def read_in_paper(filename, sentences_as_lists=False, preserve_order=False):
"""
Reads in a paper and returns it as a dictionary.
:param filename: the filename of the paper to read, of the paper file itself only not the path to the paper.
:param sentences_as_lists: if true, will return the sentences of the paper as lists of words rather than strings.
:param preserve_order: if true keeps track of which sections occured in what order in the paper.
:return: a dictionary of the form (section: list of sentences in that section)
"""
paper_to_open = PAPER_SOURCE + filename
paper_text = Reader().open_file_single_string(paper_to_open)
udata = paper_text.decode("utf-8")
paper = udata.encode("ascii", "ignore")
return paper_tokenize(paper, sentences_as_lists=sentences_as_lists, preserve_order=preserve_order)
def read_in_files():
"""Function which reads in all the scientific paper data, and parses it into a list of lists, where each item in
the list is a sentence, in the form of a list of words. This is the form needed for the Word2Vec model."""
num_files = len(
[name for name in os.listdir(DATA_SOURCE) if name.endswith(".txt")])
loading_section_size = num_files / 30
count = 0
sentences_as_lists = []
for filename in os.listdir(DATA_SOURCE):
if filename.endswith(".txt"):
# Pretty loading bar
print("Processing Files: [", end="")
for i in range(31, -1, -1):
if count > i * loading_section_size:
for j in range(0, i):
print("-", end="")
sys.stdout.flush()
for j in range(i, 30):
print(" ", end="")
sys.stdout.flush()
break
if count == num_files:
print("] ", count, end="\n")
else:
print("] ", count, end="\r")
sys.stdout.flush()
# Open the paper
paper_to_open = DATA_SOURCE + filename
paper = Reader().open_file_single_string(paper_to_open)
udata = paper.decode("utf-8")
paper = udata.encode("ascii", "ignore")
# Split the data into a list of sentences, where each sentence is a list of words
sentences = sent_tokenize(paper)
for sentence in sentences:
words = word_tokenize(sentence)
sentences_as_lists.append(words)
if DEBUG:
print(sentences_as_lists)
#wait()
count += 1
return count, sentences_as_lists
def read_stopwords():
"""
Reads the list of stop words that should not be included in the scoring of each sentence such as \"and\"
and \"or\".
"""
common_words = Reader().open_file(STOPWORD_SOURCE) + list(string.punctuation)
return set(common_words)
def read_definite_non_summary_section_titles():
"""
Reads the list of sections titles from which summary statements are very rare to come from.
:return: the list of such section titles.
"""
return set(Reader().open_file(
BASE_DIR + "/Data/Utility_Data/definite_non_summary_titles.txt"))
def read_section_titles():
"""
Reads the section titles permitted in the paper.
:return: A set of the permitted titles.
"""
return set(Reader().open_file(BASE_DIR +
"/Data/Utility_Data/permitted_titles.txt"))
def paper_tokenize(text, sentences_as_lists=False, preserve_order=False):
"""
Takes a paper with the sections delineated by '@&#' and splits them into a dictionary where the key is the section
and the value is the text under that section. This could probably be a bit more efficient but it works well enough.
:param text: the text of the paper to split
:param sentences_as_lists: if true, returns the text of each sentence as a list of sentences.
:param preserve_order: if true, tracks the order in which the paper sections occured.
:returns: a dictionary of the form (section: section_text)
"""
permitted_titles = set(Reader().open_file(PERMITTED_TITLES_SOURCE))
# Split the text into sections
if preserve_order:
split_text_1 = re.split("@&#", text)
split_text = zip(split_text_1, range(len(split_text_1)))
else:
split_text = re.split("@&#", text)
# The key value. This value is changed if a permitted section title is encountered in the list.
state = ""
# After the for loop, this dictionary will have keys relating to each permitted section, and values corresponding
# to the text of that section
sentences_with_states = defaultdict(str)
section_counts = defaultdict(int)
if preserve_order:
for text, pos in split_text:
# Hack for proper sentence tokenization because NLTK tokeniser doesn't work properly for tokenising papers
text = text.replace("etal.", "etal")
text = text.replace("et al.", "etal")
text = text.replace("Fig.", "Fig")
text = text.replace("fig.", "fig")
text = text.replace("Eq.", "Eq")
text = text.replace("eq.", "eq")
text = text.replace("pp.", "pp")
text = text.replace("i.e.", "ie")
text = text.replace("e.g.", "eg")
text = text.replace("ref.", "ref")
text = text.replace("Ref.", "Ref")
text = text.replace("etc.", "etc")
text = text.replace("Figs.", "Figs")
text = text.replace("figs.", "figs")
text = text.replace("No.", "No")
text = text.replace("eqs.", "eqs")
# Checks if text is a section title
if text.lower() in permitted_titles:
state = text
section_counts[state] += 1
else:
if sentences_as_lists:
if section_counts[state] > 1:
state = state + "_" + str(section_counts[state])
sentences_with_states[state] = ([preprocess_sentence(x) for x in sent_tokenize(text)], pos)
else:
if section_counts[state] > 1:
state = state + "_" + str(section_counts[state])
sentences_with_states[state] = (text, pos)
if not preserve_order:
for text in split_text:
# Hack for proper sentence tokenization because NLTK tokeniser doesn't work properly for tokenising papers
text = text.replace("etal.", "etal")
text = text.replace("et al.", "etal")
text = text.replace("Fig.", "Fig")
text = text.replace("fig.", "fig")
text = text.replace("Eq.", "Eq")
text = text.replace("eq.", "eq")
text = text.replace("pp.", "pp")
text = text.replace("i.e.", "ie")
text = text.replace("e.g.", "eg")
text = text.replace("ref.", "ref")
text = text.replace("Ref.", "Ref")
text = text.replace("etc.", "etc")
text = text.replace("Figs.", "Figs")
text = text.replace("figs.", "figs")
text = text.replace("No.", "No")
text = text.replace("eqs.", "eqs")
# Checks if text is a section title
if text.lower() in permitted_titles:
state = text
section_counts[state] += 1
else:
if sentences_as_lists:
if section_counts[state] > 1:
state = state + "_" + str(section_counts[state])
sentences_with_states[state] = [preprocess_sentence(x) for x in sent_tokenize(text)]
else:
if section_counts[state] > 1:
state = state + "_" + str(section_counts[state])
sentences_with_states[state] = text
return sentences_with_states
# Location of the file with stopwords in it
STOPWORD_SOURCE = BASE_DIR + "/Data/Utility_Data/common_words.txt"
# Location of the Word2Vec model
MODEL_SOURCE = BASE_DIR + "/Word2Vec/Word2Vec_Models/summarisation_100features_5minwords_20context"
# Location to write the global filecounts of bag of words for each paper and total word count
GLOBAL_WORDCOUNT_WRITE_LOC = BASE_DIR + "/Data/Utility_Data/Global_Counts/"
# This may not be needed anymore
DD_CLASSIFICATION_MODEL_LOC_VECS = BASE_DIR + "/Trained_Models/DoubleDataset/lr_c2_5_vector.pkl"
DD_CLASSIFICATION_MODEL_LOC_FEATS = BASE_DIR + "/Trained_Models/DoubleDataset/lr_c2_5_feature.pkl"
W2V_CLASSIFICATION_MODEL_LOC = BASE_DIR + "/Trained_Models/Word2Vec_Classifier/lr_highlights_only_c3_66.pkl"
# Reads the stopwords as a set
STOPWORDS = set(Reader().open_file(STOPWORD_SOURCE) + list(string.punctuation))
# Counts how many papers there are in the paper directory
NUMBER_OF_PAPERS = len([name for name in os.listdir(PAPER_SOURCE) if name.endswith(".txt")])
# Integers defining the possible location of sentences in the paper
HIGHLIGHT = 0
ABSTRACT = 1
INTRODUCTION = 2
RESULT_DISCUSSION = 3
METHOD = 4
CONCLUSION = 5
OTHER = 6
# An object which can compare two sentences and tell how similar they are / if they are the same
SENTENCE_COMPARATOR_OBJ = SentenceComparator()