_counts(group,
ybefore,
yupto,
npapers,
yfrom=None,
verbose=False):
"""Filter authors based on restrictions in the number of
publications in different periods, searched by query_size.
Parameters
----------
group : list of str
Scopus IDs of authors to be filtered.
ybefore : int
Year to be used as first year. Publications on this year and before
need to be 0.
yupto : int
Year up to which to count publications.
npapers : list
List of count of publications, minimum and maximum.
yfrom : int (optional, default=None)
If provided, publications are counted only after this year.
Publications are still set to 0 before ybefore.
Returns
-------
group : list of str
Scopus IDs filtered.
pubs_counts : list of int
List of count of publications within the period provided for authors
in group.
older_authors : list of str
Scopus IDs filtered out because have publications before ybefore.
Notes
-----
It uses cached values first, and searches for more data if needed.
"""
group = [int(x) for x in group]
years_check = [ybefore, yupto]
if yfrom:
years_check.extend([yfrom - 1])
authors = DataFrame(list(product(group, years_check)),
columns=["auth_id", "year"],
dtype="int64")
authors_size = author_size_in_cache(authors)
au_skip = []
group_tocheck = [x for x in group]
older_authors = []
pubs_counts = []
# use information in cache
if not authors_size.empty:
# authors that can be already removed because older
mask = ((authors_size.year <= ybefore) & (authors_size.n_pubs > 0))
remove = (authors_size[mask]["auth_id"].drop_duplicates().tolist())
older_authors.extend(remove)
au_remove = [x for x in remove]
# remove if number of pubs in year is in any case too small
mask = ((authors_size.year >= yupto) &
(authors_size.n_pubs < min(npapers)))
remove = (authors_size[mask]["auth_id"].drop_duplicates().tolist())
au_remove.extend(remove)
# authors with no pubs before min year
mask = (((authors_size.year == ybefore) & (authors_size.n_pubs == 0)))
au_ok_miny = (authors_size[mask]["auth_id"].drop_duplicates().tolist())
# check publications in range
if yfrom:
# adjust count by substracting the count before period; keep
# only authors for which it is possible
mask = (authors_size.year == yfrom - 1)
authors_size_bef = authors_size[mask]
authors_size_bef["year"] = yupto
authors_size_bef.columns = ["auth_id", "year", "n_pubs_bef"]
bef_auth = set(authors_size_bef["auth_id"])
mask = ((authors_size["auth_id"].isin(bef_auth)) &
(authors_size["year"] == yupto))
authors_size = authors_size[mask]
authors_size = authors_size.merge(authors_size_bef,
"left",
on=["auth_id", "year"])
authors_size = authors_size.fillna(0)
authors_size["n_pubs"] -= authors_size["n_pubs_bef"]
# authors that can be already removed because of pubs count
mask = (((authors_size.year >= yupto) &
(authors_size.n_pubs < min(npapers))) |
((authors_size.year <= yupto) &
(authors_size.n_pubs > max(npapers))))
remove = (authors_size[mask]["auth_id"].drop_duplicates().tolist())
au_remove.extend(remove)
# authors with pubs count within the range before the given year
mask = (((authors_size.year == yupto) &
(authors_size.n_pubs >= min(npapers))) &
(authors_size.n_pubs <= max(npapers)))
au_ok_year = authors_size[mask][["auth_id",
"n_pubs"]].drop_duplicates()
# authors ok (match both conditions)
au_ok = list(set(au_ok_miny).intersection(set(au_ok_year["auth_id"])))
mask = au_ok_year["auth_id"].isin(au_ok)
pubs_counts = au_ok_year[mask]["n_pubs"].tolist()
# authors that match only the first condition, but the second is
# not known, can skip the first cindition check.
au_skip = [x for x in au_ok_miny if x not in au_remove + au_ok]
group = [x for x in group if x not in au_remove]
group_tocheck = [x for x in group if x not in au_skip + au_ok]
text = "Left with {} authors based on size information already in "\
"cache.\n{} to check\n".format(len(group), len(group_tocheck))
custom_print(text, verbose)
# Verify that publications before minimum year are 0
if group_tocheck:
text = "Searching through characteristics of {:,} authors...".format(
len(group_tocheck))
custom_print(text, verbose)
print_progress(0, len(group_tocheck), verbose)
to_loop = [x for x in group_tocheck] # Temporary copy
for i, au in enumerate(to_loop):
q = "AU-ID({}) AND PUBYEAR BEF {}".format(au, ybefore + 1)
size = base_query("docs", q, size_only=True)
tp = (au, ybefore, size)
cache_insert(tp, table="author_size")
print_progress(i + 1, len(to_loop), verbose)
if not size == 0:
group.remove(au)
group_tocheck.remove(au)
older_authors.append(au)
text = "Left with {} authors based on size information before "\
"minium year\n Filtering based on size query before "\
"provided year\n".format(len(group))
custom_print(text, verbose)
# Verify that publications before the given year falle in range
group_tocheck.extend(au_skip)
n = len(group_tocheck)
if group_tocheck:
text = "Searching through characteristics of {:,} authors".format(n)
custom_print(text, verbose)
print_progress(0, n, verbose)
for i, au in enumerate(group_tocheck):
q = "AU-ID({}) AND PUBYEAR BEF {}".format(au, yupto + 1)
n_pubs_yupto = base_query("docs", q, size_only=True)
tp = (au, yupto, n_pubs_yupto)
cache_insert(tp, table="author_size")
# Eventually decrease publication count
if yfrom and n_pubs_yupto >= min(npapers):
q = "AU-ID({}) AND PUBYEAR BEF {}".format(au, yfrom)
n_pubs_yfrom = base_query("docs", q, size_only=True)
tp = (au, yfrom - 1, n_pubs_yfrom)
cache_insert(tp, table="author_size")
n_pubs_yupto -= n_pubs_yfrom
if n_pubs_yupto < min(npapers) or n_pubs_yupto > max(npapers):
group.remove(au)
else:
pubs_counts.append(n_pubs_yupto)
print_progress(i + 1, n, verbose)
return group, pubs_counts, older_authors
def find_matches(original, stacked, verbose, refresh):
"""Find matches within the search group.
Parameters
----------
original : sosia.Original()
The object containing information for the original scientist to
search for. Attribute search_group needs to exist.
stacked : bool (optional, default=False)
Whether to combine searches in few queries or not. Cached
files will most likely not be reusable. Set to True if you
query in distinct fields or you want to minimize API key usage.
verbose : bool (optional, default=False)
Whether to report on the progress of the process.
refresh : bool (optional, default=False)
Whether to refresh cached results (if they exist) or not. If int
is passed and stacked=False, results will be refreshed if they are
older than that value in number of days.
"""
# Variables
_years = range(original.first_year - original.first_year_margin,
original.first_year + original.first_year_margin + 1)
_npapers = margin_range(len(original.publications), original.pub_margin)
_max_papers = max(_npapers)
_ncits = margin_range(original.citations, original.cits_margin)
_max_cits = max(_ncits)
_ncoauth = margin_range(len(original.coauthors), original.coauth_margin)
_max_coauth = max(_ncoauth)
if original.period:
_npapers = margin_range(len(original.publications_period),
original.pub_margin)
_ncits = margin_range(original.citations_period, original.cits_margin)
_ncoauth = margin_range(len(original.coauthors_period),
original.coauth_margin)
text = "Searching through characteristics of "\
f"{len(original.search_group):,} authors..."
custom_print(text, verbose)
conn = original.sql_conn
# First round of filtering: minimum publications and main field
# create df of authors
authors = get_authors(original.search_group,
original.sql_conn,
verbose=verbose)
same_field = authors['areas'].str.startswith(original.main_field[1])
enough_pubs = authors['documents'].astype(int) >= int(min(_npapers))
group = sorted(authors[same_field & enough_pubs]["auth_id"].tolist())
text = f"Left with {len(group):,} authors with sufficient "\
"number of publications and same main field"
custom_print(text, verbose)
# Second round of filtering:
# Check having no publications before minimum year, and if 0, the
# number of publications in the relevant period.
params = {
"group": group,
"ybefore": min(_years) - 1,
"yupto": original.year,
"npapers": _npapers,
"yfrom": original._period_year,
"verbose": verbose,
"conn": conn
}
group, _, _ = filter_pub_counts(**params)
# Screen out profiles with too many publications over the full period
if original.period:
params.update({
"npapers": [1, _max_papers],
"yfrom": None,
"group": group
})
group, _, _ = filter_pub_counts(**params)
text = f"Left with {len(group):,} researchers"
custom_print(text, verbose)
# Third round of filtering: citations (in the FULL period)
authors = pd.DataFrame({"auth_id": group, "year": original.year})
auth_cits, missing = retrieve_author_info(authors, conn, "author_ncits")
if not missing.empty:
total = missing.shape[0]
text = f"Counting citations of {total:,} authors..."
custom_print(text, verbose)
missing['n_cits'] = 0
print_progress(0, total, verbose)
start = 0
for i, au in missing.iterrows():
n_cits = count_citations([str(au['auth_id'])], original.year + 1)
missing.at[i, 'n_cits'] = n_cits
print_progress(i + 1, total, verbose)
if i % 100 == 0 or i == len(missing) - 1:
insert_data(missing.iloc[start:i + 1],
conn,
table="author_ncits")
start = i
auth_cits = pd.concat([auth_cits, missing])
auth_cits['auth_id'] = auth_cits['auth_id'].astype("uint64")
# Keep if citations are in range
custom_print("Filtering based on count of citations...", verbose)
mask = auth_cits["n_cits"].between(min(_ncits), _max_cits)
group = auth_cits[mask]['auth_id'].tolist()
# Fourth round of filtering: Download publications, verify coauthors
# (in the FULL period) and first year
text = f"Left with {len(group):,} authors\nFiltering based on "\
"coauthor count..."
custom_print(text, verbose)
authors = pd.DataFrame({
"auth_id": group,
"year": original.year
},
dtype="uint64")
_, author_year_search = retrieve_author_info(authors, conn, "author_year")
matches = []
if not author_year_search.empty:
q = Template(f"AU-ID($fill) AND PUBYEAR BEF {original.year + 1}")
auth_year_group = author_year_search["auth_id"].tolist()
params = {
"group": auth_year_group,
"template": q,
"refresh": refresh,
"joiner": ") OR AU-ID(",
"q_type": "docs",
"verbose": verbose,
"stacked": stacked
}
res = stacked_query(**params)
res = build_dict(res, auth_year_group)
if res:
# res can become empty after build_dict if a au_id is old
res = pd.DataFrame.from_dict(res, orient="index")
res["year"] = original.year
res = res[["year", "first_year", "n_pubs", "n_coauth"]]
res.index.name = "auth_id"
res = res.reset_index()
insert_data(res, original.sql_conn, table="author_year")
authors_year, _ = retrieve_author_info(authors, conn, "author_year")
# Check for number of coauthors within margin
mask = authors_year["n_coauth"].between(min(_ncoauth), _max_coauth)
# Check for year of first publication within range
if not original.first_year_name_search:
same_start = authors_year["first_year"].between(
min(_years), max(_years))
mask = mask & same_start
# Filter
matches = sorted(authors_year[mask]["auth_id"].tolist())
text = f"Left with {len(matches)} authors"
custom_print(text, verbose)
if original.period:
text = "Filtering based on citations and coauthor count during period..."
custom_print(text, verbose)
# Further screen matches based on period cits and coauths
to_loop = [m for m in matches] # temporary copy
for m in to_loop:
res = base_query("docs",
f"AU-ID({m})",
refresh=refresh,
fields=["eid", "author_ids", "coverDate"])
pubs = [
p for p in res if
original._period_year <= int(p.coverDate[:4]) <= original.year
]
coauths = set(extract_authors(pubs)) - {str(m)}
if not (min(_ncoauth) <= len(coauths) <= max(_ncoauth)):
matches.remove(m)
continue
eids_period = [p.eid for p in pubs]
n_cits = count_citations(eids_period, original.year + 1, [str(m)])
if not (min(_ncits) <= n_cits <= max(_ncits)):
matches.remove(m)
# Eventually filter on affiliations
if original.search_affiliations:
text = "Filtering based on affiliations..."
custom_print(text, verbose)
matches[:] = [
m for m in matches if same_affiliation(original, m, refresh)
]
return matches
def inform_matches(self, keywords, verbose, refresh):
"""Add match-specific information to all matches.
Parameters
----------
self : sosia.Original()
Object whose matches should receive additional information.
keywords : iterable of strings
Which information to add to matches.
verbose : bool
Whether to report on the progress of the process and the completeness
of document information.
refresh : bool
Whether to refresh all cached files or not.
Returns
-------
out : list of namedtuples
A list of namedtuples representing matches. Provided information
depend on provided keywords.
"""
from sosia.classes import Scientist
# Create Match object
fields = "ID name " + " ".join(keywords)
m = namedtuple("Match", fields)
# Preparation
doc_parse = "num_cited_refs" in keywords
if doc_parse:
focal_docs = parse_docs([d.eid for d in self.publications], refresh)
focal_refs, focal_refs_n = focal_docs
# Add selected information match-by-match
out = []
completeness = {}
total = len(self.matches)
print_progress(0, total, verbose)
for idx, auth_id in enumerate(self.matches):
period = self.year + 1 - self._period_year
p = Scientist([auth_id],
self.year,
period=period,
refresh=refresh,
sql_fname=self.sql_fname)
match_info = inform_match(p, keywords, refresh=refresh)
# Abstract and reference similarity is performed jointly
if doc_parse:
eids = [d.eid for d in p.publications]
refs, refs_n = parse_docs(eids, refresh)
completeness[auth_id] = (refs_n, len(eids))
if "num_cited_refs" in keywords:
ref_cos = compute_overlap(refs, focal_refs)
match_info["num_cited_refs"] = ref_cos
out.append(m(**match_info))
print_progress(idx + 1, total, verbose)
# Eventually print information on missing information
if verbose and doc_parse:
for auth_id, completeness in completeness.items():
_print_missing_docs([auth_id], completeness[0], completeness[1])
focal_pubs_n = len(self.publications)
_print_missing_docs(self.identifier,
focal_refs_n,
focal_pubs_n,
res_type="Original")
return out
def stacked_query(group,
res,
template,
joiner,
q_type,
refresh,
i=0,
total=None):
"""Auxiliary function to recursively perform queries until they work.
Parameters
----------
group : list of str
Scopus IDs (of authors or sources) for which the stacked query should
be conducted.
res : list
(Initially empty )Container to which the query results will be
appended.
template : Template()
A string template with one paramter named `fill` which will be used
as search query.
joiner : str
On wich the group elements should be joined to fill the query.
q_type : str
Determines the query search that will be used. Allowed values:
"author", "docs".
refresh : bool
Whether the cached files should be refreshed or not.
i : int (optional, default=0)
A count variable to be used for printing the progress bar.
total : int (optional, default=None)
The total number of elements in the group. If provided, a progress
bar will be printed.
Returns
-------
res : list
A list of namedtuples representing publications.
i : int
A running variable to indicate the progress.
Notes
-----
Results of each successful query are appended to ´res´.
"""
group = [str(g) for g in group] # make robust to passing int
q = template.substitute(fill=joiner.join(group))
try:
n = base_query(q_type, q, size_only=True)
if n > 5000 and len(group) > 1:
raise ScopusQueryError()
res.extend(base_query(q_type, q, refresh=refresh))
verbose = total is not None
i += len(group)
print_progress(i, total, verbose)
except (Scopus400Error, Scopus500Error, ScopusQueryError) as e:
# Split query group into two equally sized groups
mid = len(group) // 2
params = {
"group": group[:mid],
"res": res,
"template": template,
"i": i,
"joiner": joiner,
"q_type": q_type,
"total": total,
"refresh": refresh
}
res, i = stacked_query(**params)
params.update({"group": group[mid:], "i": i})
res, i = stacked_query(**params)
return res, i
def find_matches(self,
stacked=False,
verbose=False,
stop_words=STOPWORDS,
information=True,
refresh=False,
**tfidf_kwds):
"""Find matches within search_group based on four criteria:
1. Started publishing in about the same year
2. Has about the same number of publications in the year of treatment
3. Has about the same number of coauthors in the year of treatment
4. Has about the same number of citations in the year of treatment
5. Works in the same field as the scientist's main field
Parameters
----------
stacked : bool (optional, default=False)
Whether to combine searches in few queries or not. Cached
files will most likely not be resuable. Set to True if you
query in distinct fields or you want to minimize API key usage.
verbose : bool (optional, default=False)
Whether to report on the progress of the process.
stop_words : list (optional, default=STOPWORDS)
A list of words that should be filtered in the analysis of
abstracts. Default list is the list of english stopwords
by nltk, augmented with numbers and interpunctuation.
information : bool or iterable (optional, default=True)
Whether to return additional information on the matches that may
help in the selection process. If an iterable of keywords is
provied, only return information for these keywords. Allowed
values are "first_year", "num_coauthors", "num_publications",
"num_citations", "country", "language",
"reference_sim", "abstract_sim".
refresh : bool (optional, default=False)
Whether to refresh cached search files.
tfidf_kwds : keywords
Parameters to pass to TfidfVectorizer from the sklearn package
for abstract vectorization. Not used when `information=False` or
or when "abstract_sim" is not in `information`. See
https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html
for possible values.
Returns
-------
matches : list
A list of Scopus IDs of scientists matching all the criteria (if
information is False) or a list of namedtuples with the Scopus ID
and additional information (if information is True).
Raises
------
ValueError
If information is not bool and contains invalid keywords.
"""
# Checks
info_keys = [
"first_name", "surname", "first_year", "num_coauthors",
"num_publications", "num_citations", "num_coauthors_period",
"num_publications_period", "num_citations_period", "subjects",
"country", "affiliation_id", "affiliation", "language",
"reference_sim", "abstract_sim"
]
if isinstance(information, bool):
if information:
keywords = info_keys
elif self.search_affiliations:
information = True
keywords = ["affiliation_id"]
else:
keywords = None
else:
keywords = information
invalid = [x for x in keywords if x not in info_keys]
if invalid:
text = ("Parameter information contains invalid keywords: ",
", ".join(invalid))
raise ValueError(text)
if self.search_affiliations and "affiliation_id" not in keywords:
keywords.append("affiliation_id")
# Variables
_years = range(self.first_year - self.year_margin,
self.first_year + self.year_margin + 1)
if self.period:
_npapers = margin_range(len(self.publications_period),
self.pub_margin)
_ncits = margin_range(self.citations_period, self.cits_margin)
_ncoauth = margin_range(len(self.coauthors_period),
self.coauth_margin)
_npapers_full = margin_range(len(self.publications),
self.pub_margin)
_ncits_full = margin_range(self.citations, self.cits_margin)
_ncoauth_full = margin_range(len(self.coauthors),
self.coauth_margin)
else:
_npapers = margin_range(len(self.publications), self.pub_margin)
_ncits = margin_range(self.citations, self.cits_margin)
_ncoauth = margin_range(len(self.coauthors), self.coauth_margin)
n = len(self.search_group)
text = "Searching through characteristics of {:,} authors".format(n)
custom_print(text, verbose)
# First round of filtering: minimum publications and main field
# create df of authors
authors = query_author_data(self.search_group, verbose=verbose)
same_field = (authors.areas.str.startswith(self.main_field[1]))
enough_pubs = (authors.documents.astype(int) >= int(min(_npapers)))
group = authors[same_field & enough_pubs]["auth_id"].tolist()
group.sort()
n = len(group)
text = "Left with {} authors\nFiltering based on provided "\
"conditions...".format(n)
custom_print(text, verbose)
# Second round of filtering:
# Check having no publications before minimum year, and if 0, the
# number of publications in the relevant period.
params = {
"group": group,
"ybefore": min(_years) - 1,
"yupto": self.year,
"npapers": _npapers,
"yfrom": self.year_period,
"verbose": verbose
}
group, _, _ = filter_pub_counts(**params)
# Also screen out ids with too many publications over the full period
if self.period:
params.update({
"npapers": [1, max(_npapers_full)],
"yfrom": None,
"group": group
})
group, _, _ = filter_pub_counts(**params)
# Third round of filtering: citations (in the FULL period).
authors = pd.DataFrame({"auth_id": group, "year": self.year})
_, authors_cits_search = author_cits_in_cache(authors)
text = "Search and filter based on count of citations\n{} to search "\
"out of {}\n".format(len(authors_cits_search), len(group))
custom_print(text, verbose)
if not authors_cits_search.empty:
authors_cits_search['n_cits'] = 0
print_progress(0, len(authors_cits_search), verbose)
for i, au in authors_cits_search.iterrows():
q = "REF({}) AND PUBYEAR BEF {} AND NOT AU-ID({})".format(
au['auth_id'], self.year + 1, au['auth_id'])
n = base_query("docs", q, size_only=True)
authors_cits_search.at[i, 'n_cits'] = n
print_progress(i + 1, len(authors_cits_search), verbose)
cache_insert(authors_cits_search, table="author_cits_size")
auth_cits_incache, _ = author_cits_in_cache(
authors[["auth_id", "year"]])
# keep if citations are in range
mask = ((auth_cits_incache.n_cits <= max(_ncits)) &
(auth_cits_incache.n_cits >= min(_ncits)))
if self.period:
mask = ((auth_cits_incache.n_cits >= min(_ncits)) &
(auth_cits_incache.n_cits <= max(_ncits_full)))
group = (auth_cits_incache[mask]['auth_id'].tolist())
# Fourth round of filtering: Download publications, verify coauthors
# (in the FULL period) and first year.
n = len(group)
text = "Left with {} authors\nFiltering based on coauthors "\
"number...".format(n)
custom_print(text, verbose)
authors = pd.DataFrame({
"auth_id": group,
"year": self.year
},
dtype="uint64")
_, author_year_search = author_year_in_cache(authors)
matches = []
if stacked: # Combine searches
if not author_year_search.empty:
q = Template(
"AU-ID($fill) AND PUBYEAR BEF {}".format(self.year + 1))
auth_year_group = author_year_search.auth_id.tolist()
params = {
"group": auth_year_group,
"res": [],
"template": q,
"refresh": refresh,
"joiner": ") OR AU-ID(",
"q_type": "docs"
}
if verbose:
params.update({"total": len(auth_year_group)})
res, _ = stacked_query(**params)
res = build_dict(res, auth_year_group)
if res:
# res can become empty after build_dict if a au_id is old
res = pd.DataFrame.from_dict(res, orient="index")
res["year"] = self.year
res = res[["year", "first_year", "n_pubs", "n_coauth"]]
res.index.name = "auth_id"
res = res.reset_index()
cache_insert(res, table="author_year")
author_year_cache, _ = author_year_in_cache(authors)
if self._ignore_first_id:
# only number of coauthors should be big enough
enough = (author_year_cache.n_coauth >= min(_ncoauth))
notoomany = (author_year_cache.n_coauth <= max(_ncoauth_full))
mask = enough & notoomany
elif self.period:
# number of coauthors should be "big enough" and first year in
# window
same_start = (author_year_cache.first_year.between(
min(_years), max(_years)))
enough = (author_year_cache.n_coauth >= min(_ncoauth))
notoomany = (author_year_cache.n_coauth <= max(_ncoauth_full))
mask = same_start & enough & notoomany
else:
# all restrictions apply
same_start = (author_year_cache.first_year.between(
min(_years), max(_years)))
same_coauths = (author_year_cache.n_coauth.between(
min(_ncoauth), max(_ncoauth)))
mask = same_start & same_coauths
matches = author_year_cache[mask]["auth_id"].tolist()
else: # Query each author individually
for i, au in enumerate(group):
print_progress(i + 1, len(group), verbose)
res = base_query("docs",
"AU-ID({})".format(au),
refresh=refresh)
res = [
p for p in res
if p.coverDate and int(p.coverDate[:4]) <= self.year
]
# Filter
min_year = int(min([p.coverDate[:4] for p in res]))
authids = [p.author_ids for p in res if p.author_ids]
authors = set([a for p in authids for a in p.split(";")])
n_coauth = len(authors) - 1 # Subtract 1 for focal author
if self._ignore_first_id and (n_coauth < max(_ncoauth)):
# only number of coauthors should be big enough
continue
elif (self.period and ((n_coauth < max(_ncoauth)) or
(min_year not in _years))):
# number of coauthors should be "big enough" and first year
# in window
continue
elif ((len(res) not in _npapers) or (min_year not in _years)
or (n_coauth not in _ncoauth)):
continue
matches.append(au)
if self.period:
text = "Left with {} authors\nFiltering based on exact period "\
"citations and coauthors...".format(len(matches))
custom_print(text, verbose)
# Further screen matches based on period cits and coauths
to_loop = [m for m in matches] # temporary copy
for m in to_loop:
q = "AU-ID({})".format(m)
res = base_query("docs",
"AU-ID({})".format(m),
refresh=refresh,
fields=["eid", "author_ids", "coverDate"])
pubs = [
p for p in res if int(p.coverDate[:4]) <= self.year
and int(p.coverDate[:4]) >= self.year_period
]
coauths = set(get_authors(pubs)) - {str(m)}
if not (min(_ncoauth) <= len(coauths) <= max(_ncoauth)):
matches.remove(m)
continue
eids_period = [p.eid for p in pubs]
cits = count_citations(search_ids=eids_period,
pubyear=self.year + 1,
exclusion_key="AU-ID",
exclusion_ids=[str(m)])
if not (min(_ncits) <= cits <= max(_ncits)):
matches.remove(m)
text = "Found {:,} author(s) matching all criteria".format(
len(matches))
custom_print(text, verbose)
# Possibly add information to matches
if keywords and len(matches) > 0:
custom_print("Providing additional information...", verbose)
profiles = [
Scientist([str(a)],
self.year,
period=self.period,
refresh=refresh) for a in matches
]
matches = inform_matches(profiles, self, keywords, stop_words,
verbose, refresh, **tfidf_kwds)
if self.search_affiliations:
matches = [
m for m in matches if len(
set(m.affiliation_id.replace(" ", "").split(";")).
intersection([str(a) for a in self.search_affiliations]))
]
return matches
def inform_matches(self, keywords, stop_words, verbose, refresh, **kwds):
"""Add match-specific information to all matches.
Parameters
----------
self : sosia.Original()
Object whose matches should received additional information
keywords : iterable of strings
Which information to add to matches.
stop_words : list
A list of words that should be filtered in the analysis of abstracts.
verbose : bool
Whether to report on the progress of the process and the completeness
of document information.
refresh : bool
Whether to refresh all cached files or not.
kwds : keywords
Parameters to pass to sklearn.feature_extraction.text.TfidfVectorizer
for abstract and reference vectorization.
Returns
-------
out : list of namedtuples
A list of namedtuples representing matches. Provided information
depend on provided keywords.
"""
from sklearn.feature_extraction.text import ENGLISH_STOP_WORDS
from string import digits, punctuation
from sosia.classes import Scientist
# Create Match object
fields = "ID name " + " ".join(keywords)
m = namedtuple("Match", fields)
# Preparation
doc_parse = "reference_sim" in keywords or "abstract_sim" in keywords
if doc_parse:
focal_docs = parse_docs([d.eid for d in self.publications], refresh)
focal_refs, focal_refs_n, focal_abs, focal_abs_n = focal_docs
if not stop_words:
stop_words = list(ENGLISH_STOP_WORDS) + list(punctuation + digits)
# Add selected information match-by-match
out = []
completeness = {}
total = len(self.matches)
print_progress(0, total, verbose)
meta = namedtuple("Meta", "refs absts total")
for idx, auth_id in enumerate(self.matches):
period = self.year + 1 - self._period_year
p = Scientist([auth_id],
self.year,
period=period,
refresh=refresh,
sql_fname=self.sql_fname)
match_info = inform_match(p, keywords)
# Abstract and reference similarity is performed jointly
if doc_parse:
eids = [d.eid for d in p.publications]
refs, refs_n, absts, absts_n = parse_docs(eids, refresh)
completeness[auth_id] = meta(refs=refs_n,
absts=absts_n,
total=len(eids))
if "reference_sim" in keywords:
ref_cos = compute_similarity(refs, focal_refs, **kwds)
match_info["reference_sim"] = ref_cos
if "abstract_sim" in keywords:
kwds.update({"stop_words": stop_words})
abs_cos = compute_similarity(absts,
focal_abs,
tokenize=True,
**kwds)
match_info["abstract_sim"] = abs_cos
out.append(m(**match_info))
print_progress(idx + 1, total, verbose)
# Eventually print information on missing information
if verbose and doc_parse:
for auth_id, completeness in completeness.items():
_print_missing_docs([auth_id], completeness.absts,
completeness.refs, completeness.total)
focal_pubs_n = len(self.publications)
_print_missing_docs(self.identifier,
focal_abs_n,
focal_refs_n,
focal_pubs_n,
res_type="Original")
return out
def inform_matches(profiles, focal, keywords, stop_words, verbose, refresh,
**kwds):
"""Create namedtuple adding information to matches.
Parameters
----------
profiles : list of Scientist()
A list of Scientist objects representing matches.
focal : Scientist
Object of class Scientist representing the focal scientist.
keywords : iterable of strings
Which information to add to matches.
stop_words : list
A list of words that should be filtered in the analysis of abstracts.
verbose : bool
Whether to report on the progress of the process and the completeness
of document information.
refresh : bool
Whether to refresh all cached files or not.
kwds : keywords
Parameters to pass to sklearn.feature_extraction.text.TfidfVectorizer
for abstract and reference vectorization.
Returns
-------
m : list of namedtuples
A list of namedtuples representing matches. Provided information
depend on provided keywords.
"""
from sosia.classes import Scientist
# Create Match object
fields = "ID name " + " ".join(keywords)
m = namedtuple("Match", fields)
# Preparation
doc_parse = "reference_sim" in keywords or "abstract_sim" in keywords
total = len(profiles)
print_progress(0, total, verbose)
if doc_parse:
focal_eids = [d.eid for d in focal.publications]
focal = parse_docs(focal_eids, refresh)
focal_refs, focal_refs_n, focal_abs, focal_abs_n = focal
# Add selective information
out = []
info = {} # to collect information on missing information
for idx, p in enumerate(profiles):
# Add characteristics
match_info = {"ID": p.identifier[0], "name": p.name}
if "language" in keywords:
try:
match_info["language"] = p.get_publication_languages().language
except Scopus404Error: # Refresh profile
p = Scientist(p.identifier, p.year, refresh=True)
match_info["language"] = p.get_publication_languages().language
if "first_name" in keywords:
match_info["first_name"] = p.first_name
if "surname" in keywords:
match_info["surname"] = p.surname
if "first_year" in keywords:
match_info["first_year"] = p.first_year
if "num_coauthors" in keywords:
match_info["num_coauthors"] = len(p.coauthors)
if "num_publications" in keywords:
match_info["num_publications"] = len(p.publications)
if "num_citations" in keywords:
match_info["num_citations"] = p.citations
if "num_coauthors_period" in keywords:
match_info["num_coauthors_period"] = len(p.coauthors_period)
if "num_publications_period" in keywords:
match_info["num_publications_period"] = len(p.publications_period)
if "num_citations_period" in keywords:
match_info["num_citations_period"] = p.citations_period
if "subjects" in keywords:
match_info["subjects"] = p.subjects
if "country" in keywords:
match_info["country"] = p.country
if "city" in keywords:
match_info["city"] = p.city
if "affiliation_id" in keywords:
match_info["affiliation_id"] = p.affiliation_id
if "affiliation" in keywords:
match_info["affiliation"] = p.organization
# Abstract and reference similiarity is performed jointly
if doc_parse:
eids = [d.eid for d in p.publications]
refs, refs_n, absts, absts_n = parse_docs(eids, refresh)
vec = TfidfVectorizer(**kwds)
ref_cos = compute_cos(vec.fit_transform([refs, focal_refs]))
vec = TfidfVectorizer(stop_words=stop_words,
tokenizer=tokenize_and_stem,
**kwds)
abs_cos = compute_cos(vec.fit_transform([absts, focal_abs]))
# Save info for below print statement
meta = namedtuple("Meta", "refs absts total")
meta(refs=refs_n, absts=absts_n, total=len(eids))
key = "; ".join(p.identifier)
info[key] = meta(refs=refs_n, absts=absts_n, total=len(eids))
if "reference_sim" in keywords:
match_info["reference_sim"] = ref_cos
if "abstract_sim" in keywords:
match_info["abstract_sim"] = abs_cos
# Finalize
out.append(m(**match_info))
print_progress(idx + 1, total, verbose)
# Print information on missing information
if verbose and doc_parse:
for auth_id, info in info.items():
_print_missing_docs(auth_id, info.refs, info.absts, info.total)
label = ";".join(focal.identifier) + " (focal)"
focal_pubs_n = len(focal.publications)
_print_missing_docs(label, focal_refs_n, focal_abs_n, focal_pubs_n)
return out
def filter_pub_counts(group,
conn,
ybefore,
yupto,
npapers,
yfrom=None,
verbose=False):
"""Filter authors based on restrictions in the number of
publications in different periods, searched by query_size.
Parameters
----------
conn : sqlite3 connection
Standing connection to a SQLite3 database.
group : list of str
Scopus IDs of authors to be filtered.
ybefore : int
Year to be used as first year. Publications on this year and before
need to be 0.
yupto : int
Year up to which to count publications.
npapers : list
List of count of publications, minimum and maximum.
yfrom : int (optional, default=None)
If provided, publications are counted only after this year.
Publications are still set to 0 before ybefore.
verbose : bool (optional, default=False)
Whether to print information on the search progress.
Returns
-------
group : list of str
Scopus IDs filtered.
pubs_counts : list of int
List of count of publications within the period provided for authors
in group.
older_authors : list of str
Scopus IDs filtered out because have publications before ybefore.
"""
from itertools import product
from pandas import DataFrame
group = [int(x) for x in group]
years_check = [ybefore, yupto]
if yfrom:
years_check.extend([yfrom - 1])
authors = DataFrame(product(group, years_check),
dtype="uint64",
columns=["auth_id", "year"])
auth_npubs, _ = retrieve_author_info(authors, conn, "author_pubs")
au_skip = []
group_tocheck = set(group)
older_authors = []
pubs_counts = []
# Use information in database
if not auth_npubs.empty:
# Remove authors based on age
mask = ((auth_npubs["year"] <= ybefore) & (auth_npubs["n_pubs"] > 0))
au_remove = set(auth_npubs[mask]["auth_id"].unique())
older_authors.extend(au_remove)
# Remove if number of pubs in year is in any case too small
mask = ((auth_npubs["year"] >= yupto) &
(auth_npubs["n_pubs"] < min(npapers)))
au_remove.update(auth_npubs[mask]["auth_id"])
# Authors with no pubs before min year
mask = ((auth_npubs["year"] == ybefore) & (auth_npubs["n_pubs"] == 0))
au_ok_miny = set(auth_npubs[mask]["auth_id"].unique())
# Check publications in range
if yfrom:
# Keep authors where subtracting publications from before period
# from publication count is possible
mask = auth_npubs["year"] == yfrom - 1
rename = {"n_pubs": "n_pubs_bef"}
auth_npubs_bef = auth_npubs[mask].copy().rename(columns=rename)
auth_npubs_bef["year"] = yupto
auth_npubs = (auth_npubs.merge(auth_npubs_bef,
"inner",
on=["auth_id", "year"]).fillna(0))
auth_npubs["n_pubs"] -= auth_npubs["n_pubs_bef"]
# Remove authors because of their publication count
mask = (((auth_npubs["year"] >= yupto) &
(auth_npubs["n_pubs"] < min(npapers))) |
((auth_npubs["year"] <= yupto) &
(auth_npubs["n_pubs"] > max(npapers))))
remove = auth_npubs[mask]["auth_id"]
au_remove.update(remove)
# Authors with pubs count within the range before the given year
mask = (((auth_npubs["year"] == yupto) &
(auth_npubs["n_pubs"] >= min(npapers))) &
(auth_npubs["n_pubs"] <= max(npapers)))
au_ok_year = auth_npubs[mask][["auth_id", "n_pubs"]].drop_duplicates()
# Keep authors that match both conditions
au_ok = au_ok_miny.intersection(au_ok_year["auth_id"].unique())
mask = au_ok_year["auth_id"].isin(au_ok)
pubs_counts = au_ok_year[mask]["n_pubs"].tolist()
# Skip citation check for authors that match only the first condition,
# with the second being unknown
au_skip = set([x for x in au_ok_miny if x not in au_remove | au_ok])
group = [x for x in group if x not in au_remove]
group_tocheck = set([x for x in group if x not in au_skip | au_ok])
# Verify that publications before minimum year are 0
if group_tocheck:
n = len(group_tocheck)
text = f"Obtaining information for {n:,} authors without sufficient "\
"information in database..."
custom_print(text, verbose)
print_progress(0, n, verbose)
to_loop = [x for x in group_tocheck] # Temporary copy
for i, auth_id in enumerate(to_loop):
npubs_ybefore = auth_npubs_retrieve_insert(auth_id, ybefore, conn)
if npubs_ybefore:
group.remove(auth_id)
group_tocheck.remove(auth_id)
older_authors.append(auth_id)
print_progress(i + 1, n, verbose)
text = f"Left with {len(group):,} authors based on publication "\
f"information before {ybefore}"
custom_print(text, verbose)
# Verify that publications before the given year fall in range
group_tocheck.update(au_skip)
if group_tocheck:
n = len(group_tocheck)
text = f"Counting publications of {n:,} authors before {yupto+1}..."
custom_print(text, verbose)
print_progress(0, n, verbose)
for i, au in enumerate(group_tocheck):
n_pubs_yupto = auth_npubs_retrieve_insert(au, yupto, conn)
# Eventually decrease publication count
if yfrom and n_pubs_yupto >= min(npapers):
n_pubs_yfrom = auth_npubs_retrieve_insert(au, yfrom - 1, conn)
n_pubs_yupto -= n_pubs_yfrom
if n_pubs_yupto < min(npapers) or n_pubs_yupto > max(npapers):
group.remove(au)
else:
pubs_counts.append(n_pubs_yupto)
print_progress(i + 1, n, verbose)
return group, pubs_counts, older_authors
