textfiles-politics/pythonCode/cultureTagger-old.py
2023-04-30 18:35:36 -04:00

184 lines
9.6 KiB
Python

import spacy
from collections import Counter
import re as regex
import os
from saxonche import PySaxonProcessor
#### Loads all of the necessary variables and functions.
nlp = spacy.load("en_core_web_lg")
#########################################################################################
# ebb: After reading the NLP output, we know spaCy is making some mistakes.
# So, here let's try adding an EntityRuler to customize spaCy's classification. We need
# to configure this BEFORE we send the tokens off to nlp() for processing.
##########################################################################################
# Create the EntityRuler and set it so the ner comes after, so OUR rules take precedence
# Sources:
# W. J. B. Mattingly: https://ner.pythonhumanities.com/02_01_spaCy_Entity_Ruler.html
# spaCy documentation on NER Entity Ruler: https://spacy.io/usage/rule-based-matching#entityruler
config = {"spans_key": None, "annotate_ents": True, "overwrite": True, "validate": True}
ruler = nlp.add_pipe("span_ruler", before="ner", config=config)
# 2023-04-07: ebb: NOTE: before="ner" setting seems to allow the spaCy NER rules to prevail over these patterns where
# there is a conflict.
# after="ner" means that the spaCy NER is TOTALLY OVERWRITTEN and invalidated by our patterns.
# Notes: Mattingly has this: ruler = nlp.add_pipe("entity_ruler", after="ner", config={"validate": True})
# But this only works when spaCy doesn't recognize a word / phrase as a named entity of any kind.
# If it recognizes a named entity but tags it wrong, we correct it with the span_ruler, not the entity_ruler
patterns = [
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^-\w+?"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^.$"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^\w\w$"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^[a-z]+\s+[a-z]+$"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^.*?__{2,}.*?$"}}]},
{"label": "NORP", "pattern": [{"TEXT": {"REGEX": "CHRISTIAN(ITY|DOM)"}}]},
{"label": "NORP", "pattern": [{"TEXT": {"REGEX": "CHRISTIAN\s+NETWORK"}}]},
# ebb: Don't match on any single characters!
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "[A-Z]{2,}[A-Z][a-z]+"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "[a-z]{2,}[A-Z][a-z]+"}}]},
{"label": "NULL", "pattern": [{"TEXT": {"REGEX": "^.*?[a-z][A-Z].*?$"}}]},
# ebb: Above line attempts to stop matching things like Oak IslandThe Method
{"label": "NULL", "pattern": [{"TEXT" : {"REGEX": "^[Mm\-]+$"}}]},
# SOCIALISMBY RICHARD
# ebb: Above line attempts to stop matching things Mmm-mm or mm , etc.
{"label": "GPE", "pattern": [{"TEXT": {"REGEX": "Babylon(ia)?"}}]},
{"label": "NORP", "pattern": [{"TEXT": {"REGEX": "Christiani\s*ty"}}]},
{"label": "NULL", "pattern": "Christiani"},
{"label": "NULL", "pattern": "di"},
{"label": "NULL", "pattern": "ORG"},
{"label": "PERSON", "pattern": [{"TEXT": {"REGEX": "(Ludwig [Vv]an )?Beethoven"}}]},
{"label": "ORG", "pattern": "Falangist"},
{"label": "NORP", "pattern": "Dropa"},
{"label": "GPE", "pattern": "Nazareth"},
{"label": "NULL", "pattern": "Bab"},
]
ruler.add_patterns(patterns)
ruler.add_patterns(patterns)
workingDir = os.getcwd()
CollPath = os.path.join(workingDir, '../regexConsp')
outputPath = os.path.join(workingDir, 'cultureTaggedOutput/')
# Everything in original conspiracy directory.
insideDir = os.listdir(CollPath)
print(insideDir)
# Copies files in case they do not exist
def copyTextFiles(file):
content = []
# Reads the contents of file, and saves each line of file into the content array.
with open(CollPath + "/" + file, 'r', encoding='utf8') as inFile:
for line in inFile:
content.append(line)
print(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~ copying " + file + " ~~~~~~~~~~~~~~~~~~~~~~~~~~~ ")
inFile.close()
# With the contents copied, a loop will go through the array and write it all in a new file in output folder.
with open(outputPath + "/" + file, 'w', encoding='utf8') as f:
for line in content:
f.write(str(line))
# Function runs through the tokens of given file. Entities are stored in array, then returned. Called by regexFile().
def entitycollector(tokens):
# creates a new file that includes all of the found entities.
with open('output.txt', 'w') as f:
entities = {}
# goes through each entity in the token list.
for ent in sorted(tokens.ents):
entityInfo = [ent.text, ent.label_, spacy.explain(ent.label_)]
stringify = str(entityInfo)
f.write(stringify)
f.write('\n')
entities[ent.text] = ent.label_
# return all entities with its label and text.
return entities
# Function runs regex through given file.
def regexFile(file):
fileDir = os.path.join(outputPath, file)
with PySaxonProcessor(license=False) as proc:
# grabs the original xml file and stores it in a variable for later. this some xquery bs
xml = open(fileDir, encoding='utf-8').read()
xp = proc.new_xpath_processor()
node = proc.parse_xml(xml_text=xml)
xp.set_context(xdm_item=node)
# xquery goes through original text, and stores it all in a single string.
xpath = xp.evaluate('//p ! normalize-space() => string-join()')
string = str(xpath)
# regex goes through the text and deletes anything that is not a letter or space.
cleanedText = regex.sub(r'[^A-z ]+', ' ', string)
# gets the tokens of the clean text.
tokens = nlp(cleanedText)
wrappedText = xml
# grabs all the entities in file and stores it in a list/array.
dictEntities = entitycollector(tokens)
# if anything exists in the list, the following code will run.
if dictEntities:
# it will check through each entity in the list and see its entity type. it is looking for "PERSON" tokens
# in this instance, which includes of nouns and names.
for entity, value in dictEntities.items():
if value == "LOC" or value=="ORG" or value=="GPE" or value=="NORP":
# key_template variable is the elements we wrap around found instances.
key_template = "<ent type='" + value + "'>" + entity + "</ent>"
# loops through wrappedText until all entities are wrapped.
wrappedText = wrappedText.replace(entity, key_template)
# Saves newly wrapped elements and then writes it into new file.
with open(fileDir, 'w', encoding='utf8') as f:
f.write(wrappedText)
print("WRAPPING " + entity)
checkTags(file)
# ebb: Added above line to send the tagged file to the checkTags() function for cleaning.
# This part of the code is a WIP.
# ebb: I just activated it, and it works! (Nice job.) I altered it just a bit. May need more regexes to match.
## It tries to find weird or invalid elements/tags and fix them.
def checkTags(file):
content = []
fileDir = os.path.join(outputPath, file)
with open(fileDir, 'r', encoding='utf8') as inFile:
for line in inFile:
content.append(line)
# With the contents copied, a loop will go through the array and write it all in a new file in output folder.
with open(fileDir, 'w', encoding='utf8') as f:
for line in content:
# match = regex.search(r"(<ent type='.+?'>[^<>]*?)<ent[^>]+?>([^<>]+?)</ent>([^<>]*?</ent>)", line)
# if match:
# print("broken line found, fixing...")
# ebb: NOTE: If this function only processes a line there's a regex match, we'd have a serious problem:
# we'd not output the rest of the file--only the cleaned matches. So the output files would be mostly empty!
# Better to just string-clean every line using regex.sub(). Where there's no regex match, no substitution will happen.
origLine = line
# newLine = regex.sub(r"(<ent type='.+?'>[^<>]*?)<ent[^>]+?>([^<>]+?)</ent>([^<>]*?</ent>)", r"\1\2\3",line)
# <spe<ent type='ORG'>cia</ent>l>
newLine = regex.sub(r"(</?spe)<ent type='ORG'>(cia)</ent>(l>)", r"\1\2\3", origLine)
# newLine = regex.sub(r"(<)<ent type='ORG'>(di)</ent>(v>)", r"\1\2\3", newLine)
newLine = regex.sub(r"(<ent type='[A-z]+?'>[^<]*?)<ent type='[A-z]+?'>([^<]+?)</ent>([^<]*?</ent>)", r"\1\2\3", newLine)
newLine = regex.sub(r"(<ent type='[A-z]+?'>[^<]*?)<ent type='[A-z]+?'>([^<]+?)</ent>([^<]*?</ent>)", r"\1\2\3", newLine)
newLine = regex.sub(r"(<ent type='[A-z]+?'>[^<]*?)<ent type='[A-z]+?'>([^<]+?)</ent>([^<]*?</ent>)", r"\1\2\3", newLine)
# ebb: I'm repeating the above just in case of the weird event of triple or quadruple nested <ent> tags in <ent> tags.
# We saw it happen on the LOTR project and running it through multiple passes of the above line ultimately got rid of them all
# preserving only the outermost tags.
newLine = regex.sub(r"(<ent type=')<ent type='ORG'>(ORG)</ent>('>)", r"\1\2\3", newLine)
# ebb: noticed:
# <spe<ent type='ORG'>cia</ent>l>
# <<ent type='ORG'>di</ent>v>
if origLine != newLine:
print("broken line found, fixing...")
print(origLine + "\n INTO.")
print(newLine)
f.write(str(newLine))
print("File checking finished.")
for file in insideDir:
copyTextFiles(file)
regexFile(file)
# checkTags(file) # ebb: You don't really want to activate checkTags here,
# because it would run over the untagged input files.