def calculate_overall_ranking(self, raw_queries, settings):
api = API()
mean_ap_whole = []
mean_ap_doc = []
queries = self.__raw_queries_to_queries(raw_queries)
settings["mode"] = Mode.without_importance_to_sections
settings_sec = copy.deepcopy(settings)
settings_sec["mode"] = Mode.importance_to_sections
for i, query in enumerate(queries):
progressBar(i, len(queries))
ranked_papers_whole = api.get_papers({"whole-document": query["search_query"]}, settings)
ranked_papers_sec = api.get_papers({query["imrad"]: query["search_query"]}, settings_sec)
relevant_paper = [api.get_paper(reference["paper_id"]) for reference in query["references"]]
ap_whole = self.average_precision(ranked_papers_whole, relevant_paper)
ap_doc = self.average_precision(ranked_papers_sec, relevant_paper)
mean_ap_whole.append(ap_whole)
mean_ap_doc.append(ap_doc)
result_whole = sum(mean_ap_whole) / len(mean_ap_whole)
result_doc = sum(mean_ap_doc) / len(mean_ap_doc)
print()
print("{} & {} & {}".format(Mode.without_importance_to_sections.name.replace("_", " "), len(mean_ap_whole),
round(result_whole, 4)))
print("{} & {} & {}".format(Mode.importance_to_sections.name.replace("_", " "), len(mean_ap_doc),
round(result_doc, 4)))
def create_graph():
print('create graph')
api = API()
papers = api.get_all_paper()
g = nx.Graph()
g.clear()
for paper in papers:
references = [
x.get_paper_id() for x in paper.references if x.get_paper_id()
]
for ref_id in references:
g.add_edge(str(paper.id), str(ref_id))
degrees = [len(g.edges(node)) for node in g.nodes]
for degree in degrees:
if degree == 0:
print("nope!")
print("# nodes: ", g.number_of_nodes())
print("# edges: ", g.number_of_edges())
print("# components: ", len(list(nx.connected_components(g))))
print("max degree: ", max(degrees))
print("mean degree: ", round(mean(degrees), 4))
print("median degree: ", statistics.median(degrees))
print("diameter: ", nx.diameter(g), " (maximum eccentricity - max path)")
print("periphery: ", len(nx.periphery(g)),
" (# nodes eccentricity equal to the diameter)")
create_degree_distribution(degrees, 'Degree Distribution', '#00365A', 13,
100, 3.5)
def admin_index():
if 'logged_in' in session.keys() and session['logged_in']:
api = API()
papers = api.get_all_paper()
return render_template('admin/papers.html', papers=papers)
else:
return render_template('admin/index.html')
def __add_user():
print("Add new admin to the database")
name = input("username: "******"Welcome on board {}".format(name))
def user_info():
if not ('logged_in' in session.keys() and session['logged_in']):
return redirect('admin/')
api = API()
users = api.get_all_user()
return render_template('admin/users.html', users=users)
def remove_duplicates_from_cited_by():
print("\nRemove Duplicates")
api = API()
papers = api.get_all_paper()
for i, paper in enumerate(papers):
progressBar(i, len(papers))
paper.cited_by = list(dict.fromkeys(paper.cited_by))
api.client.update_paper(paper)
def paper_info(paper_id):
api = API()
papers = api.get_all_paper()
id_to_filename = {paper.id: paper.filename for paper in papers}
paper = api.get_paper(paper_id)
return render_template('admin/paper_info.html',
paper=paper,
id_to_filename=id_to_filename)
def analize_chapters():
api = API()
papers = api.get_all_paper()
introduction, background, methods, result, discussion = {}, {}, {}, {}, {}
print("# papers: ", len(papers))
for paper in papers:
intro_titles = [
sec.heading_proceed for sec in paper.get_introduction()
]
back_titles = [sec.heading_proceed for sec in paper.get_background()]
methods_titles = [sec.heading_proceed for sec in paper.get_methods()]
result_titles = [sec.heading_proceed for sec in paper.get_results()]
discuss_titles = [
sec.heading_proceed for sec in paper.get_discussion()
]
intro_word = __is_word_in_titles(intro_titles, ["introduct"])
back_word = __is_word_in_titles(back_titles,
["relat work", "background"])
methods_word = __is_word_in_titles(methods_titles,
["method", "approach", "model"])
results_word = __is_word_in_titles(result_titles,
["result", "experi", "evalu"])
discuss_word = __is_word_in_titles(
discuss_titles, ["discuss", "conclus", "futur work"])
if intro_word:
introduction[intro_word] = introduction[
intro_word] + 1 if intro_word in introduction else 1
if back_word:
background[back_word] = background[
back_word] + 1 if back_word in background else 1
if methods_word:
methods[methods_word] = methods[
methods_word] + 1 if methods_word in methods else 1
if results_word:
result[results_word] = result[
results_word] + 1 if results_word in result else 1
if discuss_word:
discussion[discuss_word] = discussion[
discuss_word] + 1 if discuss_word in discussion else 1
print("introduction:")
print_imrad(introduction, len(papers))
print("related work:")
print_imrad(background, len(papers))
print("methods:")
print_imrad(methods, len(papers))
print("result:")
print_imrad(result, len(papers))
print("discussion:")
print_imrad(discussion, len(papers))
def print_circles(circles):
api = API()
tmp = []
for circle in circles:
tmp_circle_array = []
for node in circle:
tmp_circle_array.append(api.get_paper(node).filename)
tmp.append(tmp_circle_array)
print(tmp)
print(circles)
def create_directed_graph():
print('\ncreate directed graph')
api = API()
papers = api.get_all_paper()
dg = nx.DiGraph()
dg.clear()
for paper in papers:
references = [
x.get_paper_id() for x in paper.references if x.get_paper_id()
]
for ref_id in references:
dg.add_edge(str(paper.id), str(ref_id))
# Data cleaning - if not done 5 papers which cite each other in dataset
# preprints cited each other
dg.remove_edge('5c52a9b9bf51c50be97c5145', '5c529cbdbf51c5359dce35f3')
dg.remove_edge('5b0565406919df52a704f32c', '5b05673b6919df52a704f375')
dg.remove_edge('5b97b226bf51c561194d9f1f', '5b05682a6919df52a704f395')
dg.remove_edge('5c52a4f9bf51c50be97c5111', '5c533345bf51c5335baca21a')
dg.remove_edge('5b97b0aebf51c561194d9f09', '5b97b31ebf51c561194d9f2a')
print("# nodes: ", dg.number_of_nodes())
print("# edges: ", dg.number_of_edges())
print("# cycles: ", len(list(nx.simple_cycles(dg))))
print("# strongly connected components: ",
len(list(nx.strongly_connected_components(dg))))
print("Dag longest path: ", len(nx.dag_longest_path(dg)))
in_degrees = []
out_degrees = []
root_nodes = []
for node in dg.nodes:
if len(dg.in_edges(node)) > 0:
in_degrees.append(len(dg.in_edges(node)))
if len(dg.out_edges(node)) > 0:
out_degrees.append(len(dg.out_edges(node)))
if len(dg.out_edges(node)) == 0:
root_nodes.append(node)
print("# root nodes: ", len(root_nodes))
print("In Degree:")
print(" max degree: ", max(in_degrees))
print(" mean degree: ", round(mean(in_degrees), 4))
print(" median degree: ", statistics.median(in_degrees))
print("\nOut Degree:")
print(" max degree: ", max(out_degrees))
print(" mean degree: ", round(mean(out_degrees), 4))
print(" median degree: ", statistics.median(out_degrees))
create_degree_distribution(in_degrees, 'In-Degree Distribution', '#33691e',
20, 100, 10)
create_degree_distribution(out_degrees, 'Out-Degree Distribution',
'#e65100')
def admin_login():
if request.method == 'GET':
return redirect('admin/')
api = API()
if api.check_user_login(request.form['username'],
request.form['password']):
session['logged_in'] = True
papers = api.get_all_paper()
return render_template('admin/papers.html', papers=papers)
else:
return redirect('admin/')
def evaluate_ranking_time(paper, settings):
print("Evaluate ", settings["algorithm"])
api = API()
settings["mode"] = Mode.importance_to_sections
settings["use-unclassified-chapters"] = True
start = time.time()
ranked_papers, queries = api.get_papers_with_paper(paper.filename,
settings)
end = time.time()
elapsed_time = end - start
print("Elapsed time: ", elapsed_time)
def test_importance_to_sections(self):
api = API()
papers = api.get_all_paper()
settings = {"mode": Mode.importance_to_sections}
queries = paper_to_queries(papers[0], settings)
self.assertEqual(len(queries["whole-document"]), 0)
self.assertGreater(len(queries[IMRaDType.INTRODUCTION.name]), 0)
self.assertGreater(len(queries[IMRaDType.BACKGROUND.name]), 0)
self.assertGreater(len(queries[IMRaDType.METHODS.name]), 0)
self.assertGreater(len(queries[IMRaDType.RESULTS.name]), 0)
self.assertGreater(len(queries[IMRaDType.DISCUSSION.name]), 0)
def test_import_intro_search_back(self):
api = API()
papers = api.get_all_paper()
settings = {
"mode": Mode.areas,
"input-area": Area.Introduction,
"search-area": Area.Background
}
queries = paper_to_queries(papers[0], settings)
self.assertEqual(len(queries["whole-document"]), 0)
self.assertEqual(len(queries[IMRaDType.INTRODUCTION.name]), 0)
self.assertGreater(len(queries[IMRaDType.BACKGROUND.name]), 0)
self.assertEqual(len(queries[IMRaDType.METHODS.name]), 0)
self.assertEqual(len(queries[IMRaDType.RESULTS.name]), 0)
self.assertEqual(len(queries[IMRaDType.DISCUSSION.name]), 0)
def __add_files(folder):
api = API()
text_file = open("newpapers.txt", "a+")
for filename in os.listdir(os.path.abspath(folder)):
print('CURRENT FILE: ' + str(filename))
if filename.endswith('.pdf'):
src = folder + "/" + filename
dst = UPLOAD_FOLDER + filename
shutil.move(src, dst)
try:
paper = api.add_paper(filename)
text_file.write(str(paper.id) + "\n")
except (IOError, OSError, ClassificationError, DocumentTooLarge,
PaperInStorage) as e:
print(e)
text_file.close()
def evaluate_query_time(num_papers):
api = API()
papers = api.get_all_paper()[:num_papers]
settings = TFIDF.get_default_config()
settings["mode"] = Mode.importance_to_sections
settings["use-unclassified-chapters"] = True
all_elapsed_times = []
for paper in papers:
start = time.time()
api.get_papers_with_paper(paper.filename, settings)
end = time.time()
elapsed_time = start - end
print("Elapsed time: ", elapsed_time)
all_elapsed_times.append(elapsed_time)
print("Overall time: ", sum(all_elapsed_times))
print("Mean: ", mean(all_elapsed_times))
def check_references():
print("\nCheck References")
api = API()
papers = api.get_all_paper()
for i, paper in enumerate(papers):
progressBar(i, len(papers))
other_papers = [p for p in papers if p.id != paper.id]
for reference in paper.references:
if not reference.get_paper_id():
continue
ref_paper = api.get_paper(reference.get_paper_id())
if ref_paper.cited_by.count(paper.id) == 0:
print()
reference.paper_id = []
api.client.update_paper(paper)
repair_corrupt_reference(reference, paper, other_papers, api)
def test_simple_ranking(self):
queries = {
IMRaDType.INTRODUCTION.name: "aaa",
IMRaDType.BACKGROUND: "",
IMRaDType.METHODS.name: "aaa bbb ccc ddd eee fff",
IMRaDType.RESULTS.name: "",
IMRaDType.DISCUSSION.name: "",
"whole-document": "ggg aaa ccc"
}
settings = {
**{
"importance_sections": False
},
**TF.get_default_config()
}
api = API()
ret = api.get_papers(queries, settings)
self.assertGreater(len(ret), 0)
def check_cited_by():
print("\nCheck Cited by")
api = API()
papers = api.get_all_paper()
for i, paper in enumerate(papers):
progressBar(i, len(papers))
for cited_paper_id in paper.cited_by:
if not api.contains_paper(cited_paper_id):
paper.cited_by.remove(cited_paper_id)
api.client.update_paper(paper)
continue
cited_paper = api.get_paper(cited_paper_id)
cited_paper_refs = [ref.get_paper_id() for ref in cited_paper.references if ref.get_paper_id()]
if cited_paper_refs.count(paper.id) == 0:
print()
paper.cited_by.remove(cited_paper_id)
api.client.update_paper(paper)
link_references_to_paper(cited_paper, paper, api)
def __link_references_to_paper():
api = API()
all_papers = api.get_all_paper()
finished_files = []
if not os.path.isfile(REQ_DATA_PATH + "finished_papers.txt"):
with open(REQ_DATA_PATH + "finished_papers.txt", 'wb') as fp:
pickle.dump(finished_files, fp)
with open(REQ_DATA_PATH + "finished_papers.txt", 'rb') as fp:
finished_files = pickle.load(fp)
if os.path.isfile("newpapers.txt"):
papers = []
with open("newpapers.txt", 'r') as fp:
for paper_id in fp:
papers.append(api.get_paper(paper_id.rstrip()))
else:
papers = api.get_all_paper()
i = 0
for paper in papers:
i += 1
print("(", i, "/", len(papers), ")")
if paper.id in finished_files:
continue
other_papers = [p for p in all_papers if p.id != paper.id]
for other_paper in other_papers:
if os.path.isfile("newpapers.txt"):
link_references_to_paper(other_paper, paper, api)
link_references_to_paper(paper, other_paper, api)
finished_files.append(paper.id)
with open(REQ_DATA_PATH + "finished_papers.txt", 'wb') as fp:
pickle.dump(finished_files, fp)
def remove_link(paper_id):
if not ('logged_in' in session.keys() and session['logged_in']):
return redirect('admin/')
api = API()
api.remove__link_of_paper(paper_id, request.form['ref_paper_id'])
papers = api.get_all_paper()
id_to_filename = {paper.id: paper.filename for paper in papers}
paper = api.get_paper(paper_id)
return render_template('admin/paper_info.html',
paper=paper,
id_to_filename=id_to_filename)
def calculate_ranking_sections(self, raw_queries, settings):
api = API()
mean_ap_intro, mean_ap_background, mean_ap_methods, mean_ap_result, mean_ap_discussion = [], [], [], [], []
queries = self.__raw_queries_to_queries(raw_queries)
for i, query in enumerate(queries):
progressBar(i, len(queries))
relevant_paper = [api.get_paper(reference["paper_id"]) for reference in query["references"]]
ranked_papers_intro = api.get_papers({IMRaDType.INTRODUCTION.name: query["search_query"]}, settings)
ranked_papers_background = api.get_papers({IMRaDType.BACKGROUND.name: query["search_query"]}, settings)
ranked_papers_methods = api.get_papers({IMRaDType.METHODS.name: query["search_query"]}, settings)
ranked_papers_result = api.get_papers({IMRaDType.RESULTS.name: query["search_query"]}, settings)
ranked_papers_discussion = api.get_papers({IMRaDType.DISCUSSION.name: query["search_query"]}, settings)
ap_intro = self.average_precision(ranked_papers_intro, relevant_paper)
ap_background = self.average_precision(ranked_papers_background, relevant_paper)
ap_methods = self.average_precision(ranked_papers_methods, relevant_paper)
ap_result = self.average_precision(ranked_papers_result, relevant_paper)
ap_discussion = self.average_precision(ranked_papers_discussion, relevant_paper)
mean_ap_intro.append(ap_intro)
mean_ap_background.append(ap_background)
mean_ap_methods.append(ap_methods)
mean_ap_result.append(ap_result)
mean_ap_discussion.append(ap_discussion)
print()
print("{} & {} & {}".format(Mode.only_introduction.name.replace("_", " "),
len(mean_ap_intro), sum(mean_ap_intro) / len(mean_ap_intro)))
print("{} & {} & {}".format(Mode.only_background.name.replace("_", " "),
len(mean_ap_background), sum(mean_ap_background) / len(mean_ap_background)))
print("{} & {} & {}".format(Mode.only_methods.name.replace("_", " "),
len(mean_ap_methods), sum(mean_ap_methods) / len(mean_ap_methods)))
print("{} & {} & {}".format(Mode.only_results.name.replace("_", " "),
len(mean_ap_result), sum(mean_ap_result) / len(mean_ap_result)))
print("{} & {} & {}".format(Mode.only_discussion.name.replace("_", " "),
len(mean_ap_discussion), sum(mean_ap_discussion) / len(mean_ap_discussion)))
# encoding: utf-8
import json
import os
from flask import Blueprint, render_template, request, current_app, send_file
from engine.api import API
from engine.datastore.ranking.ranked_boolean_retrieval import RankedBoolean
from engine.datastore.ranking.tfidf import TFIDF
from engine.datastore.models.section import IMRaDType
backend = Blueprint('backend', __name__)
api = API()
@backend.route('/', methods=["GET", "POST"])
def index():
if request.method == "GET":
return render_template('index.html',
error=None,
algorithm=api.get_ranking_algos())
queries = {
"whole-document": request.form['whole_text'],
IMRaDType.INTRODUCTION.name: request.form['intro_text'],
IMRaDType.BACKGROUND.name: request.form['background_text'],
IMRaDType.METHODS.name: request.form['methods_text'],
IMRaDType.RESULTS.name: request.form['results_text'],
IMRaDType.DISCUSSION.name: request.form['discussion_text']
}
print("Elapsed time: ", elapsed_time)
def evaluate_query_time(num_papers):
api = API()
papers = api.get_all_paper()[:num_papers]
settings = TFIDF.get_default_config()
settings["mode"] = Mode.importance_to_sections
settings["use-unclassified-chapters"] = True
all_elapsed_times = []
for paper in papers:
start = time.time()
api.get_papers_with_paper(paper.filename, settings)
end = time.time()
elapsed_time = start - end
print("Elapsed time: ", elapsed_time)
all_elapsed_times.append(elapsed_time)
print("Overall time: ", sum(all_elapsed_times))
print("Mean: ", mean(all_elapsed_times))
if __name__ == "__main__":
# evaluate_query_time(10)
api = API()
papers = api.get_all_paper()
evaluate_ranking_time(papers[0], RankedBoolean.get_default_config())
evaluate_ranking_time(papers[0],
DivergenceFromRandomness.get_default_config())
def __init__(self):
self.api = API()
